fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-22 04:50:11 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

intel/brw: Replace uses of fs_reg with brw_reg

Browse source 
And remove the fs_reg alias.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29791>
This commit is contained in:
Caio Oliveira 2024-06-18 23:42:59 -07:00 • committed by Marge Bot
parent fe46efa647
commit 3670c24740
38 changed files with 1626 additions and 1629 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

126
src/intel/compiler/brw_fs.cpp
View file

@ -48,11 +48,11 @@
using namespace brw; using namespace brw;
static void static void
initialize_sources(fs_inst *inst, const fs_reg src[], uint8_t num_sources); initialize_sources(fs_inst *inst, const brw_reg src[], uint8_t num_sources);
void void
fs_inst::init(enum opcode opcode, uint8_t exec_size, const fs_reg &dst, fs_inst::init(enum opcode opcode, uint8_t exec_size, const brw_reg &dst,
const fs_reg *src, unsigned sources) const brw_reg *src, unsigned sources)
{ {
memset((void*)this, 0, sizeof(*this)); memset((void*)this, 0, sizeof(*this));
@ -100,34 +100,34 @@ fs_inst::fs_inst(enum opcode opcode, uint8_t exec_size)
init(opcode, exec_size, reg_undef, NULL, 0); init(opcode, exec_size, reg_undef, NULL, 0);
} }
fs_inst::fs_inst(enum opcode opcode, uint8_t exec_size, const fs_reg &dst) fs_inst::fs_inst(enum opcode opcode, uint8_t exec_size, const brw_reg &dst)
{ {
init(opcode, exec_size, dst, NULL, 0); init(opcode, exec_size, dst, NULL, 0);
} }
fs_inst::fs_inst(enum opcode opcode, uint8_t exec_size, const fs_reg &dst, fs_inst::fs_inst(enum opcode opcode, uint8_t exec_size, const brw_reg &dst,
const fs_reg &src0) const brw_reg &src0)
{ {
const fs_reg src[1] = { src0 }; const brw_reg src[1] = { src0 };
init(opcode, exec_size, dst, src, 1); init(opcode, exec_size, dst, src, 1);
} }
fs_inst::fs_inst(enum opcode opcode, uint8_t exec_size, const fs_reg &dst, fs_inst::fs_inst(enum opcode opcode, uint8_t exec_size, const brw_reg &dst,
const fs_reg &src0, const fs_reg &src1) const brw_reg &src0, const brw_reg &src1)
{ {
const fs_reg src[2] = { src0, src1 }; const brw_reg src[2] = { src0, src1 };
init(opcode, exec_size, dst, src, 2); init(opcode, exec_size, dst, src, 2);
} }
fs_inst::fs_inst(enum opcode opcode, uint8_t exec_size, const fs_reg &dst, fs_inst::fs_inst(enum opcode opcode, uint8_t exec_size, const brw_reg &dst,
const fs_reg &src0, const fs_reg &src1, const fs_reg &src2) const brw_reg &src0, const brw_reg &src1, const brw_reg &src2)
{ {
const fs_reg src[3] = { src0, src1, src2 }; const brw_reg src[3] = { src0, src1, src2 };
init(opcode, exec_size, dst, src, 3); init(opcode, exec_size, dst, src, 3);
} }
fs_inst::fs_inst(enum opcode opcode, uint8_t exec_width, const fs_reg &dst, fs_inst::fs_inst(enum opcode opcode, uint8_t exec_width, const brw_reg &dst,
const fs_reg src[], unsigned sources) const brw_reg src[], unsigned sources)
{ {
init(opcode, exec_width, dst, src, sources); init(opcode, exec_width, dst, src, sources);
} }
@ -145,10 +145,10 @@ fs_inst::~fs_inst()
} }
static void static void
initialize_sources(fs_inst *inst, const fs_reg src[], uint8_t num_sources) initialize_sources(fs_inst *inst, const brw_reg src[], uint8_t num_sources)
{ {
if (num_sources > ARRAY_SIZE(inst->builtin_src)) if (num_sources > ARRAY_SIZE(inst->builtin_src))
inst->src = new fs_reg[num_sources]; inst->src = new brw_reg[num_sources];
else else
inst->src = inst->builtin_src; inst->src = inst->builtin_src;
@ -164,14 +164,14 @@ fs_inst::resize_sources(uint8_t num_sources)
if (this->sources == num_sources) if (this->sources == num_sources)
return; return;
fs_reg *old_src = this->src; brw_reg *old_src = this->src;
fs_reg *new_src; brw_reg *new_src;
const unsigned builtin_size = ARRAY_SIZE(this->builtin_src); const unsigned builtin_size = ARRAY_SIZE(this->builtin_src);
if (old_src == this->builtin_src) { if (old_src == this->builtin_src) {
if (num_sources > builtin_size) { if (num_sources > builtin_size) {
new_src = new fs_reg[num_sources]; new_src = new brw_reg[num_sources];
for (unsigned i = 0; i < this->sources; i++) for (unsigned i = 0; i < this->sources; i++)
new_src[i] = old_src[i]; new_src[i] = old_src[i];
@ -189,7 +189,7 @@ fs_inst::resize_sources(uint8_t num_sources)
new_src = old_src; new_src = old_src;
} else { } else {
new_src = new fs_reg[num_sources]; new_src = new brw_reg[num_sources];
for (unsigned i = 0; i < num_sources; i++) for (unsigned i = 0; i < num_sources; i++)
new_src[i] = old_src[i]; new_src[i] = old_src[i];
} }
@ -204,10 +204,10 @@ fs_inst::resize_sources(uint8_t num_sources)
void void
fs_visitor::VARYING_PULL_CONSTANT_LOAD(const fs_builder &bld, fs_visitor::VARYING_PULL_CONSTANT_LOAD(const fs_builder &bld,
const fs_reg &dst, const brw_reg &dst,
const fs_reg &surface, const brw_reg &surface,
const fs_reg &surface_handle, const brw_reg &surface_handle,
const fs_reg &varying_offset, const brw_reg &varying_offset,
uint32_t const_offset, uint32_t const_offset,
uint8_t alignment, uint8_t alignment,
unsigned components) unsigned components)
@ -218,7 +218,7 @@ fs_visitor::VARYING_PULL_CONSTANT_LOAD(const fs_builder &bld,
* be any component of a vector, and then we load 4 contiguous * be any component of a vector, and then we load 4 contiguous
* components starting from that. TODO: Support loading fewer than 4. * components starting from that. TODO: Support loading fewer than 4.
*/ */
fs_reg total_offset = bld.ADD(varying_offset, brw_imm_ud(const_offset)); brw_reg total_offset = bld.ADD(varying_offset, brw_imm_ud(const_offset));
/* The pull load message will load a vec4 (16 bytes). If we are loading /* The pull load message will load a vec4 (16 bytes). If we are loading
* a double this means we are only loading 2 elements worth of data. * a double this means we are only loading 2 elements worth of data.
@ -226,9 +226,9 @@ fs_visitor::VARYING_PULL_CONSTANT_LOAD(const fs_builder &bld,
* so other parts of the driver don't get confused about the size of the * so other parts of the driver don't get confused about the size of the
* result. * result.
*/ */
fs_reg vec4_result = bld.vgrf(BRW_TYPE_F, 4); brw_reg vec4_result = bld.vgrf(BRW_TYPE_F, 4);
fs_reg srcs[PULL_VARYING_CONSTANT_SRCS]; brw_reg srcs[PULL_VARYING_CONSTANT_SRCS];
srcs[PULL_VARYING_CONSTANT_SRC_SURFACE] = surface; srcs[PULL_VARYING_CONSTANT_SRC_SURFACE] = surface;
srcs[PULL_VARYING_CONSTANT_SRC_SURFACE_HANDLE] = surface_handle; srcs[PULL_VARYING_CONSTANT_SRC_SURFACE_HANDLE] = surface_handle;
srcs[PULL_VARYING_CONSTANT_SRC_OFFSET] = total_offset; srcs[PULL_VARYING_CONSTANT_SRC_OFFSET] = total_offset;
@ -668,7 +668,7 @@ fs_inst::is_partial_write() const
/* Special case UNDEF since a lot of places in the backend do things like this : /* Special case UNDEF since a lot of places in the backend do things like this :
* *
* fs_builder ubld = bld.exec_all().group(1, 0); * fs_builder ubld = bld.exec_all().group(1, 0);
* fs_reg tmp = ubld.vgrf(BRW_TYPE_UD); * brw_reg tmp = ubld.vgrf(BRW_TYPE_UD);
* ubld.UNDEF(tmp); <- partial write, even if the whole register is concerned * ubld.UNDEF(tmp); <- partial write, even if the whole register is concerned
*/ */
if (this->opcode == SHADER_OPCODE_UNDEF) { if (this->opcode == SHADER_OPCODE_UNDEF) {
@ -1151,13 +1151,13 @@ fs_visitor::emit_gs_thread_end()
if (mark_last_urb_write_with_eot()) if (mark_last_urb_write_with_eot())
return; return;
fs_reg srcs[URB_LOGICAL_NUM_SRCS]; brw_reg srcs[URB_LOGICAL_NUM_SRCS];
srcs[URB_LOGICAL_SRC_HANDLE] = gs_payload().urb_handles; srcs[URB_LOGICAL_SRC_HANDLE] = gs_payload().urb_handles;
srcs[URB_LOGICAL_SRC_COMPONENTS] = brw_imm_ud(0); srcs[URB_LOGICAL_SRC_COMPONENTS] = brw_imm_ud(0);
inst = abld.emit(SHADER_OPCODE_URB_WRITE_LOGICAL, reg_undef, inst = abld.emit(SHADER_OPCODE_URB_WRITE_LOGICAL, reg_undef,
srcs, ARRAY_SIZE(srcs)); srcs, ARRAY_SIZE(srcs));
} else { } else {
fs_reg srcs[URB_LOGICAL_NUM_SRCS]; brw_reg srcs[URB_LOGICAL_NUM_SRCS];
srcs[URB_LOGICAL_SRC_HANDLE] = gs_payload().urb_handles; srcs[URB_LOGICAL_SRC_HANDLE] = gs_payload().urb_handles;
srcs[URB_LOGICAL_SRC_DATA] = this->final_gs_vertex_count; srcs[URB_LOGICAL_SRC_DATA] = this->final_gs_vertex_count;
srcs[URB_LOGICAL_SRC_COMPONENTS] = brw_imm_ud(1); srcs[URB_LOGICAL_SRC_COMPONENTS] = brw_imm_ud(1);
@ -1212,7 +1212,7 @@ fs_visitor::assign_curb_setup()
/* The base offset for our push data is passed in as R0.0[31:6]. We have /* The base offset for our push data is passed in as R0.0[31:6]. We have
* to mask off the bottom 6 bits. * to mask off the bottom 6 bits.
*/ */
fs_reg base_addr = brw_reg base_addr =
ubld.AND(retype(brw_vec1_grf(0, 0), BRW_TYPE_UD), ubld.AND(retype(brw_vec1_grf(0, 0), BRW_TYPE_UD),
brw_imm_ud(INTEL_MASK(31, 6))); brw_imm_ud(INTEL_MASK(31, 6)));
@ -1228,17 +1228,17 @@ fs_visitor::assign_curb_setup()
/* This pass occurs after all of the optimization passes, so don't /* This pass occurs after all of the optimization passes, so don't
* emit an 'ADD addr, base_addr, 0' instruction. * emit an 'ADD addr, base_addr, 0' instruction.
*/ */
fs_reg addr = i == 0 ? base_addr : brw_reg addr = i == 0 ? base_addr :
ubld.ADD(base_addr, brw_imm_ud(i * REG_SIZE)); ubld.ADD(base_addr, brw_imm_ud(i * REG_SIZE));
fs_reg srcs[4] = { brw_reg srcs[4] = {
brw_imm_ud(0), /* desc */ brw_imm_ud(0), /* desc */
brw_imm_ud(0), /* ex_desc */ brw_imm_ud(0), /* ex_desc */
addr, /* payload */ addr, /* payload */
fs_reg(), /* payload2 */ brw_reg(), /* payload2 */
}; };
fs_reg dest = retype(brw_vec8_grf(payload().num_regs + i, 0), brw_reg dest = retype(brw_vec8_grf(payload().num_regs + i, 0),
BRW_TYPE_UD); BRW_TYPE_UD);
fs_inst *send = ubld.emit(SHADER_OPCODE_SEND, dest, srcs, 4); fs_inst *send = ubld.emit(SHADER_OPCODE_SEND, dest, srcs, 4);
@ -1310,10 +1310,10 @@ fs_visitor::assign_curb_setup()
struct brw_reg mask = brw_vec1_grf(payload().num_regs + mask_param / 8, struct brw_reg mask = brw_vec1_grf(payload().num_regs + mask_param / 8,
mask_param % 8); mask_param % 8);
fs_reg b32; brw_reg b32;
for (unsigned i = 0; i < 64; i++) { for (unsigned i = 0; i < 64; i++) {
if (i % 16 == 0 && (want_zero & BITFIELD64_RANGE(i, 16))) { if (i % 16 == 0 && (want_zero & BITFIELD64_RANGE(i, 16))) {
fs_reg shifted = ubld.vgrf(BRW_TYPE_W, 2); brw_reg shifted = ubld.vgrf(BRW_TYPE_W, 2);
ubld.SHL(horiz_offset(shifted, 8), ubld.SHL(horiz_offset(shifted, 8),
byte_offset(retype(mask, BRW_TYPE_W), i / 8), byte_offset(retype(mask, BRW_TYPE_W), i / 8),
brw_imm_v(0x01234567)); brw_imm_v(0x01234567));
@ -1625,13 +1625,13 @@ fs_visitor::assign_urb_setup()
foreach_block_and_inst(block, fs_inst, inst, cfg) { foreach_block_and_inst(block, fs_inst, inst, cfg) {
for (int i = 0; i < inst->sources; i++) { for (int i = 0; i < inst->sources; i++) {
if (inst->src[i].file == ATTR) { if (inst->src[i].file == ATTR) {
/* ATTR fs_reg::nr in the FS is in units of logical scalar /* ATTR brw_reg::nr in the FS is in units of logical scalar
* inputs each of which consumes 16B on Gfx4-Gfx12. In * inputs each of which consumes 16B on Gfx4-Gfx12. In
* single polygon mode this leads to the following layout * single polygon mode this leads to the following layout
* of the vertex setup plane parameters in the ATTR * of the vertex setup plane parameters in the ATTR
* register file: * register file:
* *
* fs_reg::nr Input Comp0 Comp1 Comp2 Comp3 * brw_reg::nr Input Comp0 Comp1 Comp2 Comp3
* 0 Attr0.x a1-a0 a2-a0 N/A a0 * 0 Attr0.x a1-a0 a2-a0 N/A a0
* 1 Attr0.y a1-a0 a2-a0 N/A a0 * 1 Attr0.y a1-a0 a2-a0 N/A a0
* 2 Attr0.z a1-a0 a2-a0 N/A a0 * 2 Attr0.z a1-a0 a2-a0 N/A a0
@ -1644,7 +1644,7 @@ fs_visitor::assign_urb_setup()
* different plane parameters, so each parameter above is * different plane parameters, so each parameter above is
* represented as a dispatch_width-wide vector: * represented as a dispatch_width-wide vector:
* *
* fs_reg::nr fs_reg::offset Input Comp0 ... CompN * brw_reg::nr brw_reg::offset Input Comp0 ... CompN
* 0 0 Attr0.x a1[0]-a0[0] ... a1[N]-a0[N] * 0 0 Attr0.x a1[0]-a0[0] ... a1[N]-a0[N]
* 0 4 * dispatch_width Attr0.x a2[0]-a0[0] ... a2[N]-a0[N] * 0 4 * dispatch_width Attr0.x a2[0]-a0[0] ... a2[N]-a0[N]
* 0 8 * dispatch_width Attr0.x N/A ... N/A * 0 8 * dispatch_width Attr0.x N/A ... N/A
@ -1955,7 +1955,7 @@ fs_visitor::assign_constant_locations()
} }
bool bool
fs_visitor::get_pull_locs(const fs_reg &src, fs_visitor::get_pull_locs(const brw_reg &src,
unsigned *out_surf_index, unsigned *out_surf_index,
unsigned *out_pull_index) unsigned *out_pull_index)
{ {
@ -1993,11 +1993,11 @@ fs_visitor::emit_repclear_shader()
assert(uniforms == 0); assert(uniforms == 0);
assume(key->nr_color_regions > 0); assume(key->nr_color_regions > 0);
fs_reg color_output = retype(brw_vec4_grf(127, 0), BRW_TYPE_UD); brw_reg color_output = retype(brw_vec4_grf(127, 0), BRW_TYPE_UD);
fs_reg header = retype(brw_vec8_grf(125, 0), BRW_TYPE_UD); brw_reg header = retype(brw_vec8_grf(125, 0), BRW_TYPE_UD);
/* We pass the clear color as a flat input. Copy it to the output. */ /* We pass the clear color as a flat input. Copy it to the output. */
fs_reg color_input = brw_reg color_input =
brw_make_reg(BRW_GENERAL_REGISTER_FILE, 2, 3, 0, 0, BRW_TYPE_UD, brw_make_reg(BRW_GENERAL_REGISTER_FILE, 2, 3, 0, 0, BRW_TYPE_UD,
BRW_VERTICAL_STRIDE_8, BRW_WIDTH_2, BRW_HORIZONTAL_STRIDE_4, BRW_VERTICAL_STRIDE_8, BRW_WIDTH_2, BRW_HORIZONTAL_STRIDE_4,
BRW_SWIZZLE_XYZW, WRITEMASK_XYZW); BRW_SWIZZLE_XYZW, WRITEMASK_XYZW);
@ -2047,7 +2047,7 @@ fs_visitor::emit_repclear_shader()
* thread payload, \p bld is required to have a dispatch_width() not greater * thread payload, \p bld is required to have a dispatch_width() not greater
* than 16 for fragment shaders. * than 16 for fragment shaders.
*/ */
fs_reg brw_reg
brw_sample_mask_reg(const fs_builder &bld) brw_sample_mask_reg(const fs_builder &bld)
{ {
const fs_visitor &s = *bld.shader; const fs_visitor &s = *bld.shader;
@ -2107,7 +2107,7 @@ brw_emit_predicate_on_sample_mask(const fs_builder &bld, fs_inst *inst)
bld.dispatch_width() == inst->exec_size); bld.dispatch_width() == inst->exec_size);
const fs_visitor &s = *bld.shader; const fs_visitor &s = *bld.shader;
const fs_reg sample_mask = brw_sample_mask_reg(bld); const brw_reg sample_mask = brw_sample_mask_reg(bld);
const unsigned subreg = sample_mask_flag_subreg(s); const unsigned subreg = sample_mask_flag_subreg(s);
if (s.devinfo->ver >= 20 || brw_wm_prog_data(s.prog_data)->uses_kill) { if (s.devinfo->ver >= 20 || brw_wm_prog_data(s.prog_data)->uses_kill) {
@ -3011,8 +3011,8 @@ fs_visitor::set_tcs_invocation_id()
* * 22:16 on gfx11+ * * 22:16 on gfx11+
* * 23:17 otherwise * * 23:17 otherwise
*/ */
fs_reg t = brw_reg t =
bld.AND(fs_reg(retype(brw_vec1_grf(0, 2), BRW_TYPE_UD)), bld.AND(brw_reg(retype(brw_vec1_grf(0, 2), BRW_TYPE_UD)),
brw_imm_ud(instance_id_mask)); brw_imm_ud(instance_id_mask));
if (vue_prog_data->dispatch_mode == INTEL_DISPATCH_MODE_TCS_MULTI_PATCH) { if (vue_prog_data->dispatch_mode == INTEL_DISPATCH_MODE_TCS_MULTI_PATCH) {
@ -3023,9 +3023,9 @@ fs_visitor::set_tcs_invocation_id()
assert(vue_prog_data->dispatch_mode == INTEL_DISPATCH_MODE_TCS_SINGLE_PATCH); assert(vue_prog_data->dispatch_mode == INTEL_DISPATCH_MODE_TCS_SINGLE_PATCH);
fs_reg channels_uw = bld.vgrf(BRW_TYPE_UW); brw_reg channels_uw = bld.vgrf(BRW_TYPE_UW);
fs_reg channels_ud = bld.vgrf(BRW_TYPE_UD); brw_reg channels_ud = bld.vgrf(BRW_TYPE_UD);
bld.MOV(channels_uw, fs_reg(brw_imm_uv(0x76543210))); bld.MOV(channels_uw, brw_reg(brw_imm_uv(0x76543210)));
bld.MOV(channels_ud, channels_uw); bld.MOV(channels_ud, channels_uw);
if (tcs_prog_data->instances == 1) { if (tcs_prog_data->instances == 1) {
@ -3054,7 +3054,7 @@ fs_visitor::emit_tcs_thread_end()
* algorithm to set it optimally). On other platforms, we simply write * algorithm to set it optimally). On other platforms, we simply write
* zero to a reserved/MBZ patch header DWord which has no consequence. * zero to a reserved/MBZ patch header DWord which has no consequence.
*/ */
fs_reg srcs[URB_LOGICAL_NUM_SRCS]; brw_reg srcs[URB_LOGICAL_NUM_SRCS];
srcs[URB_LOGICAL_SRC_HANDLE] = tcs_payload().patch_urb_output; srcs[URB_LOGICAL_SRC_HANDLE] = tcs_payload().patch_urb_output;
srcs[URB_LOGICAL_SRC_CHANNEL_MASK] = brw_imm_ud(WRITEMASK_X << 16); srcs[URB_LOGICAL_SRC_CHANNEL_MASK] = brw_imm_ud(WRITEMASK_X << 16);
srcs[URB_LOGICAL_SRC_DATA] = brw_imm_ud(0); srcs[URB_LOGICAL_SRC_DATA] = brw_imm_ud(0);
@ -3258,7 +3258,7 @@ fs_visitor::run_fs(bool allow_spilling, bool do_rep_send)
* stored in R0.15/R1.15 on gfx20+ and in R1.7/R2.7 on * stored in R0.15/R1.15 on gfx20+ and in R1.7/R2.7 on
* gfx6+. * gfx6+.
*/ */
const fs_reg dispatch_mask = const brw_reg dispatch_mask =
devinfo->ver >= 20 ? xe2_vec1_grf(i, 15) : devinfo->ver >= 20 ? xe2_vec1_grf(i, 15) :
brw_vec1_grf(i + 1, 7); brw_vec1_grf(i + 1, 7);
bld.exec_all().group(1, 0) bld.exec_all().group(1, 0)
@ -4533,18 +4533,18 @@ bool brw_should_print_shader(const nir_shader *shader, uint64_t debug_flag)
} }
namespace brw { namespace brw {
fs_reg brw_reg
fetch_payload_reg(const brw::fs_builder &bld, uint8_t regs[2], fetch_payload_reg(const brw::fs_builder &bld, uint8_t regs[2],
brw_reg_type type, unsigned n) brw_reg_type type, unsigned n)
{ {
if (!regs[0]) if (!regs[0])
return fs_reg(); return brw_reg();
if (bld.dispatch_width() > 16) { if (bld.dispatch_width() > 16) {
const fs_reg tmp = bld.vgrf(type, n); const brw_reg tmp = bld.vgrf(type, n);
const brw::fs_builder hbld = bld.exec_all().group(16, 0); const brw::fs_builder hbld = bld.exec_all().group(16, 0);
const unsigned m = bld.dispatch_width() / hbld.dispatch_width(); const unsigned m = bld.dispatch_width() / hbld.dispatch_width();
fs_reg *const components = new fs_reg[m * n]; brw_reg *const components = new brw_reg[m * n];
for (unsigned c = 0; c < n; c++) { for (unsigned c = 0; c < n; c++) {
for (unsigned g = 0; g < m; g++) for (unsigned g = 0; g < m; g++)
@ -4558,22 +4558,22 @@ namespace brw {
return tmp; return tmp;
} else { } else {
return fs_reg(retype(brw_vec8_grf(regs[0], 0), type)); return brw_reg(retype(brw_vec8_grf(regs[0], 0), type));
} }
} }
fs_reg brw_reg
fetch_barycentric_reg(const brw::fs_builder &bld, uint8_t regs[2]) fetch_barycentric_reg(const brw::fs_builder &bld, uint8_t regs[2])
{ {
if (!regs[0]) if (!regs[0])
return fs_reg(); return brw_reg();
else if (bld.shader->devinfo->ver >= 20) else if (bld.shader->devinfo->ver >= 20)
return fetch_payload_reg(bld, regs, BRW_TYPE_F, 2); return fetch_payload_reg(bld, regs, BRW_TYPE_F, 2);
const fs_reg tmp = bld.vgrf(BRW_TYPE_F, 2); const brw_reg tmp = bld.vgrf(BRW_TYPE_F, 2);
const brw::fs_builder hbld = bld.exec_all().group(8, 0); const brw::fs_builder hbld = bld.exec_all().group(8, 0);
const unsigned m = bld.dispatch_width() / hbld.dispatch_width(); const unsigned m = bld.dispatch_width() / hbld.dispatch_width();
fs_reg *const components = new fs_reg[2 * m]; brw_reg *const components = new brw_reg[2 * m];
for (unsigned c = 0; c < 2; c++) { for (unsigned c = 0; c < 2; c++) {
for (unsigned g = 0; g < m; g++) for (unsigned g = 0; g < m; g++)

118
src/intel/compiler/brw_fs.h
View file

@ -76,21 +76,21 @@ namespace brw {
~def_analysis(); ~def_analysis();
fs_inst * fs_inst *
get(const fs_reg &reg) const get(const brw_reg &reg) const
{ {
return reg.file == VGRF && reg.nr < def_count ? return reg.file == VGRF && reg.nr < def_count ?
def_insts[reg.nr] : NULL; def_insts[reg.nr] : NULL;
} }
bblock_t * bblock_t *
get_block(const fs_reg &reg) const get_block(const brw_reg &reg) const
{ {
return reg.file == VGRF && reg.nr < def_count ? return reg.file == VGRF && reg.nr < def_count ?
def_blocks[reg.nr] : NULL; def_blocks[reg.nr] : NULL;
} }
uint32_t uint32_t
get_use_count(const fs_reg &reg) const get_use_count(const brw_reg &reg) const
{ {
return reg.file == VGRF && reg.nr < def_count ? return reg.file == VGRF && reg.nr < def_count ?
def_use_counts[reg.nr] : 0; def_use_counts[reg.nr] : 0;
@ -164,33 +164,33 @@ protected:
struct vs_thread_payload : public thread_payload { struct vs_thread_payload : public thread_payload {
vs_thread_payload(const fs_visitor &v); vs_thread_payload(const fs_visitor &v);
fs_reg urb_handles; brw_reg urb_handles;
}; };
struct tcs_thread_payload : public thread_payload { struct tcs_thread_payload : public thread_payload {
tcs_thread_payload(const fs_visitor &v); tcs_thread_payload(const fs_visitor &v);
fs_reg patch_urb_output; brw_reg patch_urb_output;
fs_reg primitive_id; brw_reg primitive_id;
fs_reg icp_handle_start; brw_reg icp_handle_start;
}; };
struct tes_thread_payload : public thread_payload { struct tes_thread_payload : public thread_payload {
tes_thread_payload(const fs_visitor &v); tes_thread_payload(const fs_visitor &v);
fs_reg patch_urb_input; brw_reg patch_urb_input;
fs_reg primitive_id; brw_reg primitive_id;
fs_reg coords[3]; brw_reg coords[3];
fs_reg urb_output; brw_reg urb_output;
}; };
struct gs_thread_payload : public thread_payload { struct gs_thread_payload : public thread_payload {
gs_thread_payload(fs_visitor &v); gs_thread_payload(fs_visitor &v);
fs_reg urb_handles; brw_reg urb_handles;
fs_reg primitive_id; brw_reg primitive_id;
fs_reg instance_id; brw_reg instance_id;
fs_reg icp_handle_start; brw_reg icp_handle_start;
}; };
struct fs_thread_payload : public thread_payload { struct fs_thread_payload : public thread_payload {
@ -215,34 +215,34 @@ struct fs_thread_payload : public thread_payload {
struct cs_thread_payload : public thread_payload { struct cs_thread_payload : public thread_payload {
cs_thread_payload(const fs_visitor &v); cs_thread_payload(const fs_visitor &v);
void load_subgroup_id(const brw::fs_builder &bld, fs_reg &dest) const; void load_subgroup_id(const brw::fs_builder &bld, brw_reg &dest) const;
fs_reg local_invocation_id[3]; brw_reg local_invocation_id[3];
protected: protected:
fs_reg subgroup_id_; brw_reg subgroup_id_;
}; };
struct task_mesh_thread_payload : public cs_thread_payload { struct task_mesh_thread_payload : public cs_thread_payload {
task_mesh_thread_payload(fs_visitor &v); task_mesh_thread_payload(fs_visitor &v);
fs_reg extended_parameter_0; brw_reg extended_parameter_0;
fs_reg local_index; brw_reg local_index;
fs_reg inline_parameter; brw_reg inline_parameter;
fs_reg urb_output; brw_reg urb_output;
/* URB to read Task memory inputs. Only valid for MESH stage. */ /* URB to read Task memory inputs. Only valid for MESH stage. */
fs_reg task_urb_input; brw_reg task_urb_input;
}; };
struct bs_thread_payload : public thread_payload { struct bs_thread_payload : public thread_payload {
bs_thread_payload(const fs_visitor &v); bs_thread_payload(const fs_visitor &v);
fs_reg global_arg_ptr; brw_reg global_arg_ptr;
fs_reg local_arg_ptr; brw_reg local_arg_ptr;
void load_shader_type(const brw::fs_builder &bld, fs_reg &dest) const; void load_shader_type(const brw::fs_builder &bld, brw_reg &dest) const;
}; };
enum instruction_scheduler_mode { enum instruction_scheduler_mode {
@ -293,10 +293,10 @@ public:
void import_uniforms(fs_visitor *v); void import_uniforms(fs_visitor *v);
void VARYING_PULL_CONSTANT_LOAD(const brw::fs_builder &bld, void VARYING_PULL_CONSTANT_LOAD(const brw::fs_builder &bld,
const fs_reg &dst, const brw_reg &dst,
const fs_reg &surface, const brw_reg &surface,
const fs_reg &surface_handle, const brw_reg &surface_handle,
const fs_reg &varying_offset, const brw_reg &varying_offset,
uint32_t const_offset, uint32_t const_offset,
uint8_t alignment, uint8_t alignment,
unsigned components); unsigned components);
@ -324,7 +324,7 @@ public:
void calculate_payload_ranges(unsigned payload_node_count, void calculate_payload_ranges(unsigned payload_node_count,
int *payload_last_use_ip) const; int *payload_last_use_ip) const;
void assign_constant_locations(); void assign_constant_locations();
bool get_pull_locs(const fs_reg &src, unsigned *out_surf_index, bool get_pull_locs(const brw_reg &src, unsigned *out_surf_index,
unsigned *out_pull_index); unsigned *out_pull_index);
void invalidate_analysis(brw::analysis_dependency_class c); void invalidate_analysis(brw::analysis_dependency_class c);
@ -343,23 +343,23 @@ public:
void set_tcs_invocation_id(); void set_tcs_invocation_id();
fs_inst *emit_single_fb_write(const brw::fs_builder &bld, fs_inst *emit_single_fb_write(const brw::fs_builder &bld,
fs_reg color1, fs_reg color2, brw_reg color1, brw_reg color2,
fs_reg src0_alpha, unsigned components); brw_reg src0_alpha, unsigned components);
void do_emit_fb_writes(int nr_color_regions, bool replicate_alpha); void do_emit_fb_writes(int nr_color_regions, bool replicate_alpha);
void emit_fb_writes(); void emit_fb_writes();
void emit_urb_writes(const fs_reg &gs_vertex_count = fs_reg()); void emit_urb_writes(const brw_reg &gs_vertex_count = brw_reg());
void emit_gs_control_data_bits(const fs_reg &vertex_count); void emit_gs_control_data_bits(const brw_reg &vertex_count);
fs_reg gs_urb_channel_mask(const fs_reg &dword_index); brw_reg gs_urb_channel_mask(const brw_reg &dword_index);
fs_reg gs_urb_per_slot_dword_index(const fs_reg &vertex_count); brw_reg gs_urb_per_slot_dword_index(const brw_reg &vertex_count);
void emit_gs_thread_end(); void emit_gs_thread_end();
bool mark_last_urb_write_with_eot(); bool mark_last_urb_write_with_eot();
void emit_tcs_thread_end(); void emit_tcs_thread_end();
void emit_urb_fence(); void emit_urb_fence();
void emit_cs_terminate(); void emit_cs_terminate();
fs_reg interp_reg(const brw::fs_builder &bld, unsigned location, brw_reg interp_reg(const brw::fs_builder &bld, unsigned location,
unsigned channel, unsigned comp); unsigned channel, unsigned comp);
fs_reg per_primitive_reg(const brw::fs_builder &bld, brw_reg per_primitive_reg(const brw::fs_builder &bld,
int location, unsigned comp); int location, unsigned comp);
void dump_instruction_to_file(const fs_inst *inst, FILE *file, const brw::def_analysis *defs) const; void dump_instruction_to_file(const fs_inst *inst, FILE *file, const brw::def_analysis *defs) const;
@ -416,11 +416,11 @@ public:
*/ */
int *push_constant_loc; int *push_constant_loc;
fs_reg frag_depth; brw_reg frag_depth;
fs_reg frag_stencil; brw_reg frag_stencil;
fs_reg sample_mask; brw_reg sample_mask;
fs_reg outputs[VARYING_SLOT_MAX]; brw_reg outputs[VARYING_SLOT_MAX];
fs_reg dual_src_output; brw_reg dual_src_output;
int first_non_payload_grf; int first_non_payload_grf;
bool failed; bool failed;
@ -479,15 +479,15 @@ public:
bool source_depth_to_render_target; bool source_depth_to_render_target;
fs_reg pixel_x; brw_reg pixel_x;
fs_reg pixel_y; brw_reg pixel_y;
fs_reg pixel_z; brw_reg pixel_z;
fs_reg wpos_w; brw_reg wpos_w;
fs_reg pixel_w; brw_reg pixel_w;
fs_reg delta_xy[BRW_BARYCENTRIC_MODE_COUNT]; brw_reg delta_xy[BRW_BARYCENTRIC_MODE_COUNT];
fs_reg final_gs_vertex_count; brw_reg final_gs_vertex_count;
fs_reg control_data_bits; brw_reg control_data_bits;
fs_reg invocation_id; brw_reg invocation_id;
unsigned grf_used; unsigned grf_used;
bool spilled_any_registers; bool spilled_any_registers;
@ -597,15 +597,15 @@ private:
}; };
namespace brw { namespace brw {
fs_reg brw_reg
fetch_payload_reg(const brw::fs_builder &bld, uint8_t regs[2], fetch_payload_reg(const brw::fs_builder &bld, uint8_t regs[2],
brw_reg_type type = BRW_TYPE_F, brw_reg_type type = BRW_TYPE_F,
unsigned n = 1); unsigned n = 1);
fs_reg brw_reg
fetch_barycentric_reg(const brw::fs_builder &bld, uint8_t regs[2]); fetch_barycentric_reg(const brw::fs_builder &bld, uint8_t regs[2]);
inline fs_reg inline brw_reg
dynamic_msaa_flags(const struct brw_wm_prog_data *wm_prog_data) dynamic_msaa_flags(const struct brw_wm_prog_data *wm_prog_data)
{ {
return brw_uniform_reg(wm_prog_data->msaa_flags_param, BRW_TYPE_UD); return brw_uniform_reg(wm_prog_data->msaa_flags_param, BRW_TYPE_UD);
@ -621,8 +621,8 @@ namespace brw {
} }
void shuffle_from_32bit_read(const brw::fs_builder &bld, void shuffle_from_32bit_read(const brw::fs_builder &bld,
const fs_reg &dst, const brw_reg &dst,
const fs_reg &src, const brw_reg &src,
uint32_t first_component, uint32_t first_component,
uint32_t components); uint32_t components);
@ -636,7 +636,7 @@ void brw_compute_urb_setup_index(struct brw_wm_prog_data *wm_prog_data);
bool brw_nir_lower_simd(nir_shader *nir, unsigned dispatch_width); bool brw_nir_lower_simd(nir_shader *nir, unsigned dispatch_width);
fs_reg brw_sample_mask_reg(const brw::fs_builder &bld); brw_reg brw_sample_mask_reg(const brw::fs_builder &bld);
void brw_emit_predicate_on_sample_mask(const brw::fs_builder &bld, fs_inst *inst); void brw_emit_predicate_on_sample_mask(const brw::fs_builder &bld, fs_inst *inst);
int brw_get_subgroup_id_param_index(const intel_device_info *devinfo, int brw_get_subgroup_id_param_index(const intel_device_info *devinfo,

8
src/intel/compiler/brw_fs_bank_conflicts.cpp
View file

@ -480,7 +480,7 @@ namespace {
* possibly incur bank conflicts. * possibly incur bank conflicts.
*/ */
bool bool
is_grf(const fs_reg &r) is_grf(const brw_reg &r)
{ {
return r.file == VGRF || r.file == FIXED_GRF; return r.file == VGRF || r.file == FIXED_GRF;
} }
@ -492,7 +492,7 @@ namespace {
* allocation or whether it was part of a VGRF allocation. * allocation or whether it was part of a VGRF allocation.
*/ */
unsigned unsigned
reg_of(const fs_reg &r) reg_of(const brw_reg &r)
{ {
assert(is_grf(r)); assert(is_grf(r));
if (r.file == VGRF) if (r.file == VGRF)
@ -871,8 +871,8 @@ namespace {
* Apply the GRF atom permutation given by \p map to register \p r and * Apply the GRF atom permutation given by \p map to register \p r and
* return the result. * return the result.
*/ */
fs_reg brw_reg
transform(const partitioning &p, const permutation &map, fs_reg r) transform(const partitioning &p, const permutation &map, brw_reg r)
{ {
if (r.file == VGRF) { if (r.file == VGRF) {
const unsigned reg = reg_of(r); const unsigned reg = reg_of(r);

152
src/intel/compiler/brw_fs_builder.h
View file

@ -184,7 +184,7 @@ namespace brw {
* dispatch_width units (which is just enough space for one logical * dispatch_width units (which is just enough space for one logical
* component in this IR). * component in this IR).
*/ */
fs_reg brw_reg
vgrf(enum brw_reg_type type, unsigned n = 1) const vgrf(enum brw_reg_type type, unsigned n = 1) const
{ {
const unsigned unit = reg_unit(shader->devinfo); const unsigned unit = reg_unit(shader->devinfo);
@ -202,34 +202,34 @@ namespace brw {
/** /**
* Create a null register of floating type. * Create a null register of floating type.
*/ */
fs_reg brw_reg
null_reg_f() const null_reg_f() const
{ {
return fs_reg(retype(brw_null_reg(), BRW_TYPE_F)); return brw_reg(retype(brw_null_reg(), BRW_TYPE_F));
} }
fs_reg brw_reg
null_reg_df() const null_reg_df() const
{ {
return fs_reg(retype(brw_null_reg(), BRW_TYPE_DF)); return brw_reg(retype(brw_null_reg(), BRW_TYPE_DF));
} }
/** /**
* Create a null register of signed integer type. * Create a null register of signed integer type.
*/ */
fs_reg brw_reg
null_reg_d() const null_reg_d() const
{ {
return fs_reg(retype(brw_null_reg(), BRW_TYPE_D)); return brw_reg(retype(brw_null_reg(), BRW_TYPE_D));
} }
/** /**
* Create a null register of unsigned integer type. * Create a null register of unsigned integer type.
*/ */
fs_reg brw_reg
null_reg_ud() const null_reg_ud() const
{ {
return fs_reg(retype(brw_null_reg(), BRW_TYPE_UD)); return brw_reg(retype(brw_null_reg(), BRW_TYPE_UD));
} }
/** /**
@ -254,7 +254,7 @@ namespace brw {
* Create and insert a nullary instruction into the program. * Create and insert a nullary instruction into the program.
*/ */
fs_inst * fs_inst *
emit(enum opcode opcode, const fs_reg &dst) const emit(enum opcode opcode, const brw_reg &dst) const
{ {
return emit(fs_inst(opcode, dispatch_width(), dst)); return emit(fs_inst(opcode, dispatch_width(), dst));
} }
@ -263,7 +263,7 @@ namespace brw {
* Create and insert a unary instruction into the program. * Create and insert a unary instruction into the program.
*/ */
fs_inst * fs_inst *
emit(enum opcode opcode, const fs_reg &dst, const fs_reg &src0) const emit(enum opcode opcode, const brw_reg &dst, const brw_reg &src0) const
{ {
return emit(fs_inst(opcode, dispatch_width(), dst, src0)); return emit(fs_inst(opcode, dispatch_width(), dst, src0));
} }
@ -272,8 +272,8 @@ namespace brw {
* Create and insert a binary instruction into the program. * Create and insert a binary instruction into the program.
*/ */
fs_inst * fs_inst *
emit(enum opcode opcode, const fs_reg &dst, const fs_reg &src0, emit(enum opcode opcode, const brw_reg &dst, const brw_reg &src0,
const fs_reg &src1) const const brw_reg &src1) const
{ {
return emit(fs_inst(opcode, dispatch_width(), dst, return emit(fs_inst(opcode, dispatch_width(), dst,
src0, src1)); src0, src1));
@ -283,8 +283,8 @@ namespace brw {
* Create and insert a ternary instruction into the program. * Create and insert a ternary instruction into the program.
*/ */
fs_inst * fs_inst *
emit(enum opcode opcode, const fs_reg &dst, const fs_reg &src0, emit(enum opcode opcode, const brw_reg &dst, const brw_reg &src0,
const fs_reg &src1, const fs_reg &src2) const const brw_reg &src1, const brw_reg &src2) const
{ {
switch (opcode) { switch (opcode) {
case BRW_OPCODE_BFE: case BRW_OPCODE_BFE:
@ -307,7 +307,7 @@ namespace brw {
* into the program. * into the program.
*/ */
fs_inst * fs_inst *
emit(enum opcode opcode, const fs_reg &dst, const fs_reg srcs[], emit(enum opcode opcode, const brw_reg &dst, const brw_reg srcs[],
unsigned n) const unsigned n) const
{ {
/* Use the emit() methods for specific operand counts to ensure that /* Use the emit() methods for specific operand counts to ensure that
@ -352,8 +352,8 @@ namespace brw {
* Generally useful to get the minimum or maximum of two values. * Generally useful to get the minimum or maximum of two values.
*/ */
fs_inst * fs_inst *
emit_minmax(const fs_reg &dst, const fs_reg &src0, emit_minmax(const brw_reg &dst, const brw_reg &src0,
const fs_reg &src1, brw_conditional_mod mod) const const brw_reg &src1, brw_conditional_mod mod) const
{ {
assert(mod == BRW_CONDITIONAL_GE || mod == BRW_CONDITIONAL_L); assert(mod == BRW_CONDITIONAL_GE || mod == BRW_CONDITIONAL_L);
@ -367,8 +367,8 @@ namespace brw {
/** /**
* Copy any live channel from \p src to the first channel of the result. * Copy any live channel from \p src to the first channel of the result.
*/ */
fs_reg brw_reg
emit_uniformize(const fs_reg &src) const emit_uniformize(const brw_reg &src) const
{ {
/* FIXME: We use a vector chan_index and dst to allow constant and /* FIXME: We use a vector chan_index and dst to allow constant and
* copy propagration to move result all the way into the consuming * copy propagration to move result all the way into the consuming
@ -378,37 +378,37 @@ namespace brw {
* should go back to scalar destinations here. * should go back to scalar destinations here.
*/ */
const fs_builder ubld = exec_all(); const fs_builder ubld = exec_all();
const fs_reg chan_index = vgrf(BRW_TYPE_UD); const brw_reg chan_index = vgrf(BRW_TYPE_UD);
const fs_reg dst = vgrf(src.type); const brw_reg dst = vgrf(src.type);
ubld.emit(SHADER_OPCODE_FIND_LIVE_CHANNEL, chan_index); ubld.emit(SHADER_OPCODE_FIND_LIVE_CHANNEL, chan_index);
ubld.emit(SHADER_OPCODE_BROADCAST, dst, src, component(chan_index, 0)); ubld.emit(SHADER_OPCODE_BROADCAST, dst, src, component(chan_index, 0));
return fs_reg(component(dst, 0)); return brw_reg(component(dst, 0));
} }
fs_reg brw_reg
move_to_vgrf(const fs_reg &src, unsigned num_components) const move_to_vgrf(const brw_reg &src, unsigned num_components) const
{ {
fs_reg *const src_comps = new fs_reg[num_components]; brw_reg *const src_comps = new brw_reg[num_components];
for (unsigned i = 0; i < num_components; i++) for (unsigned i = 0; i < num_components; i++)
src_comps[i] = offset(src, dispatch_width(), i); src_comps[i] = offset(src, dispatch_width(), i);
const fs_reg dst = vgrf(src.type, num_components); const brw_reg dst = vgrf(src.type, num_components);
LOAD_PAYLOAD(dst, src_comps, num_components, 0); LOAD_PAYLOAD(dst, src_comps, num_components, 0);
delete[] src_comps; delete[] src_comps;
return fs_reg(dst); return brw_reg(dst);
} }
void void
emit_scan_step(enum opcode opcode, brw_conditional_mod mod, emit_scan_step(enum opcode opcode, brw_conditional_mod mod,
const fs_reg &tmp, const brw_reg &tmp,
unsigned left_offset, unsigned left_stride, unsigned left_offset, unsigned left_stride,
unsigned right_offset, unsigned right_stride) const unsigned right_offset, unsigned right_stride) const
{ {
fs_reg left, right; brw_reg left, right;
left = horiz_stride(horiz_offset(tmp, left_offset), left_stride); left = horiz_stride(horiz_offset(tmp, left_offset), left_stride);
right = horiz_stride(horiz_offset(tmp, right_offset), right_stride); right = horiz_stride(horiz_offset(tmp, right_offset), right_stride);
if ((tmp.type == BRW_TYPE_Q || tmp.type == BRW_TYPE_UQ) && if ((tmp.type == BRW_TYPE_Q || tmp.type == BRW_TYPE_UQ) &&
@ -430,13 +430,13 @@ namespace brw {
/* We treat the bottom 32 bits as unsigned regardless of /* We treat the bottom 32 bits as unsigned regardless of
* whether or not the integer as a whole is signed. * whether or not the integer as a whole is signed.
*/ */
fs_reg right_low = subscript(right, BRW_TYPE_UD, 0); brw_reg right_low = subscript(right, BRW_TYPE_UD, 0);
fs_reg left_low = subscript(left, BRW_TYPE_UD, 0); brw_reg left_low = subscript(left, BRW_TYPE_UD, 0);
/* The upper bits get the same sign as the 64-bit type */ /* The upper bits get the same sign as the 64-bit type */
brw_reg_type type32 = brw_type_with_size(tmp.type, 32); brw_reg_type type32 = brw_type_with_size(tmp.type, 32);
fs_reg right_high = subscript(right, type32, 1); brw_reg right_high = subscript(right, type32, 1);
fs_reg left_high = subscript(left, type32, 1); brw_reg left_high = subscript(left, type32, 1);
/* Build up our comparison: /* Build up our comparison:
* *
@ -468,7 +468,7 @@ namespace brw {
} }
void void
emit_scan(enum opcode opcode, const fs_reg &tmp, emit_scan(enum opcode opcode, const brw_reg &tmp,
unsigned cluster_size, brw_conditional_mod mod) const unsigned cluster_size, brw_conditional_mod mod) const
{ {
assert(dispatch_width() >= 8); assert(dispatch_width() >= 8);
@ -479,8 +479,8 @@ namespace brw {
if (dispatch_width() * brw_type_size_bytes(tmp.type) > 2 * REG_SIZE) { if (dispatch_width() * brw_type_size_bytes(tmp.type) > 2 * REG_SIZE) {
const unsigned half_width = dispatch_width() / 2; const unsigned half_width = dispatch_width() / 2;
const fs_builder ubld = exec_all().group(half_width, 0); const fs_builder ubld = exec_all().group(half_width, 0);
fs_reg left = tmp; brw_reg left = tmp;
fs_reg right = horiz_offset(tmp, half_width); brw_reg right = horiz_offset(tmp, half_width);
ubld.emit_scan(opcode, left, cluster_size, mod); ubld.emit_scan(opcode, left, cluster_size, mod);
ubld.emit_scan(opcode, right, cluster_size, mod); ubld.emit_scan(opcode, right, cluster_size, mod);
if (cluster_size > half_width) { if (cluster_size > half_width) {
@ -547,15 +547,15 @@ namespace brw {
*/ */
#define _ALU1(prefix, op) \ #define _ALU1(prefix, op) \
fs_inst * \ fs_inst * \
op(const fs_reg &dst, const fs_reg &src0) const \ op(const brw_reg &dst, const brw_reg &src0) const \
{ \ { \
assert(_dispatch_width == 1 || \ assert(_dispatch_width == 1 || \
(dst.file >= VGRF && dst.stride != 0) || \ (dst.file >= VGRF && dst.stride != 0) || \
(dst.file < VGRF && dst.hstride != 0)); \ (dst.file < VGRF && dst.hstride != 0)); \
return emit(prefix##op, dst, src0); \ return emit(prefix##op, dst, src0); \
} \ } \
fs_reg \ brw_reg \
op(const fs_reg &src0, fs_inst **out = NULL) const \ op(const brw_reg &src0, fs_inst **out = NULL) const \
{ \ { \
fs_inst *inst = op(vgrf(src0.type), src0); \ fs_inst *inst = op(vgrf(src0.type), src0); \
if (out) *out = inst; \ if (out) *out = inst; \
@ -565,12 +565,12 @@ namespace brw {
#define VIRT1(op) _ALU1(SHADER_OPCODE_, op) #define VIRT1(op) _ALU1(SHADER_OPCODE_, op)
fs_inst * fs_inst *
alu2(opcode op, const fs_reg &dst, const fs_reg &src0, const fs_reg &src1) const alu2(opcode op, const brw_reg &dst, const brw_reg &src0, const brw_reg &src1) const
{ {
return emit(op, dst, src0, src1); return emit(op, dst, src0, src1);
} }
fs_reg brw_reg
alu2(opcode op, const fs_reg &src0, const fs_reg &src1, fs_inst **out = NULL) const alu2(opcode op, const brw_reg &src0, const brw_reg &src1, fs_inst **out = NULL) const
{ {
enum brw_reg_type inferred_dst_type = enum brw_reg_type inferred_dst_type =
brw_type_larger_of(src0.type, src1.type); brw_type_larger_of(src0.type, src1.type);
@ -581,12 +581,12 @@ namespace brw {
#define _ALU2(prefix, op) \ #define _ALU2(prefix, op) \
fs_inst * \ fs_inst * \
op(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1) const \ op(const brw_reg &dst, const brw_reg &src0, const brw_reg &src1) const \
{ \ { \
return alu2(prefix##op, dst, src0, src1); \ return alu2(prefix##op, dst, src0, src1); \
} \ } \
fs_reg \ brw_reg \
op(const fs_reg &src0, const fs_reg &src1, fs_inst **out = NULL) const \ op(const brw_reg &src0, const brw_reg &src1, fs_inst **out = NULL) const \
{ \ { \
return alu2(prefix##op, src0, src1, out); \ return alu2(prefix##op, src0, src1, out); \
} }
@ -595,7 +595,7 @@ namespace brw {
#define ALU2_ACC(op) \ #define ALU2_ACC(op) \
fs_inst * \ fs_inst * \
op(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1) const \ op(const brw_reg &dst, const brw_reg &src0, const brw_reg &src1) const \
{ \ { \
fs_inst *inst = emit(BRW_OPCODE_##op, dst, src0, src1); \ fs_inst *inst = emit(BRW_OPCODE_##op, dst, src0, src1); \
inst->writes_accumulator = true; \ inst->writes_accumulator = true; \
@ -604,8 +604,8 @@ namespace brw {
#define ALU3(op) \ #define ALU3(op) \
fs_inst * \ fs_inst * \
op(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1, \ op(const brw_reg &dst, const brw_reg &src0, const brw_reg &src1, \
const fs_reg &src2) const \ const brw_reg &src2) const \
{ \ { \
return emit(BRW_OPCODE_##op, dst, src0, src1, src2); \ return emit(BRW_OPCODE_##op, dst, src0, src1, src2); \
} }
@ -672,13 +672,13 @@ namespace brw {
/** @} */ /** @} */
fs_inst * fs_inst *
ADD(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1) const ADD(const brw_reg &dst, const brw_reg &src0, const brw_reg &src1) const
{ {
return alu2(BRW_OPCODE_ADD, dst, src0, src1); return alu2(BRW_OPCODE_ADD, dst, src0, src1);
} }
fs_reg brw_reg
ADD(const fs_reg &src0, const fs_reg &src1, fs_inst **out = NULL) const ADD(const brw_reg &src0, const brw_reg &src1, fs_inst **out = NULL) const
{ {
if (src1.file == IMM && src1.ud == 0 && !out) if (src1.file == IMM && src1.ud == 0 && !out)
return src0; return src0;
@ -692,7 +692,7 @@ namespace brw {
* the flag register with the packed 16 bits of the result. * the flag register with the packed 16 bits of the result.
*/ */
fs_inst * fs_inst *
CMP(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1, CMP(const brw_reg &dst, const brw_reg &src0, const brw_reg &src1,
brw_conditional_mod condition) const brw_conditional_mod condition) const
{ {
/* Take the instruction: /* Take the instruction:
@ -718,7 +718,7 @@ namespace brw {
* CMPN: Behaves like CMP, but produces true if src1 is NaN. * CMPN: Behaves like CMP, but produces true if src1 is NaN.
*/ */
fs_inst * fs_inst *
CMPN(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1, CMPN(const brw_reg &dst, const brw_reg &src0, const brw_reg &src1,
brw_conditional_mod condition) const brw_conditional_mod condition) const
{ {
/* Take the instruction: /* Take the instruction:
@ -753,8 +753,8 @@ namespace brw {
* CSEL: dst = src2 <op> 0.0f ? src0 : src1 * CSEL: dst = src2 <op> 0.0f ? src0 : src1
*/ */
fs_inst * fs_inst *
CSEL(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1, CSEL(const brw_reg &dst, const brw_reg &src0, const brw_reg &src1,
const fs_reg &src2, brw_conditional_mod condition) const const brw_reg &src2, brw_conditional_mod condition) const
{ {
return set_condmod(condition, return set_condmod(condition,
emit(BRW_OPCODE_CSEL, emit(BRW_OPCODE_CSEL,
@ -768,8 +768,8 @@ namespace brw {
* Emit a linear interpolation instruction. * Emit a linear interpolation instruction.
*/ */
fs_inst * fs_inst *
LRP(const fs_reg &dst, const fs_reg &x, const fs_reg &y, LRP(const brw_reg &dst, const brw_reg &x, const brw_reg &y,
const fs_reg &a) const const brw_reg &a) const
{ {
if (shader->devinfo->ver <= 10) { if (shader->devinfo->ver <= 10) {
/* The LRP instruction actually does op1 * op0 + op2 * (1 - op0), so /* The LRP instruction actually does op1 * op0 + op2 * (1 - op0), so
@ -779,14 +779,14 @@ namespace brw {
} else { } else {
/* We can't use the LRP instruction. Emit x*(1-a) + y*a. */ /* We can't use the LRP instruction. Emit x*(1-a) + y*a. */
const fs_reg y_times_a = vgrf(dst.type); const brw_reg y_times_a = vgrf(dst.type);
const fs_reg one_minus_a = vgrf(dst.type); const brw_reg one_minus_a = vgrf(dst.type);
const fs_reg x_times_one_minus_a = vgrf(dst.type); const brw_reg x_times_one_minus_a = vgrf(dst.type);
MUL(y_times_a, y, a); MUL(y_times_a, y, a);
ADD(one_minus_a, negate(a), brw_imm_f(1.0f)); ADD(one_minus_a, negate(a), brw_imm_f(1.0f));
MUL(x_times_one_minus_a, x, fs_reg(one_minus_a)); MUL(x_times_one_minus_a, x, brw_reg(one_minus_a));
return ADD(dst, fs_reg(x_times_one_minus_a), fs_reg(y_times_a)); return ADD(dst, brw_reg(x_times_one_minus_a), brw_reg(y_times_a));
} }
} }
@ -794,7 +794,7 @@ namespace brw {
* Collect a number of registers in a contiguous range of registers. * Collect a number of registers in a contiguous range of registers.
*/ */
fs_inst * fs_inst *
LOAD_PAYLOAD(const fs_reg &dst, const fs_reg *src, LOAD_PAYLOAD(const brw_reg &dst, const brw_reg *src,
unsigned sources, unsigned header_size) const unsigned sources, unsigned header_size) const
{ {
fs_inst *inst = emit(SHADER_OPCODE_LOAD_PAYLOAD, dst, src, sources); fs_inst *inst = emit(SHADER_OPCODE_LOAD_PAYLOAD, dst, src, sources);
@ -809,7 +809,7 @@ namespace brw {
} }
fs_inst * fs_inst *
VEC(const fs_reg &dst, const fs_reg *src, unsigned sources) const VEC(const brw_reg &dst, const brw_reg *src, unsigned sources) const
{ {
return sources == 1 ? MOV(dst, src[0]) return sources == 1 ? MOV(dst, src[0])
: LOAD_PAYLOAD(dst, src, sources, 0); : LOAD_PAYLOAD(dst, src, sources, 0);
@ -822,7 +822,7 @@ namespace brw {
} }
fs_inst * fs_inst *
UNDEF(const fs_reg &dst) const UNDEF(const brw_reg &dst) const
{ {
assert(dst.file == VGRF); assert(dst.file == VGRF);
assert(dst.offset % REG_SIZE == 0); assert(dst.offset % REG_SIZE == 0);
@ -834,7 +834,7 @@ namespace brw {
} }
fs_inst * fs_inst *
DPAS(const fs_reg &dst, const fs_reg &src0, const fs_reg &src1, const fs_reg &src2, DPAS(const brw_reg &dst, const brw_reg &src0, const brw_reg &src1, const brw_reg &src2,
unsigned sdepth, unsigned rcount) const unsigned sdepth, unsigned rcount) const
{ {
assert(_dispatch_width == 8 * reg_unit(shader->devinfo)); assert(_dispatch_width == 8 * reg_unit(shader->devinfo));
@ -868,14 +868,14 @@ namespace brw {
* Workaround for negation of UD registers. See comment in * Workaround for negation of UD registers. See comment in
* fs_generator::generate_code() for more details. * fs_generator::generate_code() for more details.
*/ */
fs_reg brw_reg
fix_unsigned_negate(const fs_reg &src) const fix_unsigned_negate(const brw_reg &src) const
{ {
if (src.type == BRW_TYPE_UD && if (src.type == BRW_TYPE_UD &&
src.negate) { src.negate) {
fs_reg temp = vgrf(BRW_TYPE_UD); brw_reg temp = vgrf(BRW_TYPE_UD);
MOV(temp, src); MOV(temp, src);
return fs_reg(temp); return brw_reg(temp);
} else { } else {
return src; return src;
} }
@ -885,8 +885,8 @@ namespace brw {
* Workaround for source register modes not supported by the ternary * Workaround for source register modes not supported by the ternary
* instruction encoding. * instruction encoding.
*/ */
fs_reg brw_reg
fix_3src_operand(const fs_reg &src) const fix_3src_operand(const brw_reg &src) const
{ {
switch (src.file) { switch (src.file) {
case FIXED_GRF: case FIXED_GRF:
@ -905,7 +905,7 @@ namespace brw {
break; break;
} }
fs_reg expanded = vgrf(src.type); brw_reg expanded = vgrf(src.type);
MOV(expanded, src); MOV(expanded, src);
return expanded; return expanded;
} }
@ -925,8 +925,8 @@ namespace brw {
}; };
} }
static inline fs_reg static inline brw_reg
offset(const fs_reg &reg, const brw::fs_builder &bld, unsigned delta) offset(const brw_reg &reg, const brw::fs_builder &bld, unsigned delta)
{ {
return offset(reg, bld.dispatch_width(), delta); return offset(reg, bld.dispatch_width(), delta);
} }

12
src/intel/compiler/brw_fs_combine_constants.cpp
View file

@ -1150,7 +1150,7 @@ struct register_allocation {
uint16_t avail; uint16_t avail;
}; };
static fs_reg static brw_reg
allocate_slots(struct register_allocation *regs, unsigned num_regs, allocate_slots(struct register_allocation *regs, unsigned num_regs,
unsigned bytes, unsigned align_bytes, unsigned bytes, unsigned align_bytes,
brw::simple_allocator &alloc) brw::simple_allocator &alloc)
@ -1172,7 +1172,7 @@ allocate_slots(struct register_allocation *regs, unsigned num_regs,
regs[i].avail &= ~(mask << j); regs[i].avail &= ~(mask << j);
fs_reg reg = brw_vgrf(regs[i].nr, BRW_TYPE_F); brw_reg reg = brw_vgrf(regs[i].nr, BRW_TYPE_F);
reg.offset = j * 2; reg.offset = j * 2;
return reg; return reg;
@ -1243,7 +1243,7 @@ parcel_out_registers(struct imm *imm, unsigned len, const bblock_t *cur_block,
*/ */
const unsigned width = ver == 8 && imm[i].is_half_float ? 2 : 1; const unsigned width = ver == 8 && imm[i].is_half_float ? 2 : 1;
const fs_reg reg = allocate_slots(regs, num_regs, const brw_reg reg = allocate_slots(regs, num_regs,
imm[i].size * width, imm[i].size * width,
get_alignment_for_imm(&imm[i]), get_alignment_for_imm(&imm[i]),
alloc); alloc);
@ -1569,7 +1569,7 @@ brw_fs_opt_combine_constants(fs_visitor &s)
const uint32_t width = 1; const uint32_t width = 1;
const fs_builder ibld = fs_builder(&s, width).at(insert_block, n).exec_all(); const fs_builder ibld = fs_builder(&s, width).at(insert_block, n).exec_all();
fs_reg reg = brw_vgrf(imm->nr, BRW_TYPE_F); brw_reg reg = brw_vgrf(imm->nr, BRW_TYPE_F);
reg.offset = imm->subreg_offset; reg.offset = imm->subreg_offset;
reg.stride = 0; reg.stride = 0;
@ -1591,7 +1591,7 @@ brw_fs_opt_combine_constants(fs_visitor &s)
/* Rewrite the immediate sources to refer to the new GRFs. */ /* Rewrite the immediate sources to refer to the new GRFs. */
for (int i = 0; i < table.len; i++) { for (int i = 0; i < table.len; i++) {
foreach_list_typed(reg_link, link, link, table.imm[i].uses) { foreach_list_typed(reg_link, link, link, table.imm[i].uses) {
fs_reg *reg = &link->inst->src[link->src]; brw_reg *reg = &link->inst->src[link->src];
if (link->inst->opcode == BRW_OPCODE_SEL) { if (link->inst->opcode == BRW_OPCODE_SEL) {
if (link->type == either_type) { if (link->type == either_type) {
@ -1709,7 +1709,7 @@ brw_fs_opt_combine_constants(fs_visitor &s)
inst->conditional_mod == BRW_CONDITIONAL_GE || inst->conditional_mod == BRW_CONDITIONAL_GE ||
inst->conditional_mod == BRW_CONDITIONAL_L); inst->conditional_mod == BRW_CONDITIONAL_L);
fs_reg temp = inst->src[0]; brw_reg temp = inst->src[0];
inst->src[0] = inst->src[1]; inst->src[0] = inst->src[1];
inst->src[1] = temp; inst->src[1] = temp;

32
src/intel/compiler/brw_fs_copy_propagation.cpp
View file

@ -46,8 +46,8 @@ namespace { /* avoid conflict with opt_copy_propagation_elements */
struct acp_entry { struct acp_entry {
struct rb_node by_dst; struct rb_node by_dst;
struct rb_node by_src; struct rb_node by_src;
fs_reg dst; brw_reg dst;
fs_reg src; brw_reg src;
unsigned global_idx; unsigned global_idx;
unsigned size_written; unsigned size_written;
unsigned size_read; unsigned size_read;
@ -338,7 +338,7 @@ fs_copy_prop_dataflow::fs_copy_prop_dataflow(linear_ctx *lin_ctx, cfg_t *cfg,
* Like reg_offset, but register must be VGRF or FIXED_GRF. * Like reg_offset, but register must be VGRF or FIXED_GRF.
*/ */
static inline unsigned static inline unsigned
grf_reg_offset(const fs_reg &r) grf_reg_offset(const brw_reg &r)
{ {
return (r.file == VGRF ? 0 : r.nr) * REG_SIZE + return (r.file == VGRF ? 0 : r.nr) * REG_SIZE +
r.offset + r.offset +
@ -349,7 +349,7 @@ grf_reg_offset(const fs_reg &r)
* Like regions_overlap, but register must be VGRF or FIXED_GRF. * Like regions_overlap, but register must be VGRF or FIXED_GRF.
*/ */
static inline bool static inline bool
grf_regions_overlap(const fs_reg &r, unsigned dr, const fs_reg &s, unsigned ds) grf_regions_overlap(const brw_reg &r, unsigned dr, const brw_reg &s, unsigned ds)
{ {
return reg_space(r) == reg_space(s) && return reg_space(r) == reg_space(s) &&
!(grf_reg_offset(r) + dr <= grf_reg_offset(s) || !(grf_reg_offset(r) + dr <= grf_reg_offset(s) ||
@ -793,7 +793,7 @@ try_copy_propagate(const brw_compiler *compiler, fs_inst *inst,
* regioning restrictions that apply to integer types smaller than a dword. * regioning restrictions that apply to integer types smaller than a dword.
* See BSpec #56640 for details. * See BSpec #56640 for details.
*/ */
const fs_reg tmp = horiz_stride(entry->src, inst->src[arg].stride); const brw_reg tmp = horiz_stride(entry->src, inst->src[arg].stride);
if (has_subdword_integer_region_restriction(devinfo, inst, &tmp, 1)) if (has_subdword_integer_region_restriction(devinfo, inst, &tmp, 1))
return false; return false;
@ -936,7 +936,7 @@ try_copy_propagate(const brw_compiler *compiler, fs_inst *inst,
} }
static bool static bool
try_constant_propagate_value(fs_reg val, brw_reg_type dst_type, try_constant_propagate_value(brw_reg val, brw_reg_type dst_type,
fs_inst *inst, int arg) fs_inst *inst, int arg)
{ {
bool progress = false; bool progress = false;
@ -1394,7 +1394,7 @@ opt_copy_propagation_local(const brw_compiler *compiler, linear_ctx *lin_ctx,
inst->src[i].is_contiguous())) { inst->src[i].is_contiguous())) {
const brw_reg_type t = i < inst->header_size ? const brw_reg_type t = i < inst->header_size ?
BRW_TYPE_UD : inst->src[i].type; BRW_TYPE_UD : inst->src[i].type;
fs_reg dst = byte_offset(retype(inst->dst, t), offset); brw_reg dst = byte_offset(retype(inst->dst, t), offset);
if (!dst.equals(inst->src[i])) { if (!dst.equals(inst->src[i])) {
acp_entry *entry = linear_zalloc(lin_ctx, acp_entry); acp_entry *entry = linear_zalloc(lin_ctx, acp_entry);
entry->dst = dst; entry->dst = dst;
@ -1486,7 +1486,7 @@ brw_fs_opt_copy_propagation(fs_visitor &s)
static bool static bool
try_copy_propagate_def(const brw_compiler *compiler, try_copy_propagate_def(const brw_compiler *compiler,
const brw::simple_allocator &alloc, const brw::simple_allocator &alloc,
fs_inst *def, const fs_reg &val, fs_inst *def, const brw_reg &val,
fs_inst *inst, int arg, fs_inst *inst, int arg,
uint8_t max_polygons) uint8_t max_polygons)
{ {
@ -1716,7 +1716,7 @@ try_copy_propagate_def(const brw_compiler *compiler,
} }
static bool static bool
try_constant_propagate_def(fs_inst *def, fs_reg val, fs_inst *inst, int arg) try_constant_propagate_def(fs_inst *def, brw_reg val, fs_inst *inst, int arg)
{ {
/* Bail if inst is reading more than a single vector component of entry */ /* Bail if inst is reading more than a single vector component of entry */
if (inst->size_read(arg) > def->dst.component_size(inst->exec_size)) if (inst->size_read(arg) > def->dst.component_size(inst->exec_size))
@ -1728,8 +1728,8 @@ try_constant_propagate_def(fs_inst *def, fs_reg val, fs_inst *inst, int arg)
/** /**
* Handle cases like UW subreads of a UD immediate, with an offset. * Handle cases like UW subreads of a UD immediate, with an offset.
*/ */
static fs_reg static brw_reg
extract_imm(fs_reg val, brw_reg_type type, unsigned offset) extract_imm(brw_reg val, brw_reg_type type, unsigned offset)
{ {
assert(val.file == IMM); assert(val.file == IMM);
@ -1748,16 +1748,16 @@ extract_imm(fs_reg val, brw_reg_type type, unsigned offset)
val.d = (val.d << (bitsize * (32/bitsize - 1 - offset))) >> ((32/bitsize - 1) * bitsize); val.d = (val.d << (bitsize * (32/bitsize - 1 - offset))) >> ((32/bitsize - 1) * bitsize);
break; break;
default: default:
return fs_reg(); return brw_reg();
} }
return val; return val;
} }
static fs_reg static brw_reg
find_value_for_offset(fs_inst *def, const fs_reg &src, unsigned src_size) find_value_for_offset(fs_inst *def, const brw_reg &src, unsigned src_size)
{ {
fs_reg val; brw_reg val;
switch (def->opcode) { switch (def->opcode) {
case BRW_OPCODE_MOV: case BRW_OPCODE_MOV:
@ -1837,7 +1837,7 @@ brw_fs_opt_copy_propagation_defs(fs_visitor &s)
} }
} }
fs_reg val = brw_reg val =
find_value_for_offset(def, inst->src[i], inst->size_read(i)); find_value_for_offset(def, inst->src[i], inst->size_read(i));
if (val.file == IMM) { if (val.file == IMM) {

12
src/intel/compiler/brw_fs_cse.cpp
View file

@ -162,8 +162,8 @@ local_only(const fs_inst *inst)
static bool static bool
operands_match(const fs_inst *a, const fs_inst *b, bool *negate) operands_match(const fs_inst *a, const fs_inst *b, bool *negate)
{ {
fs_reg *xs = a->src; brw_reg *xs = a->src;
fs_reg *ys = b->src; brw_reg *ys = b->src;
if (a->opcode == BRW_OPCODE_MAD) { if (a->opcode == BRW_OPCODE_MAD) {
return xs[0].equals(ys[0]) && return xs[0].equals(ys[0]) &&
@ -251,7 +251,7 @@ instructions_match(fs_inst *a, fs_inst *b, bool *negate)
#define HASH(hash, data) XXH32(&(data), sizeof(data), hash) #define HASH(hash, data) XXH32(&(data), sizeof(data), hash)
uint32_t uint32_t
hash_reg(uint32_t hash, const fs_reg &r) hash_reg(uint32_t hash, const brw_reg &r)
{ {
struct { struct {
uint64_t u64; uint64_t u64;
@ -314,7 +314,7 @@ hash_inst(const void *v)
/* Canonicalize negations on either source (or both) and commutatively /* Canonicalize negations on either source (or both) and commutatively
* combine the hashes for both sources. * combine the hashes for both sources.
*/ */
fs_reg src[2] = { inst->src[0], inst->src[1] }; brw_reg src[2] = { inst->src[0], inst->src[1] };
uint32_t src_hash[2]; uint32_t src_hash[2];
for (int i = 0; i < 2; i++) { for (int i = 0; i < 2; i++) {
@ -383,8 +383,8 @@ remap_sources(fs_visitor &s, const brw::def_analysis &defs,
if (def_block->end_ip_delta) if (def_block->end_ip_delta)
s.cfg->adjust_block_ips(); s.cfg->adjust_block_ips();
fs_reg neg = brw_vgrf(new_nr, BRW_TYPE_F); brw_reg neg = brw_vgrf(new_nr, BRW_TYPE_F);
fs_reg tmp = dbld.MOV(negate(neg)); brw_reg tmp = dbld.MOV(negate(neg));
inst->src[i].nr = tmp.nr; inst->src[i].nr = tmp.nr;
remap_table[old_nr] = tmp.nr; remap_table[old_nr] = tmp.nr;
} else { } else {

4
src/intel/compiler/brw_fs_dead_code_eliminate.cpp
View file

@ -100,8 +100,8 @@ brw_fs_opt_dead_code_eliminate(fs_visitor &s)
if (!result_live && if (!result_live &&
(can_omit_write(inst) || can_eliminate(devinfo, inst, flag_live))) { (can_omit_write(inst) || can_eliminate(devinfo, inst, flag_live))) {
inst->dst = fs_reg(spread(retype(brw_null_reg(), inst->dst.type), inst->dst = brw_reg(spread(retype(brw_null_reg(), inst->dst.type),
inst->dst.stride)); inst->dst.stride));
progress = true; progress = true;
} }
} }

1
src/intel/compiler/brw_fs_generator.cpp
View file

@ -66,7 +66,6 @@ brw_math_function(enum opcode op)
static struct brw_reg static struct brw_reg
normalize_brw_reg_for_encoding(brw_reg *reg) normalize_brw_reg_for_encoding(brw_reg *reg)
{ {
struct brw_reg brw_reg; struct brw_reg brw_reg;

10
src/intel/compiler/brw_fs_live_variables.cpp
View file

@ -54,7 +54,7 @@ using namespace brw;
void void
fs_live_variables::setup_one_read(struct block_data *bd, fs_live_variables::setup_one_read(struct block_data *bd,
int ip, const fs_reg &reg) int ip, const brw_reg &reg)
{ {
int var = var_from_reg(reg); int var = var_from_reg(reg);
assert(var < num_vars); assert(var < num_vars);
@ -72,7 +72,7 @@ fs_live_variables::setup_one_read(struct block_data *bd,
void void
fs_live_variables::setup_one_write(struct block_data *bd, fs_inst *inst, fs_live_variables::setup_one_write(struct block_data *bd, fs_inst *inst,
int ip, const fs_reg &reg) int ip, const brw_reg &reg)
{ {
int var = var_from_reg(reg); int var = var_from_reg(reg);
assert(var < num_vars); assert(var < num_vars);
@ -115,7 +115,7 @@ fs_live_variables::setup_def_use()
foreach_inst_in_block(fs_inst, inst, block) { foreach_inst_in_block(fs_inst, inst, block) {
/* Set use[] for this instruction */ /* Set use[] for this instruction */
for (unsigned int i = 0; i < inst->sources; i++) { for (unsigned int i = 0; i < inst->sources; i++) {
fs_reg reg = inst->src[i]; brw_reg reg = inst->src[i];
if (reg.file != VGRF) if (reg.file != VGRF)
continue; continue;
@ -130,7 +130,7 @@ fs_live_variables::setup_def_use()
/* Set def[] for this instruction */ /* Set def[] for this instruction */
if (inst->dst.file == VGRF) { if (inst->dst.file == VGRF) {
fs_reg reg = inst->dst; brw_reg reg = inst->dst;
for (unsigned j = 0; j < regs_written(inst); j++) { for (unsigned j = 0; j < regs_written(inst); j++) {
setup_one_write(bd, inst, ip, reg); setup_one_write(bd, inst, ip, reg);
reg.offset += REG_SIZE; reg.offset += REG_SIZE;
@ -317,7 +317,7 @@ fs_live_variables::~fs_live_variables()
static bool static bool
check_register_live_range(const fs_live_variables *live, int ip, check_register_live_range(const fs_live_variables *live, int ip,
const fs_reg &reg, unsigned n) const brw_reg &reg, unsigned n)
{ {
const unsigned var = live->var_from_reg(reg); const unsigned var = live->var_from_reg(reg);

6
src/intel/compiler/brw_fs_live_variables.h
View file

@ -92,7 +92,7 @@ public:
bool vars_interfere(int a, int b) const; bool vars_interfere(int a, int b) const;
bool vgrfs_interfere(int a, int b) const; bool vgrfs_interfere(int a, int b) const;
int var_from_reg(const fs_reg &reg) const int var_from_reg(const brw_reg &reg) const
{ {
return var_from_vgrf[reg.nr] + reg.offset / REG_SIZE; return var_from_vgrf[reg.nr] + reg.offset / REG_SIZE;
} }
@ -132,9 +132,9 @@ public:
protected: protected:
void setup_def_use(); void setup_def_use();
void setup_one_read(struct block_data *bd, int ip, const fs_reg &reg); void setup_one_read(struct block_data *bd, int ip, const brw_reg &reg);
void setup_one_write(struct block_data *bd, fs_inst *inst, int ip, void setup_one_write(struct block_data *bd, fs_inst *inst, int ip,
const fs_reg &reg); const brw_reg &reg);
void compute_live_variables(); void compute_live_variables();
void compute_start_end(); void compute_start_end();

60
src/intel/compiler/brw_fs_lower.cpp
View file

@ -37,10 +37,10 @@ brw_fs_lower_constant_loads(fs_visitor &s)
const unsigned block_sz = 64; /* Fetch one cacheline at a time. */ const unsigned block_sz = 64; /* Fetch one cacheline at a time. */
const fs_builder ubld = ibld.exec_all().group(block_sz / 4, 0); const fs_builder ubld = ibld.exec_all().group(block_sz / 4, 0);
const fs_reg dst = ubld.vgrf(BRW_TYPE_UD); const brw_reg dst = ubld.vgrf(BRW_TYPE_UD);
const unsigned base = pull_index * 4; const unsigned base = pull_index * 4;
fs_reg srcs[PULL_UNIFORM_CONSTANT_SRCS]; brw_reg srcs[PULL_UNIFORM_CONSTANT_SRCS];
srcs[PULL_UNIFORM_CONSTANT_SRC_SURFACE] = brw_imm_ud(index); srcs[PULL_UNIFORM_CONSTANT_SRC_SURFACE] = brw_imm_ud(index);
srcs[PULL_UNIFORM_CONSTANT_SRC_OFFSET] = brw_imm_ud(base & ~(block_sz - 1)); srcs[PULL_UNIFORM_CONSTANT_SRC_OFFSET] = brw_imm_ud(base & ~(block_sz - 1));
srcs[PULL_UNIFORM_CONSTANT_SRC_SIZE] = brw_imm_ud(block_sz); srcs[PULL_UNIFORM_CONSTANT_SRC_SIZE] = brw_imm_ud(block_sz);
@ -66,7 +66,7 @@ brw_fs_lower_constant_loads(fs_visitor &s)
s.VARYING_PULL_CONSTANT_LOAD(ibld, inst->dst, s.VARYING_PULL_CONSTANT_LOAD(ibld, inst->dst,
brw_imm_ud(index), brw_imm_ud(index),
fs_reg() /* surface_handle */, brw_reg() /* surface_handle */,
inst->src[1], inst->src[1],
pull_index * 4, 4, 1); pull_index * 4, 4, 1);
inst->remove(block); inst->remove(block);
@ -90,7 +90,7 @@ brw_fs_lower_load_payload(fs_visitor &s)
assert(inst->dst.file == VGRF); assert(inst->dst.file == VGRF);
assert(inst->saturate == false); assert(inst->saturate == false);
fs_reg dst = inst->dst; brw_reg dst = inst->dst;
const fs_builder ibld(&s, block, inst); const fs_builder ibld(&s, block, inst);
const fs_builder ubld = ibld.exec_all(); const fs_builder ubld = ibld.exec_all();
@ -188,7 +188,7 @@ brw_fs_lower_csel(fs_visitor &s)
const fs_builder ibld(&s, block, inst); const fs_builder ibld(&s, block, inst);
/* CSEL: dst = src2 <op> 0 ? src0 : src1 */ /* CSEL: dst = src2 <op> 0 ? src0 : src1 */
fs_reg zero = brw_imm_reg(orig_type); brw_reg zero = brw_imm_reg(orig_type);
ibld.CMP(retype(brw_null_reg(), orig_type), ibld.CMP(retype(brw_null_reg(), orig_type),
inst->src[2], zero, inst->conditional_mod); inst->src[2], zero, inst->conditional_mod);
@ -250,7 +250,7 @@ brw_fs_lower_sub_sat(fs_visitor &s)
*/ */
if (inst->exec_size == 8 && inst->src[0].type != BRW_TYPE_Q && if (inst->exec_size == 8 && inst->src[0].type != BRW_TYPE_Q &&
inst->src[0].type != BRW_TYPE_UQ) { inst->src[0].type != BRW_TYPE_UQ) {
fs_reg acc = retype(brw_acc_reg(inst->exec_size), brw_reg acc = retype(brw_acc_reg(inst->exec_size),
inst->src[1].type); inst->src[1].type);
ibld.MOV(acc, inst->src[1]); ibld.MOV(acc, inst->src[1]);
@ -263,11 +263,11 @@ brw_fs_lower_sub_sat(fs_visitor &s)
*/ */
fs_inst *add; fs_inst *add;
fs_reg tmp = ibld.vgrf(inst->src[0].type); brw_reg tmp = ibld.vgrf(inst->src[0].type);
ibld.SHR(tmp, inst->src[1], brw_imm_d(1)); ibld.SHR(tmp, inst->src[1], brw_imm_d(1));
fs_reg s1_sub_t = ibld.ADD(inst->src[1], negate(tmp)); brw_reg s1_sub_t = ibld.ADD(inst->src[1], negate(tmp));
fs_reg sat_s0_sub_t = ibld.ADD(inst->src[0], negate(tmp), &add); brw_reg sat_s0_sub_t = ibld.ADD(inst->src[0], negate(tmp), &add);
add->saturate = true; add->saturate = true;
add = ibld.ADD(inst->dst, sat_s0_sub_t, negate(s1_sub_t)); add = ibld.ADD(inst->dst, sat_s0_sub_t, negate(s1_sub_t));
@ -331,8 +331,8 @@ brw_fs_lower_barycentrics(fs_visitor &s)
switch (inst->opcode) { switch (inst->opcode) {
case BRW_OPCODE_PLN: { case BRW_OPCODE_PLN: {
assert(inst->exec_size == 16); assert(inst->exec_size == 16);
const fs_reg tmp = ibld.vgrf(inst->src[1].type, 2); const brw_reg tmp = ibld.vgrf(inst->src[1].type, 2);
fs_reg srcs[4]; brw_reg srcs[4];
for (unsigned i = 0; i < ARRAY_SIZE(srcs); i++) for (unsigned i = 0; i < ARRAY_SIZE(srcs); i++)
srcs[i] = horiz_offset(offset(inst->src[1], ibld, i % 2), srcs[i] = horiz_offset(offset(inst->src[1], ibld, i % 2),
@ -348,7 +348,7 @@ brw_fs_lower_barycentrics(fs_visitor &s)
case FS_OPCODE_INTERPOLATE_AT_SHARED_OFFSET: case FS_OPCODE_INTERPOLATE_AT_SHARED_OFFSET:
case FS_OPCODE_INTERPOLATE_AT_PER_SLOT_OFFSET: { case FS_OPCODE_INTERPOLATE_AT_PER_SLOT_OFFSET: {
assert(inst->exec_size == 16); assert(inst->exec_size == 16);
const fs_reg tmp = ibld.vgrf(inst->dst.type, 2); const brw_reg tmp = ibld.vgrf(inst->dst.type, 2);
for (unsigned i = 0; i < 2; i++) { for (unsigned i = 0; i < 2; i++) {
for (unsigned g = 0; g < inst->exec_size / 8; g++) { for (unsigned g = 0; g < inst->exec_size / 8; g++) {
@ -386,8 +386,8 @@ lower_derivative(fs_visitor &s, bblock_t *block, fs_inst *inst,
unsigned swz0, unsigned swz1) unsigned swz0, unsigned swz1)
{ {
const fs_builder ubld = fs_builder(&s, block, inst).exec_all(); const fs_builder ubld = fs_builder(&s, block, inst).exec_all();
const fs_reg tmp0 = ubld.vgrf(inst->src[0].type); const brw_reg tmp0 = ubld.vgrf(inst->src[0].type);
const fs_reg tmp1 = ubld.vgrf(inst->src[0].type); const brw_reg tmp1 = ubld.vgrf(inst->src[0].type);
ubld.emit(SHADER_OPCODE_QUAD_SWIZZLE, tmp0, inst->src[0], brw_imm_ud(swz0)); ubld.emit(SHADER_OPCODE_QUAD_SWIZZLE, tmp0, inst->src[0], brw_imm_ud(swz0));
ubld.emit(SHADER_OPCODE_QUAD_SWIZZLE, tmp1, inst->src[0], brw_imm_ud(swz1)); ubld.emit(SHADER_OPCODE_QUAD_SWIZZLE, tmp1, inst->src[0], brw_imm_ud(swz1));
@ -469,7 +469,7 @@ brw_fs_lower_find_live_channel(fs_visitor &s)
const fs_builder ubld = fs_builder(&s, block, inst).exec_all().group(1, 0); const fs_builder ubld = fs_builder(&s, block, inst).exec_all().group(1, 0);
fs_reg exec_mask = ubld.vgrf(BRW_TYPE_UD); brw_reg exec_mask = ubld.vgrf(BRW_TYPE_UD);
ubld.UNDEF(exec_mask); ubld.UNDEF(exec_mask);
ubld.emit(SHADER_OPCODE_READ_ARCH_REG, exec_mask, ubld.emit(SHADER_OPCODE_READ_ARCH_REG, exec_mask,
retype(brw_mask_reg(0), retype(brw_mask_reg(0),
@ -483,7 +483,7 @@ brw_fs_lower_find_live_channel(fs_visitor &s)
* will appear at the front of the mask. * will appear at the front of the mask.
*/ */
if (!(first && packed_dispatch)) { if (!(first && packed_dispatch)) {
fs_reg mask = ubld.vgrf(BRW_TYPE_UD); brw_reg mask = ubld.vgrf(BRW_TYPE_UD);
ubld.UNDEF(mask); ubld.UNDEF(mask);
ubld.emit(SHADER_OPCODE_READ_ARCH_REG, mask, ubld.emit(SHADER_OPCODE_READ_ARCH_REG, mask,
retype(brw_sr0_reg(vmask ? 3 : 2), retype(brw_sr0_reg(vmask ? 3 : 2),
@ -506,7 +506,7 @@ brw_fs_lower_find_live_channel(fs_visitor &s)
break; break;
case SHADER_OPCODE_FIND_LAST_LIVE_CHANNEL: { case SHADER_OPCODE_FIND_LAST_LIVE_CHANNEL: {
fs_reg tmp = ubld.vgrf(BRW_TYPE_UD); brw_reg tmp = ubld.vgrf(BRW_TYPE_UD);
ubld.UNDEF(tmp); ubld.UNDEF(tmp);
ubld.LZD(tmp, exec_mask); ubld.LZD(tmp, exec_mask);
ubld.ADD(inst->dst, negate(tmp), brw_imm_uw(31)); ubld.ADD(inst->dst, negate(tmp), brw_imm_uw(31));
@ -553,15 +553,15 @@ brw_fs_lower_sends_overlapping_payload(fs_visitor &s)
const unsigned arg = inst->mlen < inst->ex_mlen ? 2 : 3; const unsigned arg = inst->mlen < inst->ex_mlen ? 2 : 3;
const unsigned len = MIN2(inst->mlen, inst->ex_mlen); const unsigned len = MIN2(inst->mlen, inst->ex_mlen);
fs_reg tmp = brw_vgrf(s.alloc.allocate(len), brw_reg tmp = brw_vgrf(s.alloc.allocate(len),
BRW_TYPE_UD); BRW_TYPE_UD);
/* Sadly, we've lost all notion of channels and bit sizes at this /* Sadly, we've lost all notion of channels and bit sizes at this
* point. Just WE_all it. * point. Just WE_all it.
*/ */
const fs_builder ibld = fs_builder(&s, block, inst).exec_all().group(16, 0); const fs_builder ibld = fs_builder(&s, block, inst).exec_all().group(16, 0);
fs_reg copy_src = retype(inst->src[arg], BRW_TYPE_UD); brw_reg copy_src = retype(inst->src[arg], BRW_TYPE_UD);
fs_reg copy_dst = tmp; brw_reg copy_dst = tmp;
for (unsigned i = 0; i < len; i += 2) { for (unsigned i = 0; i < len; i += 2) {
if (len == i + 1) { if (len == i + 1) {
/* Only one register left; do SIMD8 */ /* Only one register left; do SIMD8 */
@ -696,7 +696,7 @@ brw_fs_lower_alu_restrictions(fs_visitor &s)
static void static void
brw_fs_lower_vgrf_to_fixed_grf(const struct intel_device_info *devinfo, fs_inst *inst, brw_fs_lower_vgrf_to_fixed_grf(const struct intel_device_info *devinfo, fs_inst *inst,
fs_reg *reg, bool compressed) brw_reg *reg, bool compressed)
{ {
if (reg->file != VGRF) if (reg->file != VGRF)
return; return;
@ -801,7 +801,7 @@ brw_fs_lower_load_subgroup_invocation(fs_visitor &s)
if (inst->exec_size == 8) { if (inst->exec_size == 8) {
assert(inst->dst.type == BRW_TYPE_UD); assert(inst->dst.type == BRW_TYPE_UD);
fs_reg uw = retype(inst->dst, BRW_TYPE_UW); brw_reg uw = retype(inst->dst, BRW_TYPE_UW);
ubld8.MOV(uw, brw_imm_v(0x76543210)); ubld8.MOV(uw, brw_imm_v(0x76543210));
ubld8.MOV(inst->dst, uw); ubld8.MOV(inst->dst, uw);
} else { } else {
@ -852,12 +852,12 @@ brw_fs_lower_indirect_mov(fs_visitor &s)
/* Extract unaligned part */ /* Extract unaligned part */
uint16_t extra_offset = inst->src[0].offset & 0x1; uint16_t extra_offset = inst->src[0].offset & 0x1;
fs_reg offset = ibld.ADD(inst->src[1], brw_imm_uw(extra_offset)); brw_reg offset = ibld.ADD(inst->src[1], brw_imm_uw(extra_offset));
/* Check if offset is odd or even so that we can choose either high or /* Check if offset is odd or even so that we can choose either high or
* low byte from the result. * low byte from the result.
*/ */
fs_reg is_odd = ibld.AND(offset, brw_imm_ud(1)); brw_reg is_odd = ibld.AND(offset, brw_imm_ud(1));
/* Make sure offset is word (2-bytes) aligned */ /* Make sure offset is word (2-bytes) aligned */
offset = ibld.AND(offset, brw_imm_uw(~1)); offset = ibld.AND(offset, brw_imm_uw(~1));
@ -865,24 +865,24 @@ brw_fs_lower_indirect_mov(fs_visitor &s)
/* Indirect addressing(vx1 and vxh) not supported with UB/B datatype for /* Indirect addressing(vx1 and vxh) not supported with UB/B datatype for
* Src0, so change data type for src0 and dst to UW. * Src0, so change data type for src0 and dst to UW.
*/ */
fs_reg dst = ibld.vgrf(BRW_TYPE_UW); brw_reg dst = ibld.vgrf(BRW_TYPE_UW);
/* Substract unaligned offset from src0 offset since we already /* Substract unaligned offset from src0 offset since we already
* accounted unaligned part in the indirect byte offset. * accounted unaligned part in the indirect byte offset.
*/ */
fs_reg start = retype(inst->src[0], BRW_TYPE_UW); brw_reg start = retype(inst->src[0], BRW_TYPE_UW);
start.offset &= ~extra_offset; start.offset &= ~extra_offset;
/* Adjust length to account extra offset. */ /* Adjust length to account extra offset. */
assert(inst->src[2].file == IMM); assert(inst->src[2].file == IMM);
fs_reg length = brw_imm_ud(inst->src[2].ud + extra_offset); brw_reg length = brw_imm_ud(inst->src[2].ud + extra_offset);
ibld.emit(SHADER_OPCODE_MOV_INDIRECT, dst, start, offset, length); ibld.emit(SHADER_OPCODE_MOV_INDIRECT, dst, start, offset, length);
/* Select high byte if offset is odd otherwise select low byte. */ /* Select high byte if offset is odd otherwise select low byte. */
fs_reg lo = ibld.AND(dst, brw_imm_uw(0xff)); brw_reg lo = ibld.AND(dst, brw_imm_uw(0xff));
fs_reg hi = ibld.SHR(dst, brw_imm_uw(8)); brw_reg hi = ibld.SHR(dst, brw_imm_uw(8));
fs_reg result = ibld.vgrf(BRW_TYPE_UW); brw_reg result = ibld.vgrf(BRW_TYPE_UW);
ibld.CSEL(result, hi, lo, is_odd, BRW_CONDITIONAL_NZ); ibld.CSEL(result, hi, lo, is_odd, BRW_CONDITIONAL_NZ);
/* Extra MOV needed here to convert back to the corresponding B type */ /* Extra MOV needed here to convert back to the corresponding B type */

38
src/intel/compiler/brw_fs_lower_dpas.cpp
View file

@ -24,16 +24,16 @@ f16_using_mac(const fs_builder &bld, fs_inst *inst)
const brw_reg_type src1_type = BRW_TYPE_HF; const brw_reg_type src1_type = BRW_TYPE_HF;
const brw_reg_type src2_type = BRW_TYPE_HF; const brw_reg_type src2_type = BRW_TYPE_HF;
const fs_reg dest = inst->dst; const brw_reg dest = inst->dst;
fs_reg src0 = inst->src[0]; brw_reg src0 = inst->src[0];
const fs_reg src1 = retype(inst->src[1], src1_type); const brw_reg src1 = retype(inst->src[1], src1_type);
const fs_reg src2 = retype(inst->src[2], src2_type); const brw_reg src2 = retype(inst->src[2], src2_type);
const unsigned dest_stride = const unsigned dest_stride =
dest.type == BRW_TYPE_HF ? REG_SIZE / 2 : REG_SIZE; dest.type == BRW_TYPE_HF ? REG_SIZE / 2 : REG_SIZE;
for (unsigned r = 0; r < inst->rcount; r++) { for (unsigned r = 0; r < inst->rcount; r++) {
fs_reg temp = bld.vgrf(BRW_TYPE_HF); brw_reg temp = bld.vgrf(BRW_TYPE_HF);
for (unsigned subword = 0; subword < 2; subword++) { for (unsigned subword = 0; subword < 2; subword++) {
for (unsigned s = 0; s < inst->sdepth; s++) { for (unsigned s = 0; s < inst->sdepth; s++) {
@ -50,7 +50,7 @@ f16_using_mac(const fs_builder &bld, fs_inst *inst)
*/ */
if (s == 0 && subword == 0) { if (s == 0 && subword == 0) {
const unsigned acc_width = 8; const unsigned acc_width = 8;
fs_reg acc = suboffset(retype(brw_acc_reg(inst->exec_size), BRW_TYPE_UD), brw_reg acc = suboffset(retype(brw_acc_reg(inst->exec_size), BRW_TYPE_UD),
inst->group % acc_width); inst->group % acc_width);
if (bld.shader->devinfo->verx10 >= 125) { if (bld.shader->devinfo->verx10 >= 125) {
@ -69,7 +69,7 @@ f16_using_mac(const fs_builder &bld, fs_inst *inst)
->writes_accumulator = true; ->writes_accumulator = true;
} else { } else {
fs_reg result; brw_reg result;
/* As mentioned above, the MAC had an optional, explicit /* As mentioned above, the MAC had an optional, explicit
* destination register. Various optimization passes are not * destination register. Various optimization passes are not
@ -96,7 +96,7 @@ f16_using_mac(const fs_builder &bld, fs_inst *inst)
if (!src0.is_null()) { if (!src0.is_null()) {
if (src0_type != BRW_TYPE_HF) { if (src0_type != BRW_TYPE_HF) {
fs_reg temp2 = bld.vgrf(src0_type); brw_reg temp2 = bld.vgrf(src0_type);
bld.MOV(temp2, temp); bld.MOV(temp2, temp);
@ -134,10 +134,10 @@ int8_using_dp4a(const fs_builder &bld, fs_inst *inst)
const brw_reg_type src2_type = inst->src[2].type == BRW_TYPE_UB const brw_reg_type src2_type = inst->src[2].type == BRW_TYPE_UB
? BRW_TYPE_UD : BRW_TYPE_D; ? BRW_TYPE_UD : BRW_TYPE_D;
fs_reg dest = inst->dst; brw_reg dest = inst->dst;
fs_reg src0 = inst->src[0]; brw_reg src0 = inst->src[0];
const fs_reg src1 = retype(inst->src[1], src1_type); const brw_reg src1 = retype(inst->src[1], src1_type);
const fs_reg src2 = retype(inst->src[2], src2_type); const brw_reg src2 = retype(inst->src[2], src2_type);
const unsigned dest_stride = reg_unit(bld.shader->devinfo) * REG_SIZE; const unsigned dest_stride = reg_unit(bld.shader->devinfo) * REG_SIZE;
@ -183,10 +183,10 @@ int8_using_mul_add(const fs_builder &bld, fs_inst *inst)
const brw_reg_type src2_type = inst->src[2].type == BRW_TYPE_UB const brw_reg_type src2_type = inst->src[2].type == BRW_TYPE_UB
? BRW_TYPE_UD : BRW_TYPE_D; ? BRW_TYPE_UD : BRW_TYPE_D;
fs_reg dest = inst->dst; brw_reg dest = inst->dst;
fs_reg src0 = inst->src[0]; brw_reg src0 = inst->src[0];
const fs_reg src1 = retype(inst->src[1], src1_type); const brw_reg src1 = retype(inst->src[1], src1_type);
const fs_reg src2 = retype(inst->src[2], src2_type); const brw_reg src2 = retype(inst->src[2], src2_type);
const unsigned dest_stride = REG_SIZE; const unsigned dest_stride = REG_SIZE;
@ -199,9 +199,9 @@ int8_using_mul_add(const fs_builder &bld, fs_inst *inst)
} }
for (unsigned s = 0; s < inst->sdepth; s++) { for (unsigned s = 0; s < inst->sdepth; s++) {
fs_reg temp1 = bld.vgrf(BRW_TYPE_UD); brw_reg temp1 = bld.vgrf(BRW_TYPE_UD);
fs_reg temp2 = bld.vgrf(BRW_TYPE_UD); brw_reg temp2 = bld.vgrf(BRW_TYPE_UD);
fs_reg temp3 = bld.vgrf(BRW_TYPE_UD, 2); brw_reg temp3 = bld.vgrf(BRW_TYPE_UD, 2);
const brw_reg_type temp_type = const brw_reg_type temp_type =
(inst->src[1].type == BRW_TYPE_B || (inst->src[1].type == BRW_TYPE_B ||
inst->src[2].type == BRW_TYPE_B) inst->src[2].type == BRW_TYPE_B)

20
src/intel/compiler/brw_fs_lower_integer_multiplication.cpp
View file

@ -207,14 +207,14 @@ brw_fs_lower_mul_dword_inst(fs_visitor &s, fs_inst *inst, bblock_t *block)
*/ */
bool needs_mov = false; bool needs_mov = false;
fs_reg orig_dst = inst->dst; brw_reg orig_dst = inst->dst;
/* Get a new VGRF for the "low" 32x16-bit multiplication result if /* Get a new VGRF for the "low" 32x16-bit multiplication result if
* reusing the original destination is impossible due to hardware * reusing the original destination is impossible due to hardware
* restrictions, source/destination overlap, or it being the null * restrictions, source/destination overlap, or it being the null
* register. * register.
*/ */
fs_reg low = inst->dst; brw_reg low = inst->dst;
if (orig_dst.is_null() || if (orig_dst.is_null() ||
regions_overlap(inst->dst, inst->size_written, regions_overlap(inst->dst, inst->size_written,
inst->src[0], inst->size_read(0)) || inst->src[0], inst->size_read(0)) ||
@ -227,7 +227,7 @@ brw_fs_lower_mul_dword_inst(fs_visitor &s, fs_inst *inst, bblock_t *block)
} }
/* Get a new VGRF but keep the same stride as inst->dst */ /* Get a new VGRF but keep the same stride as inst->dst */
fs_reg high = brw_vgrf(s.alloc.allocate(regs_written(inst)), inst->dst.type); brw_reg high = brw_vgrf(s.alloc.allocate(regs_written(inst)), inst->dst.type);
high.stride = inst->dst.stride; high.stride = inst->dst.stride;
high.offset = inst->dst.offset % REG_SIZE; high.offset = inst->dst.offset % REG_SIZE;
@ -319,19 +319,19 @@ brw_fs_lower_mul_qword_inst(fs_visitor &s, fs_inst *inst, bblock_t *block)
unsigned int q_regs = regs_written(inst); unsigned int q_regs = regs_written(inst);
unsigned int d_regs = (q_regs + 1) / 2; unsigned int d_regs = (q_regs + 1) / 2;
fs_reg bd = brw_vgrf(s.alloc.allocate(q_regs), BRW_TYPE_UQ); brw_reg bd = brw_vgrf(s.alloc.allocate(q_regs), BRW_TYPE_UQ);
fs_reg ad = brw_vgrf(s.alloc.allocate(d_regs), BRW_TYPE_UD); brw_reg ad = brw_vgrf(s.alloc.allocate(d_regs), BRW_TYPE_UD);
fs_reg bc = brw_vgrf(s.alloc.allocate(d_regs), BRW_TYPE_UD); brw_reg bc = brw_vgrf(s.alloc.allocate(d_regs), BRW_TYPE_UD);
/* Here we need the full 64 bit result for 32b * 32b. */ /* Here we need the full 64 bit result for 32b * 32b. */
if (devinfo->has_integer_dword_mul) { if (devinfo->has_integer_dword_mul) {
ibld.MUL(bd, subscript(inst->src[0], BRW_TYPE_UD, 0), ibld.MUL(bd, subscript(inst->src[0], BRW_TYPE_UD, 0),
subscript(inst->src[1], BRW_TYPE_UD, 0)); subscript(inst->src[1], BRW_TYPE_UD, 0));
} else { } else {
fs_reg bd_high = brw_vgrf(s.alloc.allocate(d_regs), BRW_TYPE_UD); brw_reg bd_high = brw_vgrf(s.alloc.allocate(d_regs), BRW_TYPE_UD);
fs_reg bd_low = brw_vgrf(s.alloc.allocate(d_regs), BRW_TYPE_UD); brw_reg bd_low = brw_vgrf(s.alloc.allocate(d_regs), BRW_TYPE_UD);
const unsigned acc_width = reg_unit(devinfo) * 8; const unsigned acc_width = reg_unit(devinfo) * 8;
fs_reg acc = suboffset(retype(brw_acc_reg(inst->exec_size), BRW_TYPE_UD), brw_reg acc = suboffset(retype(brw_acc_reg(inst->exec_size), BRW_TYPE_UD),
inst->group % acc_width); inst->group % acc_width);
fs_inst *mul = ibld.MUL(acc, fs_inst *mul = ibld.MUL(acc,
@ -390,7 +390,7 @@ brw_fs_lower_mulh_inst(fs_visitor &s, fs_inst *inst, bblock_t *block)
/* Should have been lowered to 8-wide. */ /* Should have been lowered to 8-wide. */
assert(inst->exec_size <= brw_fs_get_lowered_simd_width(&s, inst)); assert(inst->exec_size <= brw_fs_get_lowered_simd_width(&s, inst));
const unsigned acc_width = reg_unit(devinfo) * 8; const unsigned acc_width = reg_unit(devinfo) * 8;
const fs_reg acc = suboffset(retype(brw_acc_reg(inst->exec_size), inst->dst.type), const brw_reg acc = suboffset(retype(brw_acc_reg(inst->exec_size), inst->dst.type),
inst->group % acc_width); inst->group % acc_width);
fs_inst *mul = ibld.MUL(acc, inst->src[0], inst->src[1]); fs_inst *mul = ibld.MUL(acc, inst->src[0], inst->src[1]);
ibld.MACH(inst->dst, inst->src[0], inst->src[1]); ibld.MACH(inst->dst, inst->src[0], inst->src[1]);

2
src/intel/compiler/brw_fs_lower_pack.cpp
View file

@ -40,7 +40,7 @@ brw_fs_lower_pack(fs_visitor &s)
assert(inst->dst.file == VGRF); assert(inst->dst.file == VGRF);
assert(inst->saturate == false); assert(inst->saturate == false);
fs_reg dst = inst->dst; brw_reg dst = inst->dst;
const fs_builder ibld(&s, block, inst); const fs_builder ibld(&s, block, inst);
/* The lowering generates 2 instructions for what was previously 1. This /* The lowering generates 2 instructions for what was previously 1. This

14
src/intel/compiler/brw_fs_lower_regioning.cpp
View file

@ -464,7 +464,7 @@ namespace brw {
brw_type_size_bytes(inst->src[i].type) == get_exec_type_size(inst)); brw_type_size_bytes(inst->src[i].type) == get_exec_type_size(inst));
const fs_builder ibld(v, block, inst); const fs_builder ibld(v, block, inst);
const fs_reg tmp = ibld.vgrf(get_exec_type(inst)); const brw_reg tmp = ibld.vgrf(get_exec_type(inst));
lower_instruction(v, block, ibld.MOV(tmp, inst->src[i])); lower_instruction(v, block, ibld.MOV(tmp, inst->src[i]));
inst->src[i] = tmp; inst->src[i] = tmp;
@ -495,7 +495,7 @@ namespace {
const unsigned stride = const unsigned stride =
brw_type_size_bytes(inst->dst.type) * inst->dst.stride <= brw_type_size_bytes(type) ? 1 : brw_type_size_bytes(inst->dst.type) * inst->dst.stride <= brw_type_size_bytes(type) ? 1 :
brw_type_size_bytes(inst->dst.type) * inst->dst.stride / brw_type_size_bytes(type); brw_type_size_bytes(inst->dst.type) * inst->dst.stride / brw_type_size_bytes(type);
fs_reg tmp = ibld.vgrf(type, stride); brw_reg tmp = ibld.vgrf(type, stride);
ibld.UNDEF(tmp); ibld.UNDEF(tmp);
tmp = horiz_stride(tmp, stride); tmp = horiz_stride(tmp, stride);
@ -549,7 +549,7 @@ namespace {
inst->exec_size * stride * inst->exec_size * stride *
brw_type_size_bytes(inst->src[i].type), brw_type_size_bytes(inst->src[i].type),
reg_unit(devinfo) * REG_SIZE) * reg_unit(devinfo); reg_unit(devinfo) * REG_SIZE) * reg_unit(devinfo);
fs_reg tmp = brw_vgrf(v->alloc.allocate(size), inst->src[i].type); brw_reg tmp = brw_vgrf(v->alloc.allocate(size), inst->src[i].type);
ibld.UNDEF(tmp); ibld.UNDEF(tmp);
tmp = byte_offset(horiz_stride(tmp, stride), tmp = byte_offset(horiz_stride(tmp, stride),
required_src_byte_offset(devinfo, inst, i)); required_src_byte_offset(devinfo, inst, i));
@ -560,7 +560,7 @@ namespace {
const brw_reg_type raw_type = brw_int_type(MIN2(brw_type_size_bytes(tmp.type), 4), const brw_reg_type raw_type = brw_int_type(MIN2(brw_type_size_bytes(tmp.type), 4),
false); false);
const unsigned n = brw_type_size_bytes(tmp.type) / brw_type_size_bytes(raw_type); const unsigned n = brw_type_size_bytes(tmp.type) / brw_type_size_bytes(raw_type);
fs_reg raw_src = inst->src[i]; brw_reg raw_src = inst->src[i];
raw_src.negate = false; raw_src.negate = false;
raw_src.abs = false; raw_src.abs = false;
@ -578,7 +578,7 @@ namespace {
/* Point the original instruction at the temporary, making sure to keep /* Point the original instruction at the temporary, making sure to keep
* any source modifiers in the instruction. * any source modifiers in the instruction.
*/ */
fs_reg lower_src = tmp; brw_reg lower_src = tmp;
lower_src.negate = inst->src[i].negate; lower_src.negate = inst->src[i].negate;
lower_src.abs = inst->src[i].abs; lower_src.abs = inst->src[i].abs;
inst->src[i] = lower_src; inst->src[i] = lower_src;
@ -607,7 +607,7 @@ namespace {
const unsigned stride = required_dst_byte_stride(inst) / const unsigned stride = required_dst_byte_stride(inst) /
brw_type_size_bytes(inst->dst.type); brw_type_size_bytes(inst->dst.type);
assert(stride > 0); assert(stride > 0);
fs_reg tmp = ibld.vgrf(inst->dst.type, stride); brw_reg tmp = ibld.vgrf(inst->dst.type, stride);
ibld.UNDEF(tmp); ibld.UNDEF(tmp);
tmp = horiz_stride(tmp, stride); tmp = horiz_stride(tmp, stride);
@ -665,7 +665,7 @@ namespace {
const unsigned n = get_exec_type_size(inst) / brw_type_size_bytes(raw_type); const unsigned n = get_exec_type_size(inst) / brw_type_size_bytes(raw_type);
const fs_builder ibld(v, block, inst); const fs_builder ibld(v, block, inst);
fs_reg tmp = ibld.vgrf(inst->dst.type, inst->dst.stride); brw_reg tmp = ibld.vgrf(inst->dst.type, inst->dst.stride);
ibld.UNDEF(tmp); ibld.UNDEF(tmp);
tmp = horiz_stride(tmp, inst->dst.stride); tmp = horiz_stride(tmp, inst->dst.stride);

18
src/intel/compiler/brw_fs_lower_simd_width.cpp
View file

@ -500,19 +500,19 @@ needs_src_copy(const fs_builder &lbld, const fs_inst *inst, unsigned i)
* lbld.group() from the i-th source region of instruction \p inst and return * lbld.group() from the i-th source region of instruction \p inst and return
* it as result in packed form. * it as result in packed form.
*/ */
static fs_reg static brw_reg
emit_unzip(const fs_builder &lbld, fs_inst *inst, unsigned i) emit_unzip(const fs_builder &lbld, fs_inst *inst, unsigned i)
{ {
assert(lbld.group() >= inst->group); assert(lbld.group() >= inst->group);
/* Specified channel group from the source region. */ /* Specified channel group from the source region. */
const fs_reg src = horiz_offset(inst->src[i], lbld.group() - inst->group); const brw_reg src = horiz_offset(inst->src[i], lbld.group() - inst->group);
if (needs_src_copy(lbld, inst, i)) { if (needs_src_copy(lbld, inst, i)) {
const unsigned num_components = inst->components_read(i); const unsigned num_components = inst->components_read(i);
const fs_reg tmp = lbld.vgrf(inst->src[i].type, num_components); const brw_reg tmp = lbld.vgrf(inst->src[i].type, num_components);
fs_reg comps[num_components]; brw_reg comps[num_components];
for (unsigned k = 0; k < num_components; ++k) for (unsigned k = 0; k < num_components; ++k)
comps[k] = offset(src, inst->exec_size, k); comps[k] = offset(src, inst->exec_size, k);
lbld.VEC(tmp, comps, num_components); lbld.VEC(tmp, comps, num_components);
@ -585,7 +585,7 @@ needs_dst_copy(const fs_builder &lbld, const fs_inst *inst)
* inserted using \p lbld_before and any copy instructions required for * inserted using \p lbld_before and any copy instructions required for
* zipping up the destination of \p inst will be inserted using \p lbld_after. * zipping up the destination of \p inst will be inserted using \p lbld_after.
*/ */
static fs_reg static brw_reg
emit_zip(const fs_builder &lbld_before, const fs_builder &lbld_after, emit_zip(const fs_builder &lbld_before, const fs_builder &lbld_after,
fs_inst *inst) fs_inst *inst)
{ {
@ -596,7 +596,7 @@ emit_zip(const fs_builder &lbld_before, const fs_builder &lbld_after,
const struct intel_device_info *devinfo = lbld_before.shader->devinfo; const struct intel_device_info *devinfo = lbld_before.shader->devinfo;
/* Specified channel group from the destination region. */ /* Specified channel group from the destination region. */
const fs_reg dst = horiz_offset(inst->dst, lbld_after.group() - inst->group); const brw_reg dst = horiz_offset(inst->dst, lbld_after.group() - inst->group);
if (!needs_dst_copy(lbld_after, inst)) { if (!needs_dst_copy(lbld_after, inst)) {
/* No need to allocate a temporary for the lowered instruction, just /* No need to allocate a temporary for the lowered instruction, just
@ -611,7 +611,7 @@ emit_zip(const fs_builder &lbld_before, const fs_builder &lbld_after,
const unsigned dst_size = (inst->size_written - residency_size) / const unsigned dst_size = (inst->size_written - residency_size) /
inst->dst.component_size(inst->exec_size); inst->dst.component_size(inst->exec_size);
const fs_reg tmp = lbld_after.vgrf(inst->dst.type, const brw_reg tmp = lbld_after.vgrf(inst->dst.type,
dst_size + inst->has_sampler_residency()); dst_size + inst->has_sampler_residency());
if (inst->predicate) { if (inst->predicate) {
@ -639,9 +639,9 @@ emit_zip(const fs_builder &lbld_before, const fs_builder &lbld_after,
* SIMD16 16 bit values. * SIMD16 16 bit values.
*/ */
const fs_builder rbld = lbld_after.exec_all().group(1, 0); const fs_builder rbld = lbld_after.exec_all().group(1, 0);
fs_reg local_res_reg = component( brw_reg local_res_reg = component(
retype(offset(tmp, lbld_before, dst_size), BRW_TYPE_UW), 0); retype(offset(tmp, lbld_before, dst_size), BRW_TYPE_UW), 0);
fs_reg final_res_reg = brw_reg final_res_reg =
retype(byte_offset(inst->dst, retype(byte_offset(inst->dst,
inst->size_written - residency_size + inst->size_written - residency_size +
lbld_after.group() / 8), BRW_TYPE_UW); lbld_after.group() / 8), BRW_TYPE_UW);

1160
src/intel/compiler/brw_fs_nir.cpp
View file

File diff suppressed because it is too large Load diff

8
src/intel/compiler/brw_fs_opt_algebraic.cpp
View file

@ -9,7 +9,7 @@
using namespace brw; using namespace brw;
static uint64_t static uint64_t
src_as_uint(const fs_reg &src) src_as_uint(const brw_reg &src)
{ {
assert(src.file == IMM); assert(src.file == IMM);
@ -37,7 +37,7 @@ src_as_uint(const fs_reg &src)
} }
} }
static fs_reg static brw_reg
brw_imm_for_type(uint64_t value, enum brw_reg_type type) brw_imm_for_type(uint64_t value, enum brw_reg_type type)
{ {
switch (type) { switch (type) {
@ -394,7 +394,7 @@ brw_fs_opt_algebraic(fs_visitor &s)
*/ */
assert(!inst->saturate); assert(!inst->saturate);
fs_reg result; brw_reg result;
switch (brw_type_size_bytes(inst->src[0].type)) { switch (brw_type_size_bytes(inst->src[0].type)) {
case 2: case 2:
@ -470,7 +470,7 @@ brw_fs_opt_algebraic(fs_visitor &s)
*/ */
if (progress && inst->sources == 2 && inst->is_commutative()) { if (progress && inst->sources == 2 && inst->is_commutative()) {
if (inst->src[0].file == IMM) { if (inst->src[0].file == IMM) {
fs_reg tmp = inst->src[1]; brw_reg tmp = inst->src[1];
inst->src[1] = inst->src[0]; inst->src[1] = inst->src[0];
inst->src[0] = tmp; inst->src[0] = tmp;
} }

60
src/intel/compiler/brw_fs_reg_allocate.cpp
View file

@ -38,7 +38,7 @@ using namespace brw;
static void static void
assign_reg(const struct intel_device_info *devinfo, assign_reg(const struct intel_device_info *devinfo,
unsigned *reg_hw_locations, fs_reg *reg) unsigned *reg_hw_locations, brw_reg *reg)
{ {
if (reg->file == VGRF) { if (reg->file == VGRF) {
reg->nr = reg_unit(devinfo) * reg_hw_locations[reg->nr] + reg->offset / REG_SIZE; reg->nr = reg_unit(devinfo) * reg_hw_locations[reg->nr] + reg->offset / REG_SIZE;
@ -291,21 +291,21 @@ private:
void build_interference_graph(); void build_interference_graph();
void discard_interference_graph(); void discard_interference_graph();
fs_reg build_lane_offsets(const fs_builder &bld, brw_reg build_lane_offsets(const fs_builder &bld,
uint32_t spill_offset, int ip); uint32_t spill_offset, int ip);
fs_reg build_single_offset(const fs_builder &bld, brw_reg build_single_offset(const fs_builder &bld,
uint32_t spill_offset, int ip); uint32_t spill_offset, int ip);
fs_reg build_legacy_scratch_header(const fs_builder &bld, brw_reg build_legacy_scratch_header(const fs_builder &bld,
uint32_t spill_offset, int ip); uint32_t spill_offset, int ip);
void emit_unspill(const fs_builder &bld, struct shader_stats *stats, void emit_unspill(const fs_builder &bld, struct shader_stats *stats,
fs_reg dst, uint32_t spill_offset, unsigned count, int ip); brw_reg dst, uint32_t spill_offset, unsigned count, int ip);
void emit_spill(const fs_builder &bld, struct shader_stats *stats, void emit_spill(const fs_builder &bld, struct shader_stats *stats,
fs_reg src, uint32_t spill_offset, unsigned count, int ip); brw_reg src, uint32_t spill_offset, unsigned count, int ip);
void set_spill_costs(); void set_spill_costs();
int choose_spill_reg(); int choose_spill_reg();
fs_reg alloc_spill_reg(unsigned size, int ip); brw_reg alloc_spill_reg(unsigned size, int ip);
void spill_reg(unsigned spill_reg); void spill_reg(unsigned spill_reg);
void *mem_ctx; void *mem_ctx;
@ -571,16 +571,16 @@ fs_reg_alloc::discard_interference_graph()
have_spill_costs = false; have_spill_costs = false;
} }
fs_reg brw_reg
fs_reg_alloc::build_single_offset(const fs_builder &bld, uint32_t spill_offset, int ip) fs_reg_alloc::build_single_offset(const fs_builder &bld, uint32_t spill_offset, int ip)
{ {
fs_reg offset = retype(alloc_spill_reg(1, ip), BRW_TYPE_UD); brw_reg offset = retype(alloc_spill_reg(1, ip), BRW_TYPE_UD);
fs_inst *inst = bld.MOV(offset, brw_imm_ud(spill_offset)); fs_inst *inst = bld.MOV(offset, brw_imm_ud(spill_offset));
_mesa_set_add(spill_insts, inst); _mesa_set_add(spill_insts, inst);
return offset; return offset;
} }
fs_reg brw_reg
fs_reg_alloc::build_lane_offsets(const fs_builder &bld, uint32_t spill_offset, int ip) fs_reg_alloc::build_lane_offsets(const fs_builder &bld, uint32_t spill_offset, int ip)
{ {
/* LSC messages are limited to SIMD16 */ /* LSC messages are limited to SIMD16 */
@ -589,7 +589,7 @@ fs_reg_alloc::build_lane_offsets(const fs_builder &bld, uint32_t spill_offset, i
const fs_builder ubld = bld.exec_all(); const fs_builder ubld = bld.exec_all();
const unsigned reg_count = ubld.dispatch_width() / 8; const unsigned reg_count = ubld.dispatch_width() / 8;
fs_reg offset = retype(alloc_spill_reg(reg_count, ip), BRW_TYPE_UD); brw_reg offset = retype(alloc_spill_reg(reg_count, ip), BRW_TYPE_UD);
fs_inst *inst; fs_inst *inst;
/* Build an offset per lane in SIMD8 */ /* Build an offset per lane in SIMD8 */
@ -622,7 +622,7 @@ fs_reg_alloc::build_lane_offsets(const fs_builder &bld, uint32_t spill_offset, i
/** /**
* Generate a scratch header for pre-LSC platforms. * Generate a scratch header for pre-LSC platforms.
*/ */
fs_reg brw_reg
fs_reg_alloc::build_legacy_scratch_header(const fs_builder &bld, fs_reg_alloc::build_legacy_scratch_header(const fs_builder &bld,
uint32_t spill_offset, int ip) uint32_t spill_offset, int ip)
{ {
@ -630,7 +630,7 @@ fs_reg_alloc::build_legacy_scratch_header(const fs_builder &bld,
const fs_builder ubld1 = bld.exec_all().group(1, 0); const fs_builder ubld1 = bld.exec_all().group(1, 0);
/* Allocate a spill header and make it interfere with g0 */ /* Allocate a spill header and make it interfere with g0 */
fs_reg header = retype(alloc_spill_reg(1, ip), BRW_TYPE_UD); brw_reg header = retype(alloc_spill_reg(1, ip), BRW_TYPE_UD);
ra_add_node_interference(g, first_vgrf_node + header.nr, first_payload_node); ra_add_node_interference(g, first_vgrf_node + header.nr, first_payload_node);
fs_inst *inst = ubld8.emit(SHADER_OPCODE_SCRATCH_HEADER, header); fs_inst *inst = ubld8.emit(SHADER_OPCODE_SCRATCH_HEADER, header);
@ -647,7 +647,7 @@ fs_reg_alloc::build_legacy_scratch_header(const fs_builder &bld,
void void
fs_reg_alloc::emit_unspill(const fs_builder &bld, fs_reg_alloc::emit_unspill(const fs_builder &bld,
struct shader_stats *stats, struct shader_stats *stats,
fs_reg dst, brw_reg dst,
uint32_t spill_offset, unsigned count, int ip) uint32_t spill_offset, unsigned count, int ip)
{ {
const intel_device_info *devinfo = bld.shader->devinfo; const intel_device_info *devinfo = bld.shader->devinfo;
@ -664,7 +664,7 @@ fs_reg_alloc::emit_unspill(const fs_builder &bld,
*/ */
const bool use_transpose = bld.dispatch_width() > 16; const bool use_transpose = bld.dispatch_width() > 16;
const fs_builder ubld = use_transpose ? bld.exec_all().group(1, 0) : bld; const fs_builder ubld = use_transpose ? bld.exec_all().group(1, 0) : bld;
fs_reg offset; brw_reg offset;
if (use_transpose) { if (use_transpose) {
offset = build_single_offset(ubld, spill_offset, ip); offset = build_single_offset(ubld, spill_offset, ip);
} else { } else {
@ -675,11 +675,11 @@ fs_reg_alloc::emit_unspill(const fs_builder &bld,
* register. That way we don't need to burn an additional register * register. That way we don't need to burn an additional register
* for register allocation spill/fill. * for register allocation spill/fill.
*/ */
fs_reg srcs[] = { brw_reg srcs[] = {
brw_imm_ud(0), /* desc */ brw_imm_ud(0), /* desc */
brw_imm_ud(0), /* ex_desc */ brw_imm_ud(0), /* ex_desc */
offset, /* payload */ offset, /* payload */
fs_reg(), /* payload2 */ brw_reg(), /* payload2 */
}; };
unspill_inst = ubld.emit(SHADER_OPCODE_SEND, dst, unspill_inst = ubld.emit(SHADER_OPCODE_SEND, dst,
@ -702,12 +702,12 @@ fs_reg_alloc::emit_unspill(const fs_builder &bld,
unspill_inst->send_is_volatile = true; unspill_inst->send_is_volatile = true;
unspill_inst->send_ex_desc_scratch = true; unspill_inst->send_ex_desc_scratch = true;
} else { } else {
fs_reg header = build_legacy_scratch_header(bld, spill_offset, ip); brw_reg header = build_legacy_scratch_header(bld, spill_offset, ip);
const unsigned bti = GFX8_BTI_STATELESS_NON_COHERENT; const unsigned bti = GFX8_BTI_STATELESS_NON_COHERENT;
const fs_reg ex_desc = brw_imm_ud(0); const brw_reg ex_desc = brw_imm_ud(0);
fs_reg srcs[] = { brw_imm_ud(0), ex_desc, header }; brw_reg srcs[] = { brw_imm_ud(0), ex_desc, header };
unspill_inst = bld.emit(SHADER_OPCODE_SEND, dst, unspill_inst = bld.emit(SHADER_OPCODE_SEND, dst,
srcs, ARRAY_SIZE(srcs)); srcs, ARRAY_SIZE(srcs));
unspill_inst->mlen = 1; unspill_inst->mlen = 1;
@ -732,7 +732,7 @@ fs_reg_alloc::emit_unspill(const fs_builder &bld,
void void
fs_reg_alloc::emit_spill(const fs_builder &bld, fs_reg_alloc::emit_spill(const fs_builder &bld,
struct shader_stats *stats, struct shader_stats *stats,
fs_reg src, brw_reg src,
uint32_t spill_offset, unsigned count, int ip) uint32_t spill_offset, unsigned count, int ip)
{ {
const intel_device_info *devinfo = bld.shader->devinfo; const intel_device_info *devinfo = bld.shader->devinfo;
@ -744,13 +744,13 @@ fs_reg_alloc::emit_spill(const fs_builder &bld,
fs_inst *spill_inst; fs_inst *spill_inst;
if (devinfo->verx10 >= 125) { if (devinfo->verx10 >= 125) {
fs_reg offset = build_lane_offsets(bld, spill_offset, ip); brw_reg offset = build_lane_offsets(bld, spill_offset, ip);
/* We leave the extended descriptor empty and flag the instruction /* We leave the extended descriptor empty and flag the instruction
* relocate the extended descriptor. That way the surface offset is * relocate the extended descriptor. That way the surface offset is
* directly put into the instruction and we don't need to use a * directly put into the instruction and we don't need to use a
* register to hold it. * register to hold it.
*/ */
fs_reg srcs[] = { brw_reg srcs[] = {
brw_imm_ud(0), /* desc */ brw_imm_ud(0), /* desc */
brw_imm_ud(0), /* ex_desc */ brw_imm_ud(0), /* ex_desc */
offset, /* payload */ offset, /* payload */
@ -775,12 +775,12 @@ fs_reg_alloc::emit_spill(const fs_builder &bld,
spill_inst->send_is_volatile = false; spill_inst->send_is_volatile = false;
spill_inst->send_ex_desc_scratch = true; spill_inst->send_ex_desc_scratch = true;
} else { } else {
fs_reg header = build_legacy_scratch_header(bld, spill_offset, ip); brw_reg header = build_legacy_scratch_header(bld, spill_offset, ip);
const unsigned bti = GFX8_BTI_STATELESS_NON_COHERENT; const unsigned bti = GFX8_BTI_STATELESS_NON_COHERENT;
const fs_reg ex_desc = brw_imm_ud(0); const brw_reg ex_desc = brw_imm_ud(0);
fs_reg srcs[] = { brw_imm_ud(0), ex_desc, header, src }; brw_reg srcs[] = { brw_imm_ud(0), ex_desc, header, src };
spill_inst = bld.emit(SHADER_OPCODE_SEND, bld.null_reg_f(), spill_inst = bld.emit(SHADER_OPCODE_SEND, bld.null_reg_f(),
srcs, ARRAY_SIZE(srcs)); srcs, ARRAY_SIZE(srcs));
spill_inst->mlen = 1; spill_inst->mlen = 1;
@ -903,7 +903,7 @@ fs_reg_alloc::choose_spill_reg()
return node - first_vgrf_node; return node - first_vgrf_node;
} }
fs_reg brw_reg
fs_reg_alloc::alloc_spill_reg(unsigned size, int ip) fs_reg_alloc::alloc_spill_reg(unsigned size, int ip)
{ {
int vgrf = fs->alloc.allocate(ALIGN(size, reg_unit(devinfo))); int vgrf = fs->alloc.allocate(ALIGN(size, reg_unit(devinfo)));
@ -970,7 +970,7 @@ fs_reg_alloc::spill_reg(unsigned spill_reg)
int count = regs_read(inst, i); int count = regs_read(inst, i);
int subset_spill_offset = spill_offset + int subset_spill_offset = spill_offset +
ROUND_DOWN_TO(inst->src[i].offset, REG_SIZE); ROUND_DOWN_TO(inst->src[i].offset, REG_SIZE);
fs_reg unspill_dst = alloc_spill_reg(count, ip); brw_reg unspill_dst = alloc_spill_reg(count, ip);
inst->src[i].nr = unspill_dst.nr; inst->src[i].nr = unspill_dst.nr;
inst->src[i].offset %= REG_SIZE; inst->src[i].offset %= REG_SIZE;
@ -999,7 +999,7 @@ fs_reg_alloc::spill_reg(unsigned spill_reg)
inst->opcode != SHADER_OPCODE_UNDEF) { inst->opcode != SHADER_OPCODE_UNDEF) {
int subset_spill_offset = spill_offset + int subset_spill_offset = spill_offset +
ROUND_DOWN_TO(inst->dst.offset, REG_SIZE); ROUND_DOWN_TO(inst->dst.offset, REG_SIZE);
fs_reg spill_src = alloc_spill_reg(regs_written(inst), ip); brw_reg spill_src = alloc_spill_reg(regs_written(inst), ip);
inst->dst.nr = spill_src.nr; inst->dst.nr = spill_src.nr;
inst->dst.offset %= REG_SIZE; inst->dst.offset %= REG_SIZE;

2
src/intel/compiler/brw_fs_register_coalesce.cpp
View file

@ -50,7 +50,7 @@ static bool
is_nop_mov(const fs_inst *inst) is_nop_mov(const fs_inst *inst)
{ {
if (inst->opcode == SHADER_OPCODE_LOAD_PAYLOAD) { if (inst->opcode == SHADER_OPCODE_LOAD_PAYLOAD) {
fs_reg dst = inst->dst; brw_reg dst = inst->dst;
for (int i = 0; i < inst->sources; i++) { for (int i = 0; i < inst->sources; i++) {
if (!dst.equals(inst->src[i])) { if (!dst.equals(inst->src[i])) {
return false; return false;

8
src/intel/compiler/brw_fs_scoreboard.cpp
View file

@ -659,7 +659,7 @@ namespace {
* Look up the most current data dependency for register \p r. * Look up the most current data dependency for register \p r.
*/ */
dependency dependency
get(const fs_reg &r) const get(const brw_reg &r) const
{ {
if (const dependency *p = const_cast<scoreboard *>(this)->dep(r)) if (const dependency *p = const_cast<scoreboard *>(this)->dep(r))
return *p; return *p;
@ -671,7 +671,7 @@ namespace {
* Specify the most current data dependency for register \p r. * Specify the most current data dependency for register \p r.
*/ */
void void
set(const fs_reg &r, const dependency &d) set(const brw_reg &r, const dependency &d)
{ {
if (dependency *p = dep(r)) if (dependency *p = dep(r))
*p = d; *p = d;
@ -761,7 +761,7 @@ namespace {
dependency accum_dep; dependency accum_dep;
dependency * dependency *
dep(const fs_reg &r) dep(const brw_reg &r)
{ {
const unsigned reg = (r.file == VGRF ? r.nr + r.offset / REG_SIZE : const unsigned reg = (r.file == VGRF ? r.nr + r.offset / REG_SIZE :
reg_offset(r) / REG_SIZE); reg_offset(r) / REG_SIZE);
@ -1038,7 +1038,7 @@ namespace {
dependency::done; dependency::done;
for (unsigned j = 0; j < regs_read(inst, i); j++) { for (unsigned j = 0; j < regs_read(inst, i); j++) {
const fs_reg r = byte_offset(inst->src[i], REG_SIZE * j); const brw_reg r = byte_offset(inst->src[i], REG_SIZE * j);
sb.set(r, shadow(sb.get(r), rd_dep)); sb.set(r, shadow(sb.get(r), rd_dep));
} }
} }

4
src/intel/compiler/brw_fs_sel_peephole.cpp
View file

@ -200,14 +200,14 @@ brw_fs_opt_peephole_sel(fs_visitor &s)
* in the "then" clause uses a constant, we need to put it in a * in the "then" clause uses a constant, we need to put it in a
* temporary. * temporary.
*/ */
fs_reg src0(then_mov[i]->src[0]); brw_reg src0(then_mov[i]->src[0]);
if (src0.file == IMM) { if (src0.file == IMM) {
src0 = ibld.vgrf(then_mov[i]->src[0].type); src0 = ibld.vgrf(then_mov[i]->src[0].type);
ibld.MOV(src0, then_mov[i]->src[0]); ibld.MOV(src0, then_mov[i]->src[0]);
} }
/* 64-bit immediates can't be placed in src1. */ /* 64-bit immediates can't be placed in src1. */
fs_reg src1(else_mov[i]->src[0]); brw_reg src1(else_mov[i]->src[0]);
if (src1.file == IMM && brw_type_size_bytes(src1.type) == 8) { if (src1.file == IMM && brw_type_size_bytes(src1.type) == 8) {
src1 = ibld.vgrf(else_mov[i]->src[0].type); src1 = ibld.vgrf(else_mov[i]->src[0].type);
ibld.MOV(src1, else_mov[i]->src[0]); ibld.MOV(src1, else_mov[i]->src[0]);

6
src/intel/compiler/brw_fs_thread_payload.cpp
View file

@ -386,7 +386,7 @@ cs_thread_payload::cs_thread_payload(const fs_visitor &v)
void void
cs_thread_payload::load_subgroup_id(const fs_builder &bld, cs_thread_payload::load_subgroup_id(const fs_builder &bld,
fs_reg &dest) const brw_reg &dest) const
{ {
auto devinfo = bld.shader->devinfo; auto devinfo = bld.shader->devinfo;
dest = retype(dest, BRW_TYPE_UD); dest = retype(dest, BRW_TYPE_UD);
@ -483,9 +483,9 @@ bs_thread_payload::bs_thread_payload(const fs_visitor &v)
} }
void void
bs_thread_payload::load_shader_type(const fs_builder &bld, fs_reg &dest) const bs_thread_payload::load_shader_type(const fs_builder &bld, brw_reg &dest) const
{ {
fs_reg ud_dest = retype(dest, BRW_TYPE_UD); brw_reg ud_dest = retype(dest, BRW_TYPE_UD);
bld.MOV(ud_dest, retype(brw_vec1_grf(0, 3), ud_dest.type)); bld.MOV(ud_dest, retype(brw_vec1_grf(0, 3), ud_dest.type));
bld.AND(ud_dest, ud_dest, brw_imm_ud(0xf)); bld.AND(ud_dest, ud_dest, brw_imm_ud(0xf));
} }

140
src/intel/compiler/brw_fs_visitor.cpp
View file

@ -45,7 +45,7 @@ using namespace brw;
* data. It will get adjusted to be a real location before * data. It will get adjusted to be a real location before
* generate_code() time. * generate_code() time.
*/ */
fs_reg brw_reg
fs_visitor::interp_reg(const fs_builder &bld, unsigned location, fs_visitor::interp_reg(const fs_builder &bld, unsigned location,
unsigned channel, unsigned comp) unsigned channel, unsigned comp)
{ {
@ -71,7 +71,7 @@ fs_visitor::interp_reg(const fs_builder &bld, unsigned location,
* assign_urb_setup()), so we need to use offset() instead of * assign_urb_setup()), so we need to use offset() instead of
* component() to select the specified parameter. * component() to select the specified parameter.
*/ */
const fs_reg tmp = bld.vgrf(BRW_TYPE_UD); const brw_reg tmp = bld.vgrf(BRW_TYPE_UD);
bld.MOV(tmp, offset(brw_attr_reg(regnr, BRW_TYPE_UD), bld.MOV(tmp, offset(brw_attr_reg(regnr, BRW_TYPE_UD),
dispatch_width, comp)); dispatch_width, comp));
return retype(tmp, BRW_TYPE_F); return retype(tmp, BRW_TYPE_F);
@ -84,7 +84,7 @@ fs_visitor::interp_reg(const fs_builder &bld, unsigned location,
* data. It will get adjusted to be a real location before * data. It will get adjusted to be a real location before
* generate_code() time. * generate_code() time.
*/ */
fs_reg brw_reg
fs_visitor::per_primitive_reg(const fs_builder &bld, int location, unsigned comp) fs_visitor::per_primitive_reg(const fs_builder &bld, int location, unsigned comp)
{ {
assert(stage == MESA_SHADER_FRAGMENT); assert(stage == MESA_SHADER_FRAGMENT);
@ -106,7 +106,7 @@ fs_visitor::per_primitive_reg(const fs_builder &bld, int location, unsigned comp
* assign_urb_setup()), so we need to use offset() instead of * assign_urb_setup()), so we need to use offset() instead of
* component() to select the specified parameter. * component() to select the specified parameter.
*/ */
const fs_reg tmp = bld.vgrf(BRW_TYPE_UD); const brw_reg tmp = bld.vgrf(BRW_TYPE_UD);
bld.MOV(tmp, offset(brw_attr_reg(regnr, BRW_TYPE_UD), bld.MOV(tmp, offset(brw_attr_reg(regnr, BRW_TYPE_UD),
dispatch_width, comp % 4)); dispatch_width, comp % 4));
return retype(tmp, BRW_TYPE_F); return retype(tmp, BRW_TYPE_F);
@ -128,9 +128,9 @@ fs_visitor::emit_interpolation_setup()
const struct brw_wm_prog_key *wm_key = (brw_wm_prog_key*) this->key; const struct brw_wm_prog_key *wm_key = (brw_wm_prog_key*) this->key;
struct brw_wm_prog_data *wm_prog_data = brw_wm_prog_data(prog_data); struct brw_wm_prog_data *wm_prog_data = brw_wm_prog_data(prog_data);
fs_reg int_sample_offset_x, int_sample_offset_y; /* Used on Gen12HP+ */ brw_reg int_sample_offset_x, int_sample_offset_y; /* Used on Gen12HP+ */
fs_reg int_sample_offset_xy; /* Used on Gen8+ */ brw_reg int_sample_offset_xy; /* Used on Gen8+ */
fs_reg half_int_sample_offset_x, half_int_sample_offset_y; brw_reg half_int_sample_offset_x, half_int_sample_offset_y;
if (wm_prog_data->coarse_pixel_dispatch != BRW_ALWAYS) { if (wm_prog_data->coarse_pixel_dispatch != BRW_ALWAYS) {
/* The thread payload only delivers subspan locations (ss0, ss1, /* The thread payload only delivers subspan locations (ss0, ss1,
* ss2, ...). Since subspans covers 2x2 pixels blocks, we need to * ss2, ...). Since subspans covers 2x2 pixels blocks, we need to
@ -161,9 +161,9 @@ fs_visitor::emit_interpolation_setup()
* coordinates out of 2 subspans coordinates in a single ADD instruction * coordinates out of 2 subspans coordinates in a single ADD instruction
* (twice the operation above). * (twice the operation above).
*/ */
int_sample_offset_xy = fs_reg(brw_imm_v(0x11001010)); int_sample_offset_xy = brw_reg(brw_imm_v(0x11001010));
half_int_sample_offset_x = fs_reg(brw_imm_uw(0)); half_int_sample_offset_x = brw_reg(brw_imm_uw(0));
half_int_sample_offset_y = fs_reg(brw_imm_uw(0)); half_int_sample_offset_y = brw_reg(brw_imm_uw(0));
/* On Gfx12.5, because of regioning restrictions, the interpolation code /* On Gfx12.5, because of regioning restrictions, the interpolation code
* is slightly different and works off X & Y only inputs. The ordering * is slightly different and works off X & Y only inputs. The ordering
* of the half bytes here is a bit odd, with each subspan replicated * of the half bytes here is a bit odd, with each subspan replicated
@ -173,13 +173,13 @@ fs_visitor::emit_interpolation_setup()
* X offset: 0 0 1 0 0 0 1 0 * X offset: 0 0 1 0 0 0 1 0
* Y offset: 0 0 0 0 1 0 1 0 * Y offset: 0 0 0 0 1 0 1 0
*/ */
int_sample_offset_x = fs_reg(brw_imm_v(0x01000100)); int_sample_offset_x = brw_reg(brw_imm_v(0x01000100));
int_sample_offset_y = fs_reg(brw_imm_v(0x01010000)); int_sample_offset_y = brw_reg(brw_imm_v(0x01010000));
} }
fs_reg int_coarse_offset_x, int_coarse_offset_y; /* Used on Gen12HP+ */ brw_reg int_coarse_offset_x, int_coarse_offset_y; /* Used on Gen12HP+ */
fs_reg int_coarse_offset_xy; /* Used on Gen8+ */ brw_reg int_coarse_offset_xy; /* Used on Gen8+ */
fs_reg half_int_coarse_offset_x, half_int_coarse_offset_y; brw_reg half_int_coarse_offset_x, half_int_coarse_offset_y;
if (wm_prog_data->coarse_pixel_dispatch != BRW_NEVER) { if (wm_prog_data->coarse_pixel_dispatch != BRW_NEVER) {
/* In coarse pixel dispatch we have to do the same ADD instruction that /* In coarse pixel dispatch we have to do the same ADD instruction that
* we do in normal per pixel dispatch, except this time we're not adding * we do in normal per pixel dispatch, except this time we're not adding
@ -226,9 +226,9 @@ fs_visitor::emit_interpolation_setup()
bld.SHR(half_int_coarse_offset_y, suboffset(r1_0, 1), brw_imm_ud(1)); bld.SHR(half_int_coarse_offset_y, suboffset(r1_0, 1), brw_imm_ud(1));
} }
fs_reg int_pixel_offset_x, int_pixel_offset_y; /* Used on Gen12HP+ */ brw_reg int_pixel_offset_x, int_pixel_offset_y; /* Used on Gen12HP+ */
fs_reg int_pixel_offset_xy; /* Used on Gen8+ */ brw_reg int_pixel_offset_xy; /* Used on Gen8+ */
fs_reg half_int_pixel_offset_x, half_int_pixel_offset_y; brw_reg half_int_pixel_offset_x, half_int_pixel_offset_y;
switch (wm_prog_data->coarse_pixel_dispatch) { switch (wm_prog_data->coarse_pixel_dispatch) {
case BRW_NEVER: case BRW_NEVER:
int_pixel_offset_x = int_sample_offset_x; int_pixel_offset_x = int_sample_offset_x;
@ -301,14 +301,14 @@ fs_visitor::emit_interpolation_setup()
if (devinfo->verx10 >= 125) { if (devinfo->verx10 >= 125) {
const fs_builder dbld = const fs_builder dbld =
abld.exec_all().group(hbld.dispatch_width() * 2, 0); abld.exec_all().group(hbld.dispatch_width() * 2, 0);
const fs_reg int_pixel_x = dbld.vgrf(BRW_TYPE_UW); const brw_reg int_pixel_x = dbld.vgrf(BRW_TYPE_UW);
const fs_reg int_pixel_y = dbld.vgrf(BRW_TYPE_UW); const brw_reg int_pixel_y = dbld.vgrf(BRW_TYPE_UW);
dbld.ADD(int_pixel_x, dbld.ADD(int_pixel_x,
fs_reg(stride(suboffset(gi_uw, 4), 2, 8, 0)), brw_reg(stride(suboffset(gi_uw, 4), 2, 8, 0)),
int_pixel_offset_x); int_pixel_offset_x);
dbld.ADD(int_pixel_y, dbld.ADD(int_pixel_y,
fs_reg(stride(suboffset(gi_uw, 5), 2, 8, 0)), brw_reg(stride(suboffset(gi_uw, 5), 2, 8, 0)),
int_pixel_offset_y); int_pixel_offset_y);
if (wm_prog_data->coarse_pixel_dispatch != BRW_NEVER) { if (wm_prog_data->coarse_pixel_dispatch != BRW_NEVER) {
@ -338,10 +338,10 @@ fs_visitor::emit_interpolation_setup()
*/ */
const fs_builder dbld = const fs_builder dbld =
abld.exec_all().group(hbld.dispatch_width() * 2, 0); abld.exec_all().group(hbld.dispatch_width() * 2, 0);
fs_reg int_pixel_xy = dbld.vgrf(BRW_TYPE_UW); brw_reg int_pixel_xy = dbld.vgrf(BRW_TYPE_UW);
dbld.ADD(int_pixel_xy, dbld.ADD(int_pixel_xy,
fs_reg(stride(suboffset(gi_uw, 4), 1, 4, 0)), brw_reg(stride(suboffset(gi_uw, 4), 1, 4, 0)),
int_pixel_offset_xy); int_pixel_offset_xy);
hbld.emit(FS_OPCODE_PIXEL_X, offset(pixel_x, hbld, i), int_pixel_xy, hbld.emit(FS_OPCODE_PIXEL_X, offset(pixel_x, hbld, i), int_pixel_xy,
@ -352,7 +352,7 @@ fs_visitor::emit_interpolation_setup()
} }
abld = bld.annotate("compute pos.z"); abld = bld.annotate("compute pos.z");
fs_reg coarse_z; brw_reg coarse_z;
if (wm_prog_data->coarse_pixel_dispatch != BRW_NEVER && if (wm_prog_data->coarse_pixel_dispatch != BRW_NEVER &&
wm_prog_data->uses_depth_w_coefficients) { wm_prog_data->uses_depth_w_coefficients) {
/* In coarse pixel mode, the HW doesn't interpolate Z coordinate /* In coarse pixel mode, the HW doesn't interpolate Z coordinate
@ -360,30 +360,30 @@ fs_visitor::emit_interpolation_setup()
* pixels locations, here we recompute the Z value with 2 coefficients * pixels locations, here we recompute the Z value with 2 coefficients
* in X & Y axis. * in X & Y axis.
*/ */
fs_reg coef_payload = brw_vec8_grf(fs_payload().depth_w_coef_reg, 0); brw_reg coef_payload = brw_vec8_grf(fs_payload().depth_w_coef_reg, 0);
const fs_reg x_start = brw_vec1_grf(coef_payload.nr, 2); const brw_reg x_start = brw_vec1_grf(coef_payload.nr, 2);
const fs_reg y_start = brw_vec1_grf(coef_payload.nr, 6); const brw_reg y_start = brw_vec1_grf(coef_payload.nr, 6);
const fs_reg z_cx = brw_vec1_grf(coef_payload.nr, 1); const brw_reg z_cx = brw_vec1_grf(coef_payload.nr, 1);
const fs_reg z_cy = brw_vec1_grf(coef_payload.nr, 0); const brw_reg z_cy = brw_vec1_grf(coef_payload.nr, 0);
const fs_reg z_c0 = brw_vec1_grf(coef_payload.nr, 3); const brw_reg z_c0 = brw_vec1_grf(coef_payload.nr, 3);
const fs_reg float_pixel_x = abld.vgrf(BRW_TYPE_F); const brw_reg float_pixel_x = abld.vgrf(BRW_TYPE_F);
const fs_reg float_pixel_y = abld.vgrf(BRW_TYPE_F); const brw_reg float_pixel_y = abld.vgrf(BRW_TYPE_F);
abld.ADD(float_pixel_x, this->pixel_x, negate(x_start)); abld.ADD(float_pixel_x, this->pixel_x, negate(x_start));
abld.ADD(float_pixel_y, this->pixel_y, negate(y_start)); abld.ADD(float_pixel_y, this->pixel_y, negate(y_start));
/* r1.0 - 0:7 ActualCoarsePixelShadingSize.X */ /* r1.0 - 0:7 ActualCoarsePixelShadingSize.X */
const fs_reg u8_cps_width = fs_reg(retype(brw_vec1_grf(1, 0), BRW_TYPE_UB)); const brw_reg u8_cps_width = brw_reg(retype(brw_vec1_grf(1, 0), BRW_TYPE_UB));
/* r1.0 - 15:8 ActualCoarsePixelShadingSize.Y */ /* r1.0 - 15:8 ActualCoarsePixelShadingSize.Y */
const fs_reg u8_cps_height = byte_offset(u8_cps_width, 1); const brw_reg u8_cps_height = byte_offset(u8_cps_width, 1);
const fs_reg u32_cps_width = abld.vgrf(BRW_TYPE_UD); const brw_reg u32_cps_width = abld.vgrf(BRW_TYPE_UD);
const fs_reg u32_cps_height = abld.vgrf(BRW_TYPE_UD); const brw_reg u32_cps_height = abld.vgrf(BRW_TYPE_UD);
abld.MOV(u32_cps_width, u8_cps_width); abld.MOV(u32_cps_width, u8_cps_width);
abld.MOV(u32_cps_height, u8_cps_height); abld.MOV(u32_cps_height, u8_cps_height);
const fs_reg f_cps_width = abld.vgrf(BRW_TYPE_F); const brw_reg f_cps_width = abld.vgrf(BRW_TYPE_F);
const fs_reg f_cps_height = abld.vgrf(BRW_TYPE_F); const brw_reg f_cps_height = abld.vgrf(BRW_TYPE_F);
abld.MOV(f_cps_width, u32_cps_width); abld.MOV(f_cps_width, u32_cps_width);
abld.MOV(f_cps_height, u32_cps_height); abld.MOV(f_cps_height, u32_cps_height);
@ -401,7 +401,7 @@ fs_visitor::emit_interpolation_setup()
if (wm_prog_data->uses_depth_w_coefficients || if (wm_prog_data->uses_depth_w_coefficients ||
wm_prog_data->uses_src_depth) { wm_prog_data->uses_src_depth) {
fs_reg sample_z = this->pixel_z; brw_reg sample_z = this->pixel_z;
switch (wm_prog_data->coarse_pixel_dispatch) { switch (wm_prog_data->coarse_pixel_dispatch) {
case BRW_NEVER: case BRW_NEVER:
@ -505,8 +505,8 @@ fs_visitor::emit_interpolation_setup()
if (!(centroid_modes & (1 << i))) if (!(centroid_modes & (1 << i)))
continue; continue;
const fs_reg centroid_delta_xy = delta_xy[i]; const brw_reg centroid_delta_xy = delta_xy[i];
const fs_reg &pixel_delta_xy = delta_xy[i - 1]; const brw_reg &pixel_delta_xy = delta_xy[i - 1];
delta_xy[i] = bld.vgrf(BRW_TYPE_F, 2); delta_xy[i] = bld.vgrf(BRW_TYPE_F, 2);
@ -525,15 +525,15 @@ fs_visitor::emit_interpolation_setup()
fs_inst * fs_inst *
fs_visitor::emit_single_fb_write(const fs_builder &bld, fs_visitor::emit_single_fb_write(const fs_builder &bld,
fs_reg color0, fs_reg color1, brw_reg color0, brw_reg color1,
fs_reg src0_alpha, unsigned components) brw_reg src0_alpha, unsigned components)
{ {
assert(stage == MESA_SHADER_FRAGMENT); assert(stage == MESA_SHADER_FRAGMENT);
struct brw_wm_prog_data *prog_data = brw_wm_prog_data(this->prog_data); struct brw_wm_prog_data *prog_data = brw_wm_prog_data(this->prog_data);
/* Hand over gl_FragDepth or the payload depth. */ /* Hand over gl_FragDepth or the payload depth. */
const fs_reg dst_depth = fetch_payload_reg(bld, fs_payload().dest_depth_reg); const brw_reg dst_depth = fetch_payload_reg(bld, fs_payload().dest_depth_reg);
fs_reg src_depth, src_stencil; brw_reg src_depth, src_stencil;
if (nir->info.outputs_written & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) if (nir->info.outputs_written & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
src_depth = frag_depth; src_depth = frag_depth;
@ -541,13 +541,13 @@ fs_visitor::emit_single_fb_write(const fs_builder &bld,
if (nir->info.outputs_written & BITFIELD64_BIT(FRAG_RESULT_STENCIL)) if (nir->info.outputs_written & BITFIELD64_BIT(FRAG_RESULT_STENCIL))
src_stencil = frag_stencil; src_stencil = frag_stencil;
const fs_reg sources[] = { const brw_reg sources[] = {
color0, color1, src0_alpha, src_depth, dst_depth, src_stencil, color0, color1, src0_alpha, src_depth, dst_depth, src_stencil,
(prog_data->uses_omask ? sample_mask : fs_reg()), (prog_data->uses_omask ? sample_mask : brw_reg()),
brw_imm_ud(components) brw_imm_ud(components)
}; };
assert(ARRAY_SIZE(sources) - 1 == FB_WRITE_LOGICAL_SRC_COMPONENTS); assert(ARRAY_SIZE(sources) - 1 == FB_WRITE_LOGICAL_SRC_COMPONENTS);
fs_inst *write = bld.emit(FS_OPCODE_FB_WRITE_LOGICAL, fs_reg(), fs_inst *write = bld.emit(FS_OPCODE_FB_WRITE_LOGICAL, brw_reg(),
sources, ARRAY_SIZE(sources)); sources, ARRAY_SIZE(sources));
if (prog_data->uses_kill) { if (prog_data->uses_kill) {
@ -572,7 +572,7 @@ fs_visitor::do_emit_fb_writes(int nr_color_regions, bool replicate_alpha)
const fs_builder abld = bld.annotate( const fs_builder abld = bld.annotate(
ralloc_asprintf(this->mem_ctx, "FB write target %d", target)); ralloc_asprintf(this->mem_ctx, "FB write target %d", target));
fs_reg src0_alpha; brw_reg src0_alpha;
if (replicate_alpha && target != 0) if (replicate_alpha && target != 0)
src0_alpha = offset(outputs[0], bld, 3); src0_alpha = offset(outputs[0], bld, 3);
@ -589,9 +589,9 @@ fs_visitor::do_emit_fb_writes(int nr_color_regions, bool replicate_alpha)
/* FINISHME: Factor out this frequently recurring pattern into a /* FINISHME: Factor out this frequently recurring pattern into a
* helper function. * helper function.
*/ */
const fs_reg srcs[] = { reg_undef, reg_undef, const brw_reg srcs[] = { reg_undef, reg_undef,
reg_undef, offset(this->outputs[0], bld, 3) }; reg_undef, offset(this->outputs[0], bld, 3) };
const fs_reg tmp = bld.vgrf(BRW_TYPE_UD, 4); const brw_reg tmp = bld.vgrf(BRW_TYPE_UD, 4);
bld.LOAD_PAYLOAD(tmp, srcs, 4, 0); bld.LOAD_PAYLOAD(tmp, srcs, 4, 0);
inst = emit_single_fb_write(bld, tmp, reg_undef, reg_undef, 4); inst = emit_single_fb_write(bld, tmp, reg_undef, reg_undef, 4);
@ -665,7 +665,7 @@ fs_visitor::emit_fb_writes()
} }
void void
fs_visitor::emit_urb_writes(const fs_reg &gs_vertex_count) fs_visitor::emit_urb_writes(const brw_reg &gs_vertex_count)
{ {
int slot, urb_offset, length; int slot, urb_offset, length;
int starting_urb_offset = 0; int starting_urb_offset = 0;
@ -675,8 +675,8 @@ fs_visitor::emit_urb_writes(const fs_reg &gs_vertex_count)
VARYING_BIT_LAYER | VARYING_BIT_VIEWPORT | VARYING_BIT_PSIZ | VARYING_BIT_PRIMITIVE_SHADING_RATE; VARYING_BIT_LAYER | VARYING_BIT_VIEWPORT | VARYING_BIT_PSIZ | VARYING_BIT_PRIMITIVE_SHADING_RATE;
const struct intel_vue_map *vue_map = &vue_prog_data->vue_map; const struct intel_vue_map *vue_map = &vue_prog_data->vue_map;
bool flush; bool flush;
fs_reg sources[8]; brw_reg sources[8];
fs_reg urb_handle; brw_reg urb_handle;
switch (stage) { switch (stage) {
case MESA_SHADER_VERTEX: case MESA_SHADER_VERTEX:
@ -694,7 +694,7 @@ fs_visitor::emit_urb_writes(const fs_reg &gs_vertex_count)
const fs_builder bld = fs_builder(this).at_end(); const fs_builder bld = fs_builder(this).at_end();
fs_reg per_slot_offsets; brw_reg per_slot_offsets;
if (stage == MESA_SHADER_GEOMETRY) { if (stage == MESA_SHADER_GEOMETRY) {
const struct brw_gs_prog_data *gs_prog_data = const struct brw_gs_prog_data *gs_prog_data =
@ -760,7 +760,7 @@ fs_visitor::emit_urb_writes(const fs_reg &gs_vertex_count)
break; break;
} }
fs_reg zero = brw_vgrf(alloc.allocate(dispatch_width / 8), brw_reg zero = brw_vgrf(alloc.allocate(dispatch_width / 8),
BRW_TYPE_UD); BRW_TYPE_UD);
bld.MOV(zero, brw_imm_ud(0u)); bld.MOV(zero, brw_imm_ud(0u));
@ -769,7 +769,7 @@ fs_visitor::emit_urb_writes(const fs_reg &gs_vertex_count)
sources[length++] = this->outputs[VARYING_SLOT_PRIMITIVE_SHADING_RATE]; sources[length++] = this->outputs[VARYING_SLOT_PRIMITIVE_SHADING_RATE];
} else if (devinfo->has_coarse_pixel_primitive_and_cb) { } else if (devinfo->has_coarse_pixel_primitive_and_cb) {
uint32_t one_fp16 = 0x3C00; uint32_t one_fp16 = 0x3C00;
fs_reg one_by_one_fp16 = brw_vgrf(alloc.allocate(dispatch_width / 8), brw_reg one_by_one_fp16 = brw_vgrf(alloc.allocate(dispatch_width / 8),
BRW_TYPE_UD); BRW_TYPE_UD);
bld.MOV(one_by_one_fp16, brw_imm_ud((one_fp16 << 16) | one_fp16)); bld.MOV(one_by_one_fp16, brw_imm_ud((one_fp16 << 16) | one_fp16));
sources[length++] = one_by_one_fp16; sources[length++] = one_by_one_fp16;
@ -839,7 +839,7 @@ fs_visitor::emit_urb_writes(const fs_reg &gs_vertex_count)
if (length == 8 || (length > 0 && slot == last_slot)) if (length == 8 || (length > 0 && slot == last_slot))
flush = true; flush = true;
if (flush) { if (flush) {
fs_reg srcs[URB_LOGICAL_NUM_SRCS]; brw_reg srcs[URB_LOGICAL_NUM_SRCS];
srcs[URB_LOGICAL_SRC_HANDLE] = urb_handle; srcs[URB_LOGICAL_SRC_HANDLE] = urb_handle;
srcs[URB_LOGICAL_SRC_PER_SLOT_OFFSETS] = per_slot_offsets; srcs[URB_LOGICAL_SRC_PER_SLOT_OFFSETS] = per_slot_offsets;
@ -883,14 +883,14 @@ fs_visitor::emit_urb_writes(const fs_reg &gs_vertex_count)
if (stage == MESA_SHADER_GEOMETRY) if (stage == MESA_SHADER_GEOMETRY)
return; return;
fs_reg uniform_urb_handle = brw_vgrf(alloc.allocate(dispatch_width / 8), brw_reg uniform_urb_handle = brw_vgrf(alloc.allocate(dispatch_width / 8),
BRW_TYPE_UD); BRW_TYPE_UD);
fs_reg payload = brw_vgrf(alloc.allocate(dispatch_width / 8), brw_reg payload = brw_vgrf(alloc.allocate(dispatch_width / 8),
BRW_TYPE_UD); BRW_TYPE_UD);
bld.exec_all().MOV(uniform_urb_handle, urb_handle); bld.exec_all().MOV(uniform_urb_handle, urb_handle);
fs_reg srcs[URB_LOGICAL_NUM_SRCS]; brw_reg srcs[URB_LOGICAL_NUM_SRCS];
srcs[URB_LOGICAL_SRC_HANDLE] = uniform_urb_handle; srcs[URB_LOGICAL_SRC_HANDLE] = uniform_urb_handle;
srcs[URB_LOGICAL_SRC_DATA] = payload; srcs[URB_LOGICAL_SRC_DATA] = payload;
srcs[URB_LOGICAL_SRC_COMPONENTS] = brw_imm_ud(1); srcs[URB_LOGICAL_SRC_COMPONENTS] = brw_imm_ud(1);
@ -910,9 +910,9 @@ fs_visitor::emit_urb_writes(const fs_reg &gs_vertex_count)
*/ */
if (intel_needs_workaround(devinfo, 1805992985) && stage == MESA_SHADER_TESS_EVAL) { if (intel_needs_workaround(devinfo, 1805992985) && stage == MESA_SHADER_TESS_EVAL) {
assert(dispatch_width == 8); assert(dispatch_width == 8);
fs_reg uniform_urb_handle = brw_vgrf(alloc.allocate(1), BRW_TYPE_UD); brw_reg uniform_urb_handle = brw_vgrf(alloc.allocate(1), BRW_TYPE_UD);
fs_reg uniform_mask = brw_vgrf(alloc.allocate(1), BRW_TYPE_UD); brw_reg uniform_mask = brw_vgrf(alloc.allocate(1), BRW_TYPE_UD);
fs_reg payload = brw_vgrf(alloc.allocate(4), BRW_TYPE_UD); brw_reg payload = brw_vgrf(alloc.allocate(4), BRW_TYPE_UD);
/* Workaround requires all 8 channels (lanes) to be valid. This is /* Workaround requires all 8 channels (lanes) to be valid. This is
* understood to mean they all need to be alive. First trick is to find * understood to mean they all need to be alive. First trick is to find
@ -941,7 +941,7 @@ fs_visitor::emit_urb_writes(const fs_reg &gs_vertex_count)
bld.exec_all().MOV(offset(payload, bld, 2), brw_imm_ud(0u)); bld.exec_all().MOV(offset(payload, bld, 2), brw_imm_ud(0u));
bld.exec_all().MOV(offset(payload, bld, 3), brw_imm_ud(0u)); bld.exec_all().MOV(offset(payload, bld, 3), brw_imm_ud(0u));
fs_reg srcs[URB_LOGICAL_NUM_SRCS]; brw_reg srcs[URB_LOGICAL_NUM_SRCS];
srcs[URB_LOGICAL_SRC_HANDLE] = uniform_urb_handle; srcs[URB_LOGICAL_SRC_HANDLE] = uniform_urb_handle;
srcs[URB_LOGICAL_SRC_CHANNEL_MASK] = uniform_mask; srcs[URB_LOGICAL_SRC_CHANNEL_MASK] = uniform_mask;
srcs[URB_LOGICAL_SRC_DATA] = payload; srcs[URB_LOGICAL_SRC_DATA] = payload;
@ -958,7 +958,7 @@ void
fs_visitor::emit_urb_fence() fs_visitor::emit_urb_fence()
{ {
const fs_builder bld = fs_builder(this).at_end(); const fs_builder bld = fs_builder(this).at_end();
fs_reg dst = bld.vgrf(BRW_TYPE_UD); brw_reg dst = bld.vgrf(BRW_TYPE_UD);
fs_inst *fence = bld.emit(SHADER_OPCODE_MEMORY_FENCE, dst, fs_inst *fence = bld.emit(SHADER_OPCODE_MEMORY_FENCE, dst,
brw_vec8_grf(0, 0), brw_vec8_grf(0, 0),
brw_imm_ud(true), brw_imm_ud(true),
@ -983,7 +983,7 @@ fs_visitor::emit_cs_terminate()
* make sure it uses the appropriate register range. * make sure it uses the appropriate register range.
*/ */
struct brw_reg g0 = retype(brw_vec8_grf(0, 0), BRW_TYPE_UD); struct brw_reg g0 = retype(brw_vec8_grf(0, 0), BRW_TYPE_UD);
fs_reg payload = brw_vgrf(alloc.allocate(reg_unit(devinfo)), brw_reg payload = brw_vgrf(alloc.allocate(reg_unit(devinfo)),
BRW_TYPE_UD); BRW_TYPE_UD);
ubld.group(8 * reg_unit(devinfo), 0).MOV(payload, g0); ubld.group(8 * reg_unit(devinfo), 0).MOV(payload, g0);
@ -999,11 +999,11 @@ fs_visitor::emit_cs_terminate()
if (devinfo->ver < 11) if (devinfo->ver < 11)
desc |= (1 << 4); /* Do not dereference URB */ desc |= (1 << 4); /* Do not dereference URB */
fs_reg srcs[4] = { brw_reg srcs[4] = {
brw_imm_ud(desc), /* desc */ brw_imm_ud(desc), /* desc */
brw_imm_ud(0), /* ex_desc */ brw_imm_ud(0), /* ex_desc */
payload, /* payload */ payload, /* payload */
fs_reg(), /* payload2 */ brw_reg(), /* payload2 */
}; };
fs_inst *send = ubld.emit(SHADER_OPCODE_SEND, reg_undef, srcs, 4); fs_inst *send = ubld.emit(SHADER_OPCODE_SEND, reg_undef, srcs, 4);

6
src/intel/compiler/brw_fs_workaround.cpp
View file

@ -103,7 +103,7 @@ brw_fs_workaround_memory_fence_before_eot(fs_visitor &s)
const fs_builder ibld(&s, block, inst); const fs_builder ibld(&s, block, inst);
const fs_builder ubld = ibld.exec_all().group(1, 0); const fs_builder ubld = ibld.exec_all().group(1, 0);
fs_reg dst = ubld.vgrf(BRW_TYPE_UD); brw_reg dst = ubld.vgrf(BRW_TYPE_UD);
fs_inst *dummy_fence = ubld.emit(SHADER_OPCODE_MEMORY_FENCE, fs_inst *dummy_fence = ubld.emit(SHADER_OPCODE_MEMORY_FENCE,
dst, brw_vec8_grf(0, 0), dst, brw_vec8_grf(0, 0),
/* commit enable */ brw_imm_ud(1), /* commit enable */ brw_imm_ud(1),
@ -230,7 +230,7 @@ brw_fs_workaround_nomask_control_flow(fs_visitor &s)
*/ */
const fs_builder ubld = fs_builder(&s, block, inst) const fs_builder ubld = fs_builder(&s, block, inst)
.exec_all().group(s.dispatch_width, 0); .exec_all().group(s.dispatch_width, 0);
const fs_reg flag = retype(brw_flag_reg(0, 0), const brw_reg flag = retype(brw_flag_reg(0, 0),
BRW_TYPE_UD); BRW_TYPE_UD);
/* Due to the lack of flag register allocation we need to save /* Due to the lack of flag register allocation we need to save
@ -238,7 +238,7 @@ brw_fs_workaround_nomask_control_flow(fs_visitor &s)
*/ */
const bool save_flag = flag_liveout & const bool save_flag = flag_liveout &
brw_fs_flag_mask(flag, s.dispatch_width / 8); brw_fs_flag_mask(flag, s.dispatch_width / 8);
const fs_reg tmp = ubld.group(8, 0).vgrf(flag.type); const brw_reg tmp = ubld.group(8, 0).vgrf(flag.type);
if (save_flag) { if (save_flag) {
ubld.group(8, 0).UNDEF(tmp); ubld.group(8, 0).UNDEF(tmp);

82
src/intel/compiler/brw_ir_fs.h
View file

@ -28,10 +28,8 @@
#include "brw_ir.h" #include "brw_ir.h"
#include "brw_ir_allocator.h" #include "brw_ir_allocator.h"
using fs_reg = brw_reg; static inline brw_reg
horiz_offset(const brw_reg &reg, unsigned delta)
static inline fs_reg
horiz_offset(const fs_reg &reg, unsigned delta)
{ {
switch (reg.file) { switch (reg.file) {
case BAD_FILE: case BAD_FILE:
@ -65,8 +63,8 @@ horiz_offset(const fs_reg &reg, unsigned delta)
unreachable("Invalid register file"); unreachable("Invalid register file");
} }
static inline fs_reg static inline brw_reg
offset(fs_reg reg, unsigned width, unsigned delta) offset(brw_reg reg, unsigned width, unsigned delta)
{ {
switch (reg.file) { switch (reg.file) {
case BAD_FILE: case BAD_FILE:
@ -87,8 +85,8 @@ offset(fs_reg reg, unsigned width, unsigned delta)
* Get the scalar channel of \p reg given by \p idx and replicate it to all * Get the scalar channel of \p reg given by \p idx and replicate it to all
* channels of the result. * channels of the result.
*/ */
static inline fs_reg static inline brw_reg
component(fs_reg reg, unsigned idx) component(brw_reg reg, unsigned idx)
{ {
reg = horiz_offset(reg, idx); reg = horiz_offset(reg, idx);
reg.stride = 0; reg.stride = 0;
@ -109,7 +107,7 @@ component(fs_reg reg, unsigned idx)
* address spaces, one for each allocation and input attribute respectively. * address spaces, one for each allocation and input attribute respectively.
*/ */
static inline uint32_t static inline uint32_t
reg_space(const fs_reg &r) reg_space(const brw_reg &r)
{ {
return r.file << 16 | (r.file == VGRF || r.file == ATTR ? r.nr : 0); return r.file << 16 | (r.file == VGRF || r.file == ATTR ? r.nr : 0);
} }
@ -119,7 +117,7 @@ reg_space(const fs_reg &r)
* reg_space(). * reg_space().
*/ */
static inline unsigned static inline unsigned
reg_offset(const fs_reg &r) reg_offset(const brw_reg &r)
{ {
return (r.file == VGRF || r.file == IMM || r.file == ATTR ? 0 : r.nr) * return (r.file == VGRF || r.file == IMM || r.file == ATTR ? 0 : r.nr) *
(r.file == UNIFORM ? 4 : REG_SIZE) + r.offset + (r.file == UNIFORM ? 4 : REG_SIZE) + r.offset +
@ -132,7 +130,7 @@ reg_offset(const fs_reg &r)
* one, or zero if components are tightly packed in the register file. * one, or zero if components are tightly packed in the register file.
*/ */
static inline unsigned static inline unsigned
reg_padding(const fs_reg &r) reg_padding(const brw_reg &r)
{ {
const unsigned stride = ((r.file != ARF && r.file != FIXED_GRF) ? r.stride : const unsigned stride = ((r.file != ARF && r.file != FIXED_GRF) ? r.stride :
r.hstride == 0 ? 0 : r.hstride == 0 ? 0 :
@ -146,7 +144,7 @@ reg_padding(const fs_reg &r)
* spanning \p ds bytes. * spanning \p ds bytes.
*/ */
static inline bool static inline bool
regions_overlap(const fs_reg &r, unsigned dr, const fs_reg &s, unsigned ds) regions_overlap(const brw_reg &r, unsigned dr, const brw_reg &s, unsigned ds)
{ {
if (r.file != s.file) if (r.file != s.file)
return false; return false;
@ -166,7 +164,7 @@ regions_overlap(const fs_reg &r, unsigned dr, const fs_reg &s, unsigned ds)
* [s.offset, s.offset + ds[. * [s.offset, s.offset + ds[.
*/ */
static inline bool static inline bool
region_contained_in(const fs_reg &r, unsigned dr, const fs_reg &s, unsigned ds) region_contained_in(const brw_reg &r, unsigned dr, const brw_reg &s, unsigned ds)
{ {
return reg_space(r) == reg_space(s) && return reg_space(r) == reg_space(s) &&
reg_offset(r) >= reg_offset(s) && reg_offset(r) >= reg_offset(s) &&
@ -179,7 +177,7 @@ region_contained_in(const fs_reg &r, unsigned dr, const fs_reg &s, unsigned ds)
* channels. * channels.
*/ */
static inline bool static inline bool
is_periodic(const fs_reg &reg, unsigned n) is_periodic(const brw_reg &reg, unsigned n)
{ {
if (reg.file == BAD_FILE || reg.is_null()) { if (reg.file == BAD_FILE || reg.is_null()) {
return true; return true;
@ -203,7 +201,7 @@ is_periodic(const fs_reg &reg, unsigned n)
} }
static inline bool static inline bool
is_uniform(const fs_reg &reg) is_uniform(const brw_reg &reg)
{ {
return is_periodic(reg, 1); return is_periodic(reg, 1);
} }
@ -211,47 +209,47 @@ is_uniform(const fs_reg &reg)
/** /**
* Get the specified 8-component quarter of a register. * Get the specified 8-component quarter of a register.
*/ */
static inline fs_reg static inline brw_reg
quarter(const fs_reg &reg, unsigned idx) quarter(const brw_reg &reg, unsigned idx)
{ {
assert(idx < 4); assert(idx < 4);
return horiz_offset(reg, 8 * idx); return horiz_offset(reg, 8 * idx);
} }
static inline fs_reg static inline brw_reg
horiz_stride(fs_reg reg, unsigned s) horiz_stride(brw_reg reg, unsigned s)
{ {
reg.stride *= s; reg.stride *= s;
return reg; return reg;
} }
bool fs_reg_saturate_immediate(fs_reg *reg); bool fs_reg_saturate_immediate(brw_reg *reg);
bool fs_reg_negate_immediate(fs_reg *reg); bool fs_reg_negate_immediate(brw_reg *reg);
bool fs_reg_abs_immediate(fs_reg *reg); bool fs_reg_abs_immediate(brw_reg *reg);
static const fs_reg reg_undef; static const brw_reg reg_undef;
struct fs_inst : public exec_node { struct fs_inst : public exec_node {
private: private:
fs_inst &operator=(const fs_inst &); fs_inst &operator=(const fs_inst &);
void init(enum opcode opcode, uint8_t exec_width, const fs_reg &dst, void init(enum opcode opcode, uint8_t exec_width, const brw_reg &dst,
const fs_reg *src, unsigned sources); const brw_reg *src, unsigned sources);
public: public:
DECLARE_RALLOC_CXX_OPERATORS(fs_inst) DECLARE_RALLOC_CXX_OPERATORS(fs_inst)
fs_inst(); fs_inst();
fs_inst(enum opcode opcode, uint8_t exec_size); fs_inst(enum opcode opcode, uint8_t exec_size);
fs_inst(enum opcode opcode, uint8_t exec_size, const fs_reg &dst); fs_inst(enum opcode opcode, uint8_t exec_size, const brw_reg &dst);
fs_inst(enum opcode opcode, uint8_t exec_size, const fs_reg &dst, fs_inst(enum opcode opcode, uint8_t exec_size, const brw_reg &dst,
const fs_reg &src0); const brw_reg &src0);
fs_inst(enum opcode opcode, uint8_t exec_size, const fs_reg &dst, fs_inst(enum opcode opcode, uint8_t exec_size, const brw_reg &dst,
const fs_reg &src0, const fs_reg &src1); const brw_reg &src0, const brw_reg &src1);
fs_inst(enum opcode opcode, uint8_t exec_size, const fs_reg &dst, fs_inst(enum opcode opcode, uint8_t exec_size, const brw_reg &dst,
const fs_reg &src0, const fs_reg &src1, const fs_reg &src2); const brw_reg &src0, const brw_reg &src1, const brw_reg &src2);
fs_inst(enum opcode opcode, uint8_t exec_size, const fs_reg &dst, fs_inst(enum opcode opcode, uint8_t exec_size, const brw_reg &dst,
const fs_reg src[], unsigned sources); const brw_reg src[], unsigned sources);
fs_inst(const fs_inst &that); fs_inst(const fs_inst &that);
~fs_inst(); ~fs_inst();
@ -332,7 +330,7 @@ public:
const char *annotation; const char *annotation;
/** @} */ /** @} */
uint8_t sources; /**< Number of fs_reg sources. */ uint8_t sources; /**< Number of brw_reg sources. */
/** /**
* Execution size of the instruction. This is used by the generator to * Execution size of the instruction. This is used by the generator to
@ -425,9 +423,9 @@ public:
uint32_t bits; uint32_t bits;
}; };
fs_reg dst; brw_reg dst;
fs_reg *src; brw_reg *src;
fs_reg builtin_src[4]; brw_reg builtin_src[4];
}; };
/** /**
@ -587,7 +585,7 @@ is_unordered(const intel_device_info *devinfo, const fs_inst *inst)
* single one-dimensional stride. * single one-dimensional stride.
*/ */
static inline unsigned static inline unsigned
byte_stride(const fs_reg &reg) byte_stride(const brw_reg &reg)
{ {
switch (reg.file) { switch (reg.file) {
case BAD_FILE: case BAD_FILE:
@ -675,7 +673,7 @@ has_dst_aligned_region_restriction(const intel_device_info *devinfo,
static inline bool static inline bool
has_subdword_integer_region_restriction(const intel_device_info *devinfo, has_subdword_integer_region_restriction(const intel_device_info *devinfo,
const fs_inst *inst, const fs_inst *inst,
const fs_reg *srcs, unsigned num_srcs) const brw_reg *srcs, unsigned num_srcs)
{ {
if (devinfo->ver >= 20 && if (devinfo->ver >= 20 &&
brw_type_is_int(inst->dst.type) && brw_type_is_int(inst->dst.type) &&
@ -741,7 +739,7 @@ is_copy_payload(brw_reg_file file, const fs_inst *inst)
inline bool inline bool
is_identity_payload(brw_reg_file file, const fs_inst *inst) { is_identity_payload(brw_reg_file file, const fs_inst *inst) {
if (is_copy_payload(file, inst)) { if (is_copy_payload(file, inst)) {
fs_reg reg = inst->src[0]; brw_reg reg = inst->src[0];
for (unsigned i = 0; i < inst->sources; i++) { for (unsigned i = 0; i < inst->sources; i++) {
reg.type = inst->src[i].type; reg.type = inst->src[i].type;
@ -823,7 +821,7 @@ brw_fs_bit_mask(unsigned n)
} }
static inline unsigned static inline unsigned
brw_fs_flag_mask(const fs_reg &r, unsigned sz) brw_fs_flag_mask(const brw_reg &r, unsigned sz)
{ {
if (r.file == ARF) { if (r.file == ARF) {
const unsigned start = (r.nr - BRW_ARF_FLAG) * 4 + r.subnr; const unsigned start = (r.nr - BRW_ARF_FLAG) * 4 + r.subnr;

2
src/intel/compiler/brw_ir_performance.cpp
View file

@ -778,7 +778,7 @@ namespace {
* Return the dependency ID of a backend_reg, offset by \p delta GRFs. * Return the dependency ID of a backend_reg, offset by \p delta GRFs.
*/ */
enum intel_eu_dependency_id enum intel_eu_dependency_id
reg_dependency_id(const intel_device_info *devinfo, const fs_reg &r, reg_dependency_id(const intel_device_info *devinfo, const brw_reg &r,
const int delta) const int delta)
{ {
if (r.file == VGRF) { if (r.file == VGRF) {

320
src/intel/compiler/brw_lower_logical_sends.cpp
View file

@ -41,13 +41,13 @@ lower_urb_read_logical_send(const fs_builder &bld, fs_inst *inst)
assert(inst->size_written % REG_SIZE == 0); assert(inst->size_written % REG_SIZE == 0);
assert(inst->header_size == 0); assert(inst->header_size == 0);
fs_reg payload_sources[2]; brw_reg payload_sources[2];
unsigned header_size = 0; unsigned header_size = 0;
payload_sources[header_size++] = inst->src[URB_LOGICAL_SRC_HANDLE]; payload_sources[header_size++] = inst->src[URB_LOGICAL_SRC_HANDLE];
if (per_slot_present) if (per_slot_present)
payload_sources[header_size++] = inst->src[URB_LOGICAL_SRC_PER_SLOT_OFFSETS]; payload_sources[header_size++] = inst->src[URB_LOGICAL_SRC_PER_SLOT_OFFSETS];
fs_reg payload = brw_vgrf(bld.shader->alloc.allocate(header_size), brw_reg payload = brw_vgrf(bld.shader->alloc.allocate(header_size),
BRW_TYPE_F); BRW_TYPE_F);
bld.LOAD_PAYLOAD(payload, payload_sources, header_size, header_size); bld.LOAD_PAYLOAD(payload, payload_sources, header_size, header_size);
@ -84,12 +84,12 @@ lower_urb_read_logical_send_xe2(const fs_builder &bld, fs_inst *inst)
assert(inst->header_size == 0); assert(inst->header_size == 0);
/* Get the logical send arguments. */ /* Get the logical send arguments. */
const fs_reg handle = inst->src[URB_LOGICAL_SRC_HANDLE]; const brw_reg handle = inst->src[URB_LOGICAL_SRC_HANDLE];
/* Calculate the total number of components of the payload. */ /* Calculate the total number of components of the payload. */
const unsigned dst_comps = inst->size_written / (REG_SIZE * reg_unit(devinfo)); const unsigned dst_comps = inst->size_written / (REG_SIZE * reg_unit(devinfo));
fs_reg payload = bld.vgrf(BRW_TYPE_UD); brw_reg payload = bld.vgrf(BRW_TYPE_UD);
bld.MOV(payload, handle); bld.MOV(payload, handle);
@ -101,7 +101,7 @@ lower_urb_read_logical_send_xe2(const fs_builder &bld, fs_inst *inst)
inst->offset = 0; inst->offset = 0;
} }
fs_reg offsets = inst->src[URB_LOGICAL_SRC_PER_SLOT_OFFSETS]; brw_reg offsets = inst->src[URB_LOGICAL_SRC_PER_SLOT_OFFSETS];
if (offsets.file != BAD_FILE) { if (offsets.file != BAD_FILE) {
bld.ADD(payload, payload, offsets); bld.ADD(payload, payload, offsets);
} }
@ -147,8 +147,8 @@ lower_urb_write_logical_send(const fs_builder &bld, fs_inst *inst)
const unsigned length = 1 + per_slot_present + channel_mask_present + const unsigned length = 1 + per_slot_present + channel_mask_present +
inst->components_read(URB_LOGICAL_SRC_DATA); inst->components_read(URB_LOGICAL_SRC_DATA);
fs_reg *payload_sources = new fs_reg[length]; brw_reg *payload_sources = new brw_reg[length];
fs_reg payload = brw_vgrf(bld.shader->alloc.allocate(length), brw_reg payload = brw_vgrf(bld.shader->alloc.allocate(length),
BRW_TYPE_F); BRW_TYPE_F);
unsigned header_size = 0; unsigned header_size = 0;
@ -197,16 +197,16 @@ lower_urb_write_logical_send_xe2(const fs_builder &bld, fs_inst *inst)
assert(devinfo->has_lsc); assert(devinfo->has_lsc);
/* Get the logical send arguments. */ /* Get the logical send arguments. */
const fs_reg handle = inst->src[URB_LOGICAL_SRC_HANDLE]; const brw_reg handle = inst->src[URB_LOGICAL_SRC_HANDLE];
const fs_reg src = inst->components_read(URB_LOGICAL_SRC_DATA) ? const brw_reg src = inst->components_read(URB_LOGICAL_SRC_DATA) ?
inst->src[URB_LOGICAL_SRC_DATA] : fs_reg(brw_imm_ud(0)); inst->src[URB_LOGICAL_SRC_DATA] : brw_reg(brw_imm_ud(0));
assert(brw_type_size_bytes(src.type) == 4); assert(brw_type_size_bytes(src.type) == 4);
/* Calculate the total number of components of the payload. */ /* Calculate the total number of components of the payload. */
const unsigned src_comps = MAX2(1, inst->components_read(URB_LOGICAL_SRC_DATA)); const unsigned src_comps = MAX2(1, inst->components_read(URB_LOGICAL_SRC_DATA));
const unsigned src_sz = brw_type_size_bytes(src.type); const unsigned src_sz = brw_type_size_bytes(src.type);
fs_reg payload = bld.vgrf(BRW_TYPE_UD); brw_reg payload = bld.vgrf(BRW_TYPE_UD);
bld.MOV(payload, handle); bld.MOV(payload, handle);
@ -218,12 +218,12 @@ lower_urb_write_logical_send_xe2(const fs_builder &bld, fs_inst *inst)
inst->offset = 0; inst->offset = 0;
} }
fs_reg offsets = inst->src[URB_LOGICAL_SRC_PER_SLOT_OFFSETS]; brw_reg offsets = inst->src[URB_LOGICAL_SRC_PER_SLOT_OFFSETS];
if (offsets.file != BAD_FILE) { if (offsets.file != BAD_FILE) {
bld.ADD(payload, payload, offsets); bld.ADD(payload, payload, offsets);
} }
const fs_reg cmask = inst->src[URB_LOGICAL_SRC_CHANNEL_MASK]; const brw_reg cmask = inst->src[URB_LOGICAL_SRC_CHANNEL_MASK];
unsigned mask = 0; unsigned mask = 0;
if (cmask.file != BAD_FILE) { if (cmask.file != BAD_FILE) {
@ -232,7 +232,7 @@ lower_urb_write_logical_send_xe2(const fs_builder &bld, fs_inst *inst)
mask = cmask.ud >> 16; mask = cmask.ud >> 16;
} }
fs_reg payload2 = bld.move_to_vgrf(src, src_comps); brw_reg payload2 = bld.move_to_vgrf(src, src_comps);
const unsigned ex_mlen = (src_comps * src_sz * inst->exec_size) / REG_SIZE; const unsigned ex_mlen = (src_comps * src_sz * inst->exec_size) / REG_SIZE;
inst->sfid = BRW_SFID_URB; inst->sfid = BRW_SFID_URB;
@ -265,10 +265,10 @@ lower_urb_write_logical_send_xe2(const fs_builder &bld, fs_inst *inst)
static void static void
setup_color_payload(const fs_builder &bld, const brw_wm_prog_key *key, setup_color_payload(const fs_builder &bld, const brw_wm_prog_key *key,
fs_reg *dst, fs_reg color, unsigned components) brw_reg *dst, brw_reg color, unsigned components)
{ {
if (key->clamp_fragment_color) { if (key->clamp_fragment_color) {
fs_reg tmp = bld.vgrf(BRW_TYPE_F, 4); brw_reg tmp = bld.vgrf(BRW_TYPE_F, 4);
assert(color.type == BRW_TYPE_F); assert(color.type == BRW_TYPE_F);
for (unsigned i = 0; i < components; i++) for (unsigned i = 0; i < components; i++)
@ -290,19 +290,19 @@ lower_fb_write_logical_send(const fs_builder &bld, fs_inst *inst,
{ {
assert(inst->src[FB_WRITE_LOGICAL_SRC_COMPONENTS].file == IMM); assert(inst->src[FB_WRITE_LOGICAL_SRC_COMPONENTS].file == IMM);
const intel_device_info *devinfo = bld.shader->devinfo; const intel_device_info *devinfo = bld.shader->devinfo;
const fs_reg color0 = inst->src[FB_WRITE_LOGICAL_SRC_COLOR0]; const brw_reg color0 = inst->src[FB_WRITE_LOGICAL_SRC_COLOR0];
const fs_reg color1 = inst->src[FB_WRITE_LOGICAL_SRC_COLOR1]; const brw_reg color1 = inst->src[FB_WRITE_LOGICAL_SRC_COLOR1];
const fs_reg src0_alpha = inst->src[FB_WRITE_LOGICAL_SRC_SRC0_ALPHA]; const brw_reg src0_alpha = inst->src[FB_WRITE_LOGICAL_SRC_SRC0_ALPHA];
const fs_reg src_depth = inst->src[FB_WRITE_LOGICAL_SRC_SRC_DEPTH]; const brw_reg src_depth = inst->src[FB_WRITE_LOGICAL_SRC_SRC_DEPTH];
const fs_reg dst_depth = inst->src[FB_WRITE_LOGICAL_SRC_DST_DEPTH]; const brw_reg dst_depth = inst->src[FB_WRITE_LOGICAL_SRC_DST_DEPTH];
const fs_reg src_stencil = inst->src[FB_WRITE_LOGICAL_SRC_SRC_STENCIL]; const brw_reg src_stencil = inst->src[FB_WRITE_LOGICAL_SRC_SRC_STENCIL];
fs_reg sample_mask = inst->src[FB_WRITE_LOGICAL_SRC_OMASK]; brw_reg sample_mask = inst->src[FB_WRITE_LOGICAL_SRC_OMASK];
const unsigned components = const unsigned components =
inst->src[FB_WRITE_LOGICAL_SRC_COMPONENTS].ud; inst->src[FB_WRITE_LOGICAL_SRC_COMPONENTS].ud;
assert(inst->target != 0 || src0_alpha.file == BAD_FILE); assert(inst->target != 0 || src0_alpha.file == BAD_FILE);
fs_reg sources[15]; brw_reg sources[15];
int header_size = 2, payload_header_size; int header_size = 2, payload_header_size;
unsigned length = 0; unsigned length = 0;
@ -319,7 +319,7 @@ lower_fb_write_logical_send(const fs_builder &bld, fs_inst *inst,
*/ */
const fs_builder ubld = bld.exec_all().group(8, 0); const fs_builder ubld = bld.exec_all().group(8, 0);
fs_reg header = ubld.vgrf(BRW_TYPE_UD, 2); brw_reg header = ubld.vgrf(BRW_TYPE_UD, 2);
if (bld.group() < 16) { if (bld.group() < 16) {
/* The header starts off as g0 and g1 for the first half */ /* The header starts off as g0 and g1 for the first half */
ubld.group(16, 0).MOV(header, retype(brw_vec8_grf(0, 0), ubld.group(16, 0).MOV(header, retype(brw_vec8_grf(0, 0),
@ -327,7 +327,7 @@ lower_fb_write_logical_send(const fs_builder &bld, fs_inst *inst,
} else { } else {
/* The header starts off as g0 and g2 for the second half */ /* The header starts off as g0 and g2 for the second half */
assert(bld.group() < 32); assert(bld.group() < 32);
const fs_reg header_sources[2] = { const brw_reg header_sources[2] = {
retype(brw_vec8_grf(0, 0), BRW_TYPE_UD), retype(brw_vec8_grf(0, 0), BRW_TYPE_UD),
retype(brw_vec8_grf(2, 0), BRW_TYPE_UD), retype(brw_vec8_grf(2, 0), BRW_TYPE_UD),
}; };
@ -379,7 +379,7 @@ lower_fb_write_logical_send(const fs_builder &bld, fs_inst *inst,
sources[length] = brw_vgrf(bld.shader->alloc.allocate(1), BRW_TYPE_F); sources[length] = brw_vgrf(bld.shader->alloc.allocate(1), BRW_TYPE_F);
bld.group(8, 0).exec_all().annotate("FB write stencil/AA alpha") bld.group(8, 0).exec_all().annotate("FB write stencil/AA alpha")
.MOV(sources[length], .MOV(sources[length],
fs_reg(brw_vec8_grf(fs_payload.aa_dest_stencil_reg[0], 0))); brw_reg(brw_vec8_grf(fs_payload.aa_dest_stencil_reg[0], 0)));
length++; length++;
} }
@ -387,7 +387,7 @@ lower_fb_write_logical_send(const fs_builder &bld, fs_inst *inst,
for (unsigned i = 0; i < bld.dispatch_width() / 8; i++) { for (unsigned i = 0; i < bld.dispatch_width() / 8; i++) {
const fs_builder &ubld = bld.exec_all().group(8, i) const fs_builder &ubld = bld.exec_all().group(8, i)
.annotate("FB write src0 alpha"); .annotate("FB write src0 alpha");
const fs_reg tmp = ubld.vgrf(BRW_TYPE_F); const brw_reg tmp = ubld.vgrf(BRW_TYPE_F);
ubld.MOV(tmp, horiz_offset(src0_alpha, i * 8)); ubld.MOV(tmp, horiz_offset(src0_alpha, i * 8));
setup_color_payload(ubld, key, &sources[length], tmp, 1); setup_color_payload(ubld, key, &sources[length], tmp, 1);
length++; length++;
@ -395,7 +395,7 @@ lower_fb_write_logical_send(const fs_builder &bld, fs_inst *inst,
} }
if (sample_mask.file != BAD_FILE) { if (sample_mask.file != BAD_FILE) {
const fs_reg tmp = brw_vgrf(bld.shader->alloc.allocate(reg_unit(devinfo)), const brw_reg tmp = brw_vgrf(bld.shader->alloc.allocate(reg_unit(devinfo)),
BRW_TYPE_UD); BRW_TYPE_UD);
/* Hand over gl_SampleMask. Only the lower 16 bits of each channel are /* Hand over gl_SampleMask. Only the lower 16 bits of each channel are
@ -455,7 +455,7 @@ lower_fb_write_logical_send(const fs_builder &bld, fs_inst *inst,
} }
/* Send from the GRF */ /* Send from the GRF */
fs_reg payload = brw_vgrf(-1, BRW_TYPE_F); brw_reg payload = brw_vgrf(-1, BRW_TYPE_F);
fs_inst *load = bld.LOAD_PAYLOAD(payload, sources, length, payload_header_size); fs_inst *load = bld.LOAD_PAYLOAD(payload, sources, length, payload_header_size);
payload.nr = bld.shader->alloc.allocate(regs_written(load)); payload.nr = bld.shader->alloc.allocate(regs_written(load));
load->dst = payload; load->dst = payload;
@ -468,7 +468,7 @@ lower_fb_write_logical_send(const fs_builder &bld, fs_inst *inst,
brw_fb_write_desc(devinfo, inst->target, msg_ctl, inst->last_rt, brw_fb_write_desc(devinfo, inst->target, msg_ctl, inst->last_rt,
0 /* coarse_rt_write */); 0 /* coarse_rt_write */);
fs_reg desc = brw_imm_ud(0); brw_reg desc = brw_imm_ud(0);
if (prog_data->coarse_pixel_dispatch == BRW_ALWAYS) { if (prog_data->coarse_pixel_dispatch == BRW_ALWAYS) {
inst->desc |= (1 << 18); inst->desc |= (1 << 18);
} else if (prog_data->coarse_pixel_dispatch == BRW_SOMETIMES) { } else if (prog_data->coarse_pixel_dispatch == BRW_SOMETIMES) {
@ -519,7 +519,7 @@ lower_fb_read_logical_send(const fs_builder &bld, fs_inst *inst,
const intel_device_info *devinfo = bld.shader->devinfo; const intel_device_info *devinfo = bld.shader->devinfo;
const fs_builder &ubld = bld.exec_all().group(8, 0); const fs_builder &ubld = bld.exec_all().group(8, 0);
const unsigned length = 2; const unsigned length = 2;
const fs_reg header = ubld.vgrf(BRW_TYPE_UD, length); const brw_reg header = ubld.vgrf(BRW_TYPE_UD, length);
assert(devinfo->ver >= 9 && devinfo->ver < 20); assert(devinfo->ver >= 9 && devinfo->ver < 20);
@ -528,7 +528,7 @@ lower_fb_read_logical_send(const fs_builder &bld, fs_inst *inst,
BRW_TYPE_UD)); BRW_TYPE_UD));
} else { } else {
assert(bld.group() < 32); assert(bld.group() < 32);
const fs_reg header_sources[] = { const brw_reg header_sources[] = {
retype(brw_vec8_grf(0, 0), BRW_TYPE_UD), retype(brw_vec8_grf(0, 0), BRW_TYPE_UD),
retype(brw_vec8_grf(2, 0), BRW_TYPE_UD) retype(brw_vec8_grf(2, 0), BRW_TYPE_UD)
}; };
@ -567,7 +567,7 @@ lower_fb_read_logical_send(const fs_builder &bld, fs_inst *inst,
inst->src[0] = brw_imm_ud(0); inst->src[0] = brw_imm_ud(0);
inst->src[1] = brw_imm_ud(0); inst->src[1] = brw_imm_ud(0);
inst->src[2] = header; inst->src[2] = header;
inst->src[3] = fs_reg(); inst->src[3] = brw_reg();
inst->mlen = length; inst->mlen = length;
inst->header_size = length; inst->header_size = length;
inst->sfid = GFX6_SFID_DATAPORT_RENDER_CACHE; inst->sfid = GFX6_SFID_DATAPORT_RENDER_CACHE;
@ -580,7 +580,7 @@ lower_fb_read_logical_send(const fs_builder &bld, fs_inst *inst,
} }
static bool static bool
is_high_sampler(const struct intel_device_info *devinfo, const fs_reg &sampler) is_high_sampler(const struct intel_device_info *devinfo, const brw_reg &sampler)
{ {
return sampler.file != IMM || sampler.ud >= 16; return sampler.file != IMM || sampler.ud >= 16;
} }
@ -676,15 +676,15 @@ sampler_msg_type(const intel_device_info *devinfo,
* the given requested_alignment_sz. * the given requested_alignment_sz.
*/ */
static fs_inst * static fs_inst *
emit_load_payload_with_padding(const fs_builder &bld, const fs_reg &dst, emit_load_payload_with_padding(const fs_builder &bld, const brw_reg &dst,
const fs_reg *src, unsigned sources, const brw_reg *src, unsigned sources,
unsigned header_size, unsigned header_size,
unsigned requested_alignment_sz) unsigned requested_alignment_sz)
{ {
unsigned length = 0; unsigned length = 0;
unsigned num_srcs = unsigned num_srcs =
sources * DIV_ROUND_UP(requested_alignment_sz, bld.dispatch_width()); sources * DIV_ROUND_UP(requested_alignment_sz, bld.dispatch_width());
fs_reg *src_comps = new fs_reg[num_srcs]; brw_reg *src_comps = new brw_reg[num_srcs];
for (unsigned i = 0; i < header_size; i++) for (unsigned i = 0; i < header_size; i++)
src_comps[length++] = src[i]; src_comps[length++] = src[i];
@ -702,7 +702,7 @@ emit_load_payload_with_padding(const fs_builder &bld, const fs_reg &dst,
*/ */
if (src_sz < requested_alignment_sz) { if (src_sz < requested_alignment_sz) {
for (unsigned j = 0; j < (requested_alignment_sz / src_sz) - 1; j++) { for (unsigned j = 0; j < (requested_alignment_sz / src_sz) - 1; j++) {
src_comps[length++] = retype(fs_reg(), padding_payload_type); src_comps[length++] = retype(brw_reg(), padding_payload_type);
} }
} }
} }
@ -735,17 +735,17 @@ shader_opcode_needs_header(opcode op)
static void static void
lower_sampler_logical_send(const fs_builder &bld, fs_inst *inst, lower_sampler_logical_send(const fs_builder &bld, fs_inst *inst,
const fs_reg &coordinate, const brw_reg &coordinate,
const fs_reg &shadow_c, const brw_reg &shadow_c,
fs_reg lod, const fs_reg &lod2, brw_reg lod, const brw_reg &lod2,
const fs_reg &min_lod, const brw_reg &min_lod,
const fs_reg &sample_index, const brw_reg &sample_index,
const fs_reg &mcs, const brw_reg &mcs,
const fs_reg &surface, const brw_reg &surface,
const fs_reg &sampler, const brw_reg &sampler,
const fs_reg &surface_handle, const brw_reg &surface_handle,
const fs_reg &sampler_handle, const brw_reg &sampler_handle,
const fs_reg &tg4_offset, const brw_reg &tg4_offset,
unsigned payload_type_bit_size, unsigned payload_type_bit_size,
unsigned coord_components, unsigned coord_components,
unsigned grad_components, unsigned grad_components,
@ -762,7 +762,7 @@ lower_sampler_logical_send(const fs_builder &bld, fs_inst *inst,
unsigned reg_width = bld.dispatch_width() / 8; unsigned reg_width = bld.dispatch_width() / 8;
unsigned header_size = 0, length = 0; unsigned header_size = 0, length = 0;
opcode op = inst->opcode; opcode op = inst->opcode;
fs_reg sources[1 + MAX_SAMPLER_MESSAGE_SIZE]; brw_reg sources[1 + MAX_SAMPLER_MESSAGE_SIZE];
for (unsigned i = 0; i < ARRAY_SIZE(sources); i++) for (unsigned i = 0; i < ARRAY_SIZE(sources); i++)
sources[i] = bld.vgrf(payload_type); sources[i] = bld.vgrf(payload_type);
@ -782,7 +782,7 @@ lower_sampler_logical_send(const fs_builder &bld, fs_inst *inst,
* larger sampler numbers we need to offset the Sampler State Pointer in * larger sampler numbers we need to offset the Sampler State Pointer in
* the header. * the header.
*/ */
fs_reg header = retype(sources[0], BRW_TYPE_UD); brw_reg header = retype(sources[0], BRW_TYPE_UD);
for (header_size = 0; header_size < reg_unit(devinfo); header_size++) for (header_size = 0; header_size < reg_unit(devinfo); header_size++)
sources[length++] = byte_offset(header, REG_SIZE * header_size); sources[length++] = byte_offset(header, REG_SIZE * header_size);
@ -840,7 +840,7 @@ lower_sampler_logical_send(const fs_builder &bld, fs_inst *inst,
ubld1.MOV(component(header, 3), sampler_handle); ubld1.MOV(component(header, 3), sampler_handle);
} }
} else if (is_high_sampler(devinfo, sampler)) { } else if (is_high_sampler(devinfo, sampler)) {
fs_reg sampler_state_ptr = brw_reg sampler_state_ptr =
retype(brw_vec1_grf(0, 3), BRW_TYPE_UD); retype(brw_vec1_grf(0, 3), BRW_TYPE_UD);
/* Gfx11+ sampler message headers include bits in 4:0 which conflict /* Gfx11+ sampler message headers include bits in 4:0 which conflict
@ -860,7 +860,7 @@ lower_sampler_logical_send(const fs_builder &bld, fs_inst *inst,
ubld1.ADD(component(header, 3), sampler_state_ptr, ubld1.ADD(component(header, 3), sampler_state_ptr,
brw_imm_ud(16 * (sampler.ud / 16) * sampler_state_size)); brw_imm_ud(16 * (sampler.ud / 16) * sampler_state_size));
} else { } else {
fs_reg tmp = ubld1.vgrf(BRW_TYPE_UD); brw_reg tmp = ubld1.vgrf(BRW_TYPE_UD);
ubld1.AND(tmp, sampler, brw_imm_ud(0x0f0)); ubld1.AND(tmp, sampler, brw_imm_ud(0x0f0));
ubld1.SHL(tmp, tmp, brw_imm_ud(4)); ubld1.SHL(tmp, tmp, brw_imm_ud(4));
ubld1.ADD(component(header, 3), sampler_state_ptr, tmp); ubld1.ADD(component(header, 3), sampler_state_ptr, tmp);
@ -991,16 +991,16 @@ lower_sampler_logical_send(const fs_builder &bld, fs_inst *inst,
* ld2dms_w si mcs0 mcs1 mcs2 mcs3 u v r * ld2dms_w si mcs0 mcs1 mcs2 mcs3 u v r
*/ */
if (op == SHADER_OPCODE_TXF_CMS_W_GFX12_LOGICAL) { if (op == SHADER_OPCODE_TXF_CMS_W_GFX12_LOGICAL) {
fs_reg tmp = offset(mcs, bld, i); brw_reg tmp = offset(mcs, bld, i);
sources[length] = retype(sources[length], payload_unsigned_type); sources[length] = retype(sources[length], payload_unsigned_type);
bld.MOV(sources[length++], bld.MOV(sources[length++],
mcs.file == IMM ? mcs : mcs.file == IMM ? mcs :
fs_reg(subscript(tmp, payload_unsigned_type, 0))); brw_reg(subscript(tmp, payload_unsigned_type, 0)));
sources[length] = retype(sources[length], payload_unsigned_type); sources[length] = retype(sources[length], payload_unsigned_type);
bld.MOV(sources[length++], bld.MOV(sources[length++],
mcs.file == IMM ? mcs : mcs.file == IMM ? mcs :
fs_reg(subscript(tmp, payload_unsigned_type, 1))); brw_reg(subscript(tmp, payload_unsigned_type, 1)));
} else { } else {
sources[length] = retype(sources[length], payload_unsigned_type); sources[length] = retype(sources[length], payload_unsigned_type);
bld.MOV(sources[length++], bld.MOV(sources[length++],
@ -1087,7 +1087,7 @@ lower_sampler_logical_send(const fs_builder &bld, fs_inst *inst,
bld.MOV(sources[length++], min_lod); bld.MOV(sources[length++], min_lod);
} }
const fs_reg src_payload = const brw_reg src_payload =
brw_vgrf(bld.shader->alloc.allocate(length * reg_width), brw_vgrf(bld.shader->alloc.allocate(length * reg_width),
BRW_TYPE_F); BRW_TYPE_F);
/* In case of 16-bit payload each component takes one full register in /* In case of 16-bit payload each component takes one full register in
@ -1149,7 +1149,7 @@ lower_sampler_logical_send(const fs_builder &bld, fs_inst *inst,
inst->src[0] = brw_imm_ud(0); inst->src[0] = brw_imm_ud(0);
} else { } else {
const fs_builder ubld = bld.group(1, 0).exec_all(); const fs_builder ubld = bld.group(1, 0).exec_all();
fs_reg desc = ubld.vgrf(BRW_TYPE_UD); brw_reg desc = ubld.vgrf(BRW_TYPE_UD);
ubld.SHL(desc, sampler, brw_imm_ud(8)); ubld.SHL(desc, sampler, brw_imm_ud(8));
inst->src[0] = component(desc, 0); inst->src[0] = component(desc, 0);
} }
@ -1168,7 +1168,7 @@ lower_sampler_logical_send(const fs_builder &bld, fs_inst *inst,
simd_mode, simd_mode,
0 /* return_format unused on gfx7+ */); 0 /* return_format unused on gfx7+ */);
const fs_builder ubld = bld.group(1, 0).exec_all(); const fs_builder ubld = bld.group(1, 0).exec_all();
fs_reg desc = ubld.vgrf(BRW_TYPE_UD); brw_reg desc = ubld.vgrf(BRW_TYPE_UD);
if (surface.equals(sampler)) { if (surface.equals(sampler)) {
/* This case is common in GL */ /* This case is common in GL */
ubld.MUL(desc, surface, brw_imm_ud(0x101)); ubld.MUL(desc, surface, brw_imm_ud(0x101));
@ -1212,7 +1212,7 @@ get_sampler_msg_payload_type_bit_size(const intel_device_info *devinfo,
const fs_inst *inst) const fs_inst *inst)
{ {
assert(inst); assert(inst);
const fs_reg *src = inst->src; const brw_reg *src = inst->src;
unsigned src_type_size = 0; unsigned src_type_size = 0;
/* All sources need to have the same size, therefore seek the first valid /* All sources need to have the same size, therefore seek the first valid
@ -1263,18 +1263,18 @@ static void
lower_sampler_logical_send(const fs_builder &bld, fs_inst *inst) lower_sampler_logical_send(const fs_builder &bld, fs_inst *inst)
{ {
const intel_device_info *devinfo = bld.shader->devinfo; const intel_device_info *devinfo = bld.shader->devinfo;
const fs_reg coordinate = inst->src[TEX_LOGICAL_SRC_COORDINATE]; const brw_reg coordinate = inst->src[TEX_LOGICAL_SRC_COORDINATE];
const fs_reg shadow_c = inst->src[TEX_LOGICAL_SRC_SHADOW_C]; const brw_reg shadow_c = inst->src[TEX_LOGICAL_SRC_SHADOW_C];
const fs_reg lod = inst->src[TEX_LOGICAL_SRC_LOD]; const brw_reg lod = inst->src[TEX_LOGICAL_SRC_LOD];
const fs_reg lod2 = inst->src[TEX_LOGICAL_SRC_LOD2]; const brw_reg lod2 = inst->src[TEX_LOGICAL_SRC_LOD2];
const fs_reg min_lod = inst->src[TEX_LOGICAL_SRC_MIN_LOD]; const brw_reg min_lod = inst->src[TEX_LOGICAL_SRC_MIN_LOD];
const fs_reg sample_index = inst->src[TEX_LOGICAL_SRC_SAMPLE_INDEX]; const brw_reg sample_index = inst->src[TEX_LOGICAL_SRC_SAMPLE_INDEX];
const fs_reg mcs = inst->src[TEX_LOGICAL_SRC_MCS]; const brw_reg mcs = inst->src[TEX_LOGICAL_SRC_MCS];
const fs_reg surface = inst->src[TEX_LOGICAL_SRC_SURFACE]; const brw_reg surface = inst->src[TEX_LOGICAL_SRC_SURFACE];
const fs_reg sampler = inst->src[TEX_LOGICAL_SRC_SAMPLER]; const brw_reg sampler = inst->src[TEX_LOGICAL_SRC_SAMPLER];
const fs_reg surface_handle = inst->src[TEX_LOGICAL_SRC_SURFACE_HANDLE]; const brw_reg surface_handle = inst->src[TEX_LOGICAL_SRC_SURFACE_HANDLE];
const fs_reg sampler_handle = inst->src[TEX_LOGICAL_SRC_SAMPLER_HANDLE]; const brw_reg sampler_handle = inst->src[TEX_LOGICAL_SRC_SAMPLER_HANDLE];
const fs_reg tg4_offset = inst->src[TEX_LOGICAL_SRC_TG4_OFFSET]; const brw_reg tg4_offset = inst->src[TEX_LOGICAL_SRC_TG4_OFFSET];
assert(inst->src[TEX_LOGICAL_SRC_COORD_COMPONENTS].file == IMM); assert(inst->src[TEX_LOGICAL_SRC_COORD_COMPONENTS].file == IMM);
const unsigned coord_components = inst->src[TEX_LOGICAL_SRC_COORD_COMPONENTS].ud; const unsigned coord_components = inst->src[TEX_LOGICAL_SRC_COORD_COMPONENTS].ud;
assert(inst->src[TEX_LOGICAL_SRC_GRAD_COMPONENTS].file == IMM); assert(inst->src[TEX_LOGICAL_SRC_GRAD_COMPONENTS].file == IMM);
@ -1312,7 +1312,7 @@ emit_predicate_on_vector_mask(const fs_builder &bld, fs_inst *inst)
const fs_builder ubld = bld.exec_all().group(1, 0); const fs_builder ubld = bld.exec_all().group(1, 0);
const fs_visitor &s = *bld.shader; const fs_visitor &s = *bld.shader;
const fs_reg vector_mask = ubld.vgrf(BRW_TYPE_UW); const brw_reg vector_mask = ubld.vgrf(BRW_TYPE_UW);
ubld.UNDEF(vector_mask); ubld.UNDEF(vector_mask);
ubld.emit(SHADER_OPCODE_READ_ARCH_REG, vector_mask, retype(brw_sr0_reg(3), ubld.emit(SHADER_OPCODE_READ_ARCH_REG, vector_mask, retype(brw_sr0_reg(3),
BRW_TYPE_UD)); BRW_TYPE_UD));
@ -1338,7 +1338,7 @@ emit_predicate_on_vector_mask(const fs_builder &bld, fs_inst *inst)
static void static void
setup_surface_descriptors(const fs_builder &bld, fs_inst *inst, uint32_t desc, setup_surface_descriptors(const fs_builder &bld, fs_inst *inst, uint32_t desc,
const fs_reg &surface, const fs_reg &surface_handle) const brw_reg &surface, const brw_reg &surface_handle)
{ {
const brw_compiler *compiler = bld.shader->compiler; const brw_compiler *compiler = bld.shader->compiler;
@ -1362,7 +1362,7 @@ setup_surface_descriptors(const fs_builder &bld, fs_inst *inst, uint32_t desc,
} else { } else {
inst->desc = desc; inst->desc = desc;
const fs_builder ubld = bld.exec_all().group(1, 0); const fs_builder ubld = bld.exec_all().group(1, 0);
fs_reg tmp = ubld.vgrf(BRW_TYPE_UD); brw_reg tmp = ubld.vgrf(BRW_TYPE_UD);
ubld.AND(tmp, surface, brw_imm_ud(0xff)); ubld.AND(tmp, surface, brw_imm_ud(0xff));
inst->src[0] = component(tmp, 0); inst->src[0] = component(tmp, 0);
inst->src[1] = brw_imm_ud(0); /* ex_desc */ inst->src[1] = brw_imm_ud(0); /* ex_desc */
@ -1371,7 +1371,7 @@ setup_surface_descriptors(const fs_builder &bld, fs_inst *inst, uint32_t desc,
static void static void
setup_lsc_surface_descriptors(const fs_builder &bld, fs_inst *inst, setup_lsc_surface_descriptors(const fs_builder &bld, fs_inst *inst,
uint32_t desc, const fs_reg &surface) uint32_t desc, const brw_reg &surface)
{ {
const ASSERTED intel_device_info *devinfo = bld.shader->devinfo; const ASSERTED intel_device_info *devinfo = bld.shader->devinfo;
const brw_compiler *compiler = bld.shader->compiler; const brw_compiler *compiler = bld.shader->compiler;
@ -1397,7 +1397,7 @@ setup_lsc_surface_descriptors(const fs_builder &bld, fs_inst *inst,
inst->src[1] = brw_imm_ud(lsc_bti_ex_desc(devinfo, surface.ud)); inst->src[1] = brw_imm_ud(lsc_bti_ex_desc(devinfo, surface.ud));
} else { } else {
const fs_builder ubld = bld.exec_all().group(1, 0); const fs_builder ubld = bld.exec_all().group(1, 0);
fs_reg tmp = ubld.vgrf(BRW_TYPE_UD); brw_reg tmp = ubld.vgrf(BRW_TYPE_UD);
ubld.SHL(tmp, surface, brw_imm_ud(24)); ubld.SHL(tmp, surface, brw_imm_ud(24));
inst->src[1] = component(tmp, 0); inst->src[1] = component(tmp, 0);
} }
@ -1419,13 +1419,13 @@ lower_surface_logical_send(const fs_builder &bld, fs_inst *inst)
const intel_device_info *devinfo = bld.shader->devinfo; const intel_device_info *devinfo = bld.shader->devinfo;
/* Get the logical send arguments. */ /* Get the logical send arguments. */
const fs_reg addr = inst->src[SURFACE_LOGICAL_SRC_ADDRESS]; const brw_reg addr = inst->src[SURFACE_LOGICAL_SRC_ADDRESS];
const fs_reg src = inst->src[SURFACE_LOGICAL_SRC_DATA]; const brw_reg src = inst->src[SURFACE_LOGICAL_SRC_DATA];
const fs_reg surface = inst->src[SURFACE_LOGICAL_SRC_SURFACE]; const brw_reg surface = inst->src[SURFACE_LOGICAL_SRC_SURFACE];
const fs_reg surface_handle = inst->src[SURFACE_LOGICAL_SRC_SURFACE_HANDLE]; const brw_reg surface_handle = inst->src[SURFACE_LOGICAL_SRC_SURFACE_HANDLE];
const UNUSED fs_reg dims = inst->src[SURFACE_LOGICAL_SRC_IMM_DIMS]; const UNUSED brw_reg dims = inst->src[SURFACE_LOGICAL_SRC_IMM_DIMS];
const fs_reg arg = inst->src[SURFACE_LOGICAL_SRC_IMM_ARG]; const brw_reg arg = inst->src[SURFACE_LOGICAL_SRC_IMM_ARG];
const fs_reg allow_sample_mask = const brw_reg allow_sample_mask =
inst->src[SURFACE_LOGICAL_SRC_ALLOW_SAMPLE_MASK]; inst->src[SURFACE_LOGICAL_SRC_ALLOW_SAMPLE_MASK];
assert(arg.file == IMM); assert(arg.file == IMM);
assert(allow_sample_mask.file == IMM); assert(allow_sample_mask.file == IMM);
@ -1450,10 +1450,10 @@ lower_surface_logical_send(const fs_builder &bld, fs_inst *inst)
const bool has_side_effects = inst->has_side_effects(); const bool has_side_effects = inst->has_side_effects();
fs_reg sample_mask = allow_sample_mask.ud ? brw_sample_mask_reg(bld) : brw_reg sample_mask = allow_sample_mask.ud ? brw_sample_mask_reg(bld) :
fs_reg(brw_imm_ud(0xffffffff)); brw_reg(brw_imm_ud(0xffffffff));
fs_reg header; brw_reg header;
if (is_stateless) { if (is_stateless) {
assert(!is_surface_access); assert(!is_surface_access);
fs_builder ubld = bld.exec_all().group(8, 0); fs_builder ubld = bld.exec_all().group(8, 0);
@ -1462,7 +1462,7 @@ lower_surface_logical_send(const fs_builder &bld, fs_inst *inst)
} }
const unsigned header_sz = header.file != BAD_FILE ? 1 : 0; const unsigned header_sz = header.file != BAD_FILE ? 1 : 0;
fs_reg payload, payload2; brw_reg payload, payload2;
unsigned mlen, ex_mlen = 0; unsigned mlen, ex_mlen = 0;
if (src.file == BAD_FILE || header.file == BAD_FILE) { if (src.file == BAD_FILE || header.file == BAD_FILE) {
/* We have split sends on gfx9 and above */ /* We have split sends on gfx9 and above */
@ -1482,7 +1482,7 @@ lower_surface_logical_send(const fs_builder &bld, fs_inst *inst)
/* Allocate space for the payload. */ /* Allocate space for the payload. */
const unsigned sz = header_sz + addr_sz + src_sz; const unsigned sz = header_sz + addr_sz + src_sz;
payload = bld.vgrf(BRW_TYPE_UD, sz); payload = bld.vgrf(BRW_TYPE_UD, sz);
fs_reg *const components = new fs_reg[sz]; brw_reg *const components = new brw_reg[sz];
unsigned n = 0; unsigned n = 0;
/* Construct the payload. */ /* Construct the payload. */
@ -1657,13 +1657,13 @@ lower_lsc_surface_logical_send(const fs_builder &bld, fs_inst *inst)
assert(devinfo->has_lsc); assert(devinfo->has_lsc);
/* Get the logical send arguments. */ /* Get the logical send arguments. */
const fs_reg addr = inst->src[SURFACE_LOGICAL_SRC_ADDRESS]; const brw_reg addr = inst->src[SURFACE_LOGICAL_SRC_ADDRESS];
const fs_reg src = inst->src[SURFACE_LOGICAL_SRC_DATA]; const brw_reg src = inst->src[SURFACE_LOGICAL_SRC_DATA];
const fs_reg surface = inst->src[SURFACE_LOGICAL_SRC_SURFACE]; const brw_reg surface = inst->src[SURFACE_LOGICAL_SRC_SURFACE];
const fs_reg surface_handle = inst->src[SURFACE_LOGICAL_SRC_SURFACE_HANDLE]; const brw_reg surface_handle = inst->src[SURFACE_LOGICAL_SRC_SURFACE_HANDLE];
const fs_reg dims = inst->src[SURFACE_LOGICAL_SRC_IMM_DIMS]; const brw_reg dims = inst->src[SURFACE_LOGICAL_SRC_IMM_DIMS];
const fs_reg arg = inst->src[SURFACE_LOGICAL_SRC_IMM_ARG]; const brw_reg arg = inst->src[SURFACE_LOGICAL_SRC_IMM_ARG];
const fs_reg allow_sample_mask = const brw_reg allow_sample_mask =
inst->src[SURFACE_LOGICAL_SRC_ALLOW_SAMPLE_MASK]; inst->src[SURFACE_LOGICAL_SRC_ALLOW_SAMPLE_MASK];
assert(arg.file == IMM); assert(arg.file == IMM);
assert(allow_sample_mask.file == IMM); assert(allow_sample_mask.file == IMM);
@ -1685,7 +1685,7 @@ lower_lsc_surface_logical_send(const fs_builder &bld, fs_inst *inst)
unsigned num_components = 0; unsigned num_components = 0;
unsigned ex_mlen = 0; unsigned ex_mlen = 0;
fs_reg payload, payload2; brw_reg payload, payload2;
payload = bld.move_to_vgrf(addr, addr_sz); payload = bld.move_to_vgrf(addr, addr_sz);
if (src.file != BAD_FILE) { if (src.file != BAD_FILE) {
payload2 = bld.move_to_vgrf(src, src_comps); payload2 = bld.move_to_vgrf(src, src_comps);
@ -1693,8 +1693,8 @@ lower_lsc_surface_logical_send(const fs_builder &bld, fs_inst *inst)
} }
/* Predicate the instruction on the sample mask if needed */ /* Predicate the instruction on the sample mask if needed */
fs_reg sample_mask = allow_sample_mask.ud ? brw_sample_mask_reg(bld) : brw_reg sample_mask = allow_sample_mask.ud ? brw_sample_mask_reg(bld) :
fs_reg(brw_imm_ud(0xffffffff)); brw_reg(brw_imm_ud(0xffffffff));
if (sample_mask.file != BAD_FILE && sample_mask.file != IMM) if (sample_mask.file != BAD_FILE && sample_mask.file != IMM)
brw_emit_predicate_on_sample_mask(bld, inst); brw_emit_predicate_on_sample_mask(bld, inst);
@ -1831,11 +1831,11 @@ lower_lsc_block_logical_send(const fs_builder &bld, fs_inst *inst)
assert(devinfo->has_lsc); assert(devinfo->has_lsc);
/* Get the logical send arguments. */ /* Get the logical send arguments. */
const fs_reg addr = inst->src[SURFACE_LOGICAL_SRC_ADDRESS]; const brw_reg addr = inst->src[SURFACE_LOGICAL_SRC_ADDRESS];
const fs_reg src = inst->src[SURFACE_LOGICAL_SRC_DATA]; const brw_reg src = inst->src[SURFACE_LOGICAL_SRC_DATA];
const fs_reg surface = inst->src[SURFACE_LOGICAL_SRC_SURFACE]; const brw_reg surface = inst->src[SURFACE_LOGICAL_SRC_SURFACE];
const fs_reg surface_handle = inst->src[SURFACE_LOGICAL_SRC_SURFACE_HANDLE]; const brw_reg surface_handle = inst->src[SURFACE_LOGICAL_SRC_SURFACE_HANDLE];
const fs_reg arg = inst->src[SURFACE_LOGICAL_SRC_IMM_ARG]; const brw_reg arg = inst->src[SURFACE_LOGICAL_SRC_IMM_ARG];
assert(arg.file == IMM); assert(arg.file == IMM);
assert(inst->src[SURFACE_LOGICAL_SRC_IMM_DIMS].file == BAD_FILE); assert(inst->src[SURFACE_LOGICAL_SRC_IMM_DIMS].file == BAD_FILE);
assert(inst->src[SURFACE_LOGICAL_SRC_ALLOW_SAMPLE_MASK].file == BAD_FILE); assert(inst->src[SURFACE_LOGICAL_SRC_ALLOW_SAMPLE_MASK].file == BAD_FILE);
@ -1849,7 +1849,7 @@ lower_lsc_block_logical_send(const fs_builder &bld, fs_inst *inst)
const bool write = inst->opcode == SHADER_OPCODE_OWORD_BLOCK_WRITE_LOGICAL; const bool write = inst->opcode == SHADER_OPCODE_OWORD_BLOCK_WRITE_LOGICAL;
fs_builder ubld = bld.exec_all().group(1, 0); fs_builder ubld = bld.exec_all().group(1, 0);
fs_reg stateless_ex_desc; brw_reg stateless_ex_desc;
if (is_stateless) { if (is_stateless) {
stateless_ex_desc = ubld.vgrf(BRW_TYPE_UD); stateless_ex_desc = ubld.vgrf(BRW_TYPE_UD);
ubld.AND(stateless_ex_desc, ubld.AND(stateless_ex_desc,
@ -1859,7 +1859,7 @@ lower_lsc_block_logical_send(const fs_builder &bld, fs_inst *inst)
ubld.SHR(stateless_ex_desc, stateless_ex_desc, brw_imm_ud(4)); ubld.SHR(stateless_ex_desc, stateless_ex_desc, brw_imm_ud(4));
} }
fs_reg data; brw_reg data;
if (write) { if (write) {
const unsigned src_sz = inst->components_read(SURFACE_LOGICAL_SRC_DATA); const unsigned src_sz = inst->components_read(SURFACE_LOGICAL_SRC_DATA);
data = retype(bld.move_to_vgrf(src, src_sz), BRW_TYPE_UD); data = retype(bld.move_to_vgrf(src, src_sz), BRW_TYPE_UD);
@ -1913,11 +1913,11 @@ lower_surface_block_logical_send(const fs_builder &bld, fs_inst *inst)
const intel_device_info *devinfo = bld.shader->devinfo; const intel_device_info *devinfo = bld.shader->devinfo;
/* Get the logical send arguments. */ /* Get the logical send arguments. */
const fs_reg addr = inst->src[SURFACE_LOGICAL_SRC_ADDRESS]; const brw_reg addr = inst->src[SURFACE_LOGICAL_SRC_ADDRESS];
const fs_reg src = inst->src[SURFACE_LOGICAL_SRC_DATA]; const brw_reg src = inst->src[SURFACE_LOGICAL_SRC_DATA];
const fs_reg surface = inst->src[SURFACE_LOGICAL_SRC_SURFACE]; const brw_reg surface = inst->src[SURFACE_LOGICAL_SRC_SURFACE];
const fs_reg surface_handle = inst->src[SURFACE_LOGICAL_SRC_SURFACE_HANDLE]; const brw_reg surface_handle = inst->src[SURFACE_LOGICAL_SRC_SURFACE_HANDLE];
const fs_reg arg = inst->src[SURFACE_LOGICAL_SRC_IMM_ARG]; const brw_reg arg = inst->src[SURFACE_LOGICAL_SRC_IMM_ARG];
assert(arg.file == IMM); assert(arg.file == IMM);
assert(inst->src[SURFACE_LOGICAL_SRC_IMM_DIMS].file == BAD_FILE); assert(inst->src[SURFACE_LOGICAL_SRC_IMM_DIMS].file == BAD_FILE);
assert(inst->src[SURFACE_LOGICAL_SRC_ALLOW_SAMPLE_MASK].file == BAD_FILE); assert(inst->src[SURFACE_LOGICAL_SRC_ALLOW_SAMPLE_MASK].file == BAD_FILE);
@ -1941,7 +1941,7 @@ lower_surface_block_logical_send(const fs_builder &bld, fs_inst *inst)
/* The address is stored in the header. See MH_A32_GO and MH_BTS_GO. */ /* The address is stored in the header. See MH_A32_GO and MH_BTS_GO. */
fs_builder ubld = bld.exec_all().group(8, 0); fs_builder ubld = bld.exec_all().group(8, 0);
fs_reg header = ubld.vgrf(BRW_TYPE_UD); brw_reg header = ubld.vgrf(BRW_TYPE_UD);
if (is_stateless) if (is_stateless)
ubld.emit(SHADER_OPCODE_SCRATCH_HEADER, header); ubld.emit(SHADER_OPCODE_SCRATCH_HEADER, header);
@ -1954,7 +1954,7 @@ lower_surface_block_logical_send(const fs_builder &bld, fs_inst *inst)
else else
ubld.group(1, 0).MOV(component(header, 2), addr); ubld.group(1, 0).MOV(component(header, 2), addr);
fs_reg data; brw_reg data;
unsigned ex_mlen = 0; unsigned ex_mlen = 0;
if (write) { if (write) {
const unsigned src_sz = inst->components_read(SURFACE_LOGICAL_SRC_DATA); const unsigned src_sz = inst->components_read(SURFACE_LOGICAL_SRC_DATA);
@ -1981,14 +1981,14 @@ lower_surface_block_logical_send(const fs_builder &bld, fs_inst *inst)
inst->src[3] = data; inst->src[3] = data;
} }
static fs_reg static brw_reg
emit_a64_oword_block_header(const fs_builder &bld, const fs_reg &addr) emit_a64_oword_block_header(const fs_builder &bld, const brw_reg &addr)
{ {
const fs_builder ubld = bld.exec_all().group(8, 0); const fs_builder ubld = bld.exec_all().group(8, 0);
assert(brw_type_size_bytes(addr.type) == 8 && addr.stride == 0); assert(brw_type_size_bytes(addr.type) == 8 && addr.stride == 0);
fs_reg expanded_addr = addr; brw_reg expanded_addr = addr;
if (addr.file == UNIFORM) { if (addr.file == UNIFORM) {
/* We can't do stride 1 with the UNIFORM file, it requires stride 0 */ /* We can't do stride 1 with the UNIFORM file, it requires stride 0 */
expanded_addr = ubld.vgrf(BRW_TYPE_UQ); expanded_addr = ubld.vgrf(BRW_TYPE_UQ);
@ -1996,11 +1996,11 @@ emit_a64_oword_block_header(const fs_builder &bld, const fs_reg &addr)
ubld.MOV(expanded_addr, retype(addr, BRW_TYPE_UQ)); ubld.MOV(expanded_addr, retype(addr, BRW_TYPE_UQ));
} }
fs_reg header = ubld.vgrf(BRW_TYPE_UD); brw_reg header = ubld.vgrf(BRW_TYPE_UD);
ubld.MOV(header, brw_imm_ud(0)); ubld.MOV(header, brw_imm_ud(0));
/* Use a 2-wide MOV to fill out the address */ /* Use a 2-wide MOV to fill out the address */
fs_reg addr_vec2 = expanded_addr; brw_reg addr_vec2 = expanded_addr;
addr_vec2.type = BRW_TYPE_UD; addr_vec2.type = BRW_TYPE_UD;
addr_vec2.stride = 1; addr_vec2.stride = 1;
ubld.group(2, 0).MOV(header, addr_vec2); ubld.group(2, 0).MOV(header, addr_vec2);
@ -2034,8 +2034,8 @@ lower_lsc_a64_logical_send(const fs_builder &bld, fs_inst *inst)
const intel_device_info *devinfo = bld.shader->devinfo; const intel_device_info *devinfo = bld.shader->devinfo;
/* Get the logical send arguments. */ /* Get the logical send arguments. */
const fs_reg addr = inst->src[A64_LOGICAL_ADDRESS]; const brw_reg addr = inst->src[A64_LOGICAL_ADDRESS];
const fs_reg src = inst->src[A64_LOGICAL_SRC]; const brw_reg src = inst->src[A64_LOGICAL_SRC];
const unsigned src_sz = brw_type_size_bytes(src.type); const unsigned src_sz = brw_type_size_bytes(src.type);
const unsigned dst_sz = brw_type_size_bytes(inst->dst.type); const unsigned dst_sz = brw_type_size_bytes(inst->dst.type);
@ -2044,8 +2044,8 @@ lower_lsc_a64_logical_send(const fs_builder &bld, fs_inst *inst)
const unsigned arg = inst->src[A64_LOGICAL_ARG].ud; const unsigned arg = inst->src[A64_LOGICAL_ARG].ud;
const bool has_side_effects = inst->has_side_effects(); const bool has_side_effects = inst->has_side_effects();
fs_reg payload = retype(bld.move_to_vgrf(addr, 1), BRW_TYPE_UD); brw_reg payload = retype(bld.move_to_vgrf(addr, 1), BRW_TYPE_UD);
fs_reg payload2 = retype(bld.move_to_vgrf(src, src_comps), BRW_TYPE_UD); brw_reg payload2 = retype(bld.move_to_vgrf(src, src_comps), BRW_TYPE_UD);
unsigned ex_mlen = src_comps * src_sz * inst->exec_size / REG_SIZE; unsigned ex_mlen = src_comps * src_sz * inst->exec_size / REG_SIZE;
unsigned num_components = 0; unsigned num_components = 0;
@ -2155,14 +2155,14 @@ lower_a64_logical_send(const fs_builder &bld, fs_inst *inst)
{ {
const intel_device_info *devinfo = bld.shader->devinfo; const intel_device_info *devinfo = bld.shader->devinfo;
const fs_reg addr = inst->src[A64_LOGICAL_ADDRESS]; const brw_reg addr = inst->src[A64_LOGICAL_ADDRESS];
const fs_reg src = inst->src[A64_LOGICAL_SRC]; const brw_reg src = inst->src[A64_LOGICAL_SRC];
const unsigned src_comps = inst->components_read(1); const unsigned src_comps = inst->components_read(1);
assert(inst->src[A64_LOGICAL_ARG].file == IMM); assert(inst->src[A64_LOGICAL_ARG].file == IMM);
const unsigned arg = inst->src[A64_LOGICAL_ARG].ud; const unsigned arg = inst->src[A64_LOGICAL_ARG].ud;
const bool has_side_effects = inst->has_side_effects(); const bool has_side_effects = inst->has_side_effects();
fs_reg payload, payload2; brw_reg payload, payload2;
unsigned mlen, ex_mlen = 0, header_size = 0; unsigned mlen, ex_mlen = 0, header_size = 0;
if (inst->opcode == SHADER_OPCODE_A64_OWORD_BLOCK_READ_LOGICAL || if (inst->opcode == SHADER_OPCODE_A64_OWORD_BLOCK_READ_LOGICAL ||
inst->opcode == SHADER_OPCODE_A64_OWORD_BLOCK_WRITE_LOGICAL || inst->opcode == SHADER_OPCODE_A64_OWORD_BLOCK_WRITE_LOGICAL ||
@ -2279,16 +2279,16 @@ lower_lsc_varying_pull_constant_logical_send(const fs_builder &bld,
const intel_device_info *devinfo = bld.shader->devinfo; const intel_device_info *devinfo = bld.shader->devinfo;
ASSERTED const brw_compiler *compiler = bld.shader->compiler; ASSERTED const brw_compiler *compiler = bld.shader->compiler;
fs_reg surface = inst->src[PULL_VARYING_CONSTANT_SRC_SURFACE]; brw_reg surface = inst->src[PULL_VARYING_CONSTANT_SRC_SURFACE];
fs_reg surface_handle = inst->src[PULL_VARYING_CONSTANT_SRC_SURFACE_HANDLE]; brw_reg surface_handle = inst->src[PULL_VARYING_CONSTANT_SRC_SURFACE_HANDLE];
fs_reg offset_B = inst->src[PULL_VARYING_CONSTANT_SRC_OFFSET]; brw_reg offset_B = inst->src[PULL_VARYING_CONSTANT_SRC_OFFSET];
fs_reg alignment_B = inst->src[PULL_VARYING_CONSTANT_SRC_ALIGNMENT]; brw_reg alignment_B = inst->src[PULL_VARYING_CONSTANT_SRC_ALIGNMENT];
/* We are switching the instruction from an ALU-like instruction to a /* We are switching the instruction from an ALU-like instruction to a
* send-from-grf instruction. Since sends can't handle strides or * send-from-grf instruction. Since sends can't handle strides or
* source modifiers, we have to make a copy of the offset source. * source modifiers, we have to make a copy of the offset source.
*/ */
fs_reg ubo_offset = bld.move_to_vgrf(offset_B, 1); brw_reg ubo_offset = bld.move_to_vgrf(offset_B, 1);
enum lsc_addr_surface_type surf_type = enum lsc_addr_surface_type surf_type =
surface_handle.file == BAD_FILE ? surface_handle.file == BAD_FILE ?
@ -2365,15 +2365,15 @@ lower_varying_pull_constant_logical_send(const fs_builder &bld, fs_inst *inst)
const intel_device_info *devinfo = bld.shader->devinfo; const intel_device_info *devinfo = bld.shader->devinfo;
const brw_compiler *compiler = bld.shader->compiler; const brw_compiler *compiler = bld.shader->compiler;
fs_reg surface = inst->src[PULL_VARYING_CONSTANT_SRC_SURFACE]; brw_reg surface = inst->src[PULL_VARYING_CONSTANT_SRC_SURFACE];
fs_reg surface_handle = inst->src[PULL_VARYING_CONSTANT_SRC_SURFACE_HANDLE]; brw_reg surface_handle = inst->src[PULL_VARYING_CONSTANT_SRC_SURFACE_HANDLE];
fs_reg offset_B = inst->src[PULL_VARYING_CONSTANT_SRC_OFFSET]; brw_reg offset_B = inst->src[PULL_VARYING_CONSTANT_SRC_OFFSET];
/* We are switching the instruction from an ALU-like instruction to a /* We are switching the instruction from an ALU-like instruction to a
* send-from-grf instruction. Since sends can't handle strides or * send-from-grf instruction. Since sends can't handle strides or
* source modifiers, we have to make a copy of the offset source. * source modifiers, we have to make a copy of the offset source.
*/ */
fs_reg ubo_offset = bld.vgrf(BRW_TYPE_UD); brw_reg ubo_offset = bld.vgrf(BRW_TYPE_UD);
bld.MOV(ubo_offset, offset_B); bld.MOV(ubo_offset, offset_B);
assert(inst->src[PULL_VARYING_CONSTANT_SRC_ALIGNMENT].file == BRW_IMMEDIATE_VALUE); assert(inst->src[PULL_VARYING_CONSTANT_SRC_ALIGNMENT].file == BRW_IMMEDIATE_VALUE);
@ -2445,7 +2445,7 @@ lower_interpolator_logical_send(const fs_builder &bld, fs_inst *inst,
const intel_device_info *devinfo = bld.shader->devinfo; const intel_device_info *devinfo = bld.shader->devinfo;
/* We have to send something */ /* We have to send something */
fs_reg payload = brw_vec8_grf(0, 0); brw_reg payload = brw_vec8_grf(0, 0);
unsigned mlen = 1; unsigned mlen = 1;
unsigned mode; unsigned mode;
@ -2473,7 +2473,7 @@ lower_interpolator_logical_send(const fs_builder &bld, fs_inst *inst,
const bool dynamic_mode = const bool dynamic_mode =
inst->src[INTERP_SRC_DYNAMIC_MODE].file != BAD_FILE; inst->src[INTERP_SRC_DYNAMIC_MODE].file != BAD_FILE;
fs_reg desc = inst->src[INTERP_SRC_MSG_DESC]; brw_reg desc = inst->src[INTERP_SRC_MSG_DESC];
uint32_t desc_imm = uint32_t desc_imm =
brw_pixel_interp_desc(devinfo, brw_pixel_interp_desc(devinfo,
/* Leave the mode at 0 if persample_dispatch is /* Leave the mode at 0 if persample_dispatch is
@ -2488,7 +2488,7 @@ lower_interpolator_logical_send(const fs_builder &bld, fs_inst *inst,
desc_imm |= (1 << 15); desc_imm |= (1 << 15);
} else if (wm_prog_data->coarse_pixel_dispatch == BRW_SOMETIMES) { } else if (wm_prog_data->coarse_pixel_dispatch == BRW_SOMETIMES) {
STATIC_ASSERT(INTEL_MSAA_FLAG_COARSE_PI_MSG == (1 << 15)); STATIC_ASSERT(INTEL_MSAA_FLAG_COARSE_PI_MSG == (1 << 15));
fs_reg orig_desc = desc; brw_reg orig_desc = desc;
const fs_builder &ubld = bld.exec_all().group(8, 0); const fs_builder &ubld = bld.exec_all().group(8, 0);
desc = ubld.vgrf(BRW_TYPE_UD); desc = ubld.vgrf(BRW_TYPE_UD);
ubld.AND(desc, dynamic_msaa_flags(wm_prog_data), ubld.AND(desc, dynamic_msaa_flags(wm_prog_data),
@ -2518,7 +2518,7 @@ lower_interpolator_logical_send(const fs_builder &bld, fs_inst *inst,
* components of "Per Message Offset”, which will give us the pixel offset 0x0. * components of "Per Message Offset”, which will give us the pixel offset 0x0.
*/ */
if (dynamic_mode) { if (dynamic_mode) {
fs_reg orig_desc = desc; brw_reg orig_desc = desc;
const fs_builder &ubld = bld.exec_all().group(8, 0); const fs_builder &ubld = bld.exec_all().group(8, 0);
desc = ubld.vgrf(BRW_TYPE_UD); desc = ubld.vgrf(BRW_TYPE_UD);
@ -2565,13 +2565,13 @@ static void
lower_btd_logical_send(const fs_builder &bld, fs_inst *inst) lower_btd_logical_send(const fs_builder &bld, fs_inst *inst)
{ {
const intel_device_info *devinfo = bld.shader->devinfo; const intel_device_info *devinfo = bld.shader->devinfo;
fs_reg global_addr = inst->src[0]; brw_reg global_addr = inst->src[0];
const fs_reg btd_record = inst->src[1]; const brw_reg btd_record = inst->src[1];
const unsigned unit = reg_unit(devinfo); const unsigned unit = reg_unit(devinfo);
const unsigned mlen = 2 * unit; const unsigned mlen = 2 * unit;
const fs_builder ubld = bld.exec_all(); const fs_builder ubld = bld.exec_all();
fs_reg header = ubld.vgrf(BRW_TYPE_UD, 2 * unit); brw_reg header = ubld.vgrf(BRW_TYPE_UD, 2 * unit);
ubld.MOV(header, brw_imm_ud(0)); ubld.MOV(header, brw_imm_ud(0));
switch (inst->opcode) { switch (inst->opcode) {
@ -2595,12 +2595,12 @@ lower_btd_logical_send(const fs_builder &bld, fs_inst *inst)
/* Stack IDs are always in R1 regardless of whether we're coming from a /* Stack IDs are always in R1 regardless of whether we're coming from a
* bindless shader or a regular compute shader. * bindless shader or a regular compute shader.
*/ */
fs_reg stack_ids = retype(offset(header, bld, 1), BRW_TYPE_UW); brw_reg stack_ids = retype(offset(header, bld, 1), BRW_TYPE_UW);
bld.exec_all().MOV(stack_ids, retype(brw_vec8_grf(1 * unit, 0), bld.exec_all().MOV(stack_ids, retype(brw_vec8_grf(1 * unit, 0),
BRW_TYPE_UW)); BRW_TYPE_UW));
unsigned ex_mlen = 0; unsigned ex_mlen = 0;
fs_reg payload; brw_reg payload;
if (inst->opcode == SHADER_OPCODE_BTD_SPAWN_LOGICAL) { if (inst->opcode == SHADER_OPCODE_BTD_SPAWN_LOGICAL) {
ex_mlen = 2 * (inst->exec_size / 8); ex_mlen = 2 * (inst->exec_size / 8);
payload = bld.move_to_vgrf(btd_record, 1); payload = bld.move_to_vgrf(btd_record, 1);
@ -2643,33 +2643,33 @@ lower_trace_ray_logical_send(const fs_builder &bld, fs_inst *inst)
* so that the MOV operates on 2 components rather than twice the same * so that the MOV operates on 2 components rather than twice the same
* component. * component.
*/ */
fs_reg globals_addr = retype(inst->src[RT_LOGICAL_SRC_GLOBALS], BRW_TYPE_UD); brw_reg globals_addr = retype(inst->src[RT_LOGICAL_SRC_GLOBALS], BRW_TYPE_UD);
globals_addr.stride = 1; globals_addr.stride = 1;
const fs_reg bvh_level = const brw_reg bvh_level =
inst->src[RT_LOGICAL_SRC_BVH_LEVEL].file == BRW_IMMEDIATE_VALUE ? inst->src[RT_LOGICAL_SRC_BVH_LEVEL].file == BRW_IMMEDIATE_VALUE ?
inst->src[RT_LOGICAL_SRC_BVH_LEVEL] : inst->src[RT_LOGICAL_SRC_BVH_LEVEL] :
bld.move_to_vgrf(inst->src[RT_LOGICAL_SRC_BVH_LEVEL], bld.move_to_vgrf(inst->src[RT_LOGICAL_SRC_BVH_LEVEL],
inst->components_read(RT_LOGICAL_SRC_BVH_LEVEL)); inst->components_read(RT_LOGICAL_SRC_BVH_LEVEL));
const fs_reg trace_ray_control = const brw_reg trace_ray_control =
inst->src[RT_LOGICAL_SRC_TRACE_RAY_CONTROL].file == BRW_IMMEDIATE_VALUE ? inst->src[RT_LOGICAL_SRC_TRACE_RAY_CONTROL].file == BRW_IMMEDIATE_VALUE ?
inst->src[RT_LOGICAL_SRC_TRACE_RAY_CONTROL] : inst->src[RT_LOGICAL_SRC_TRACE_RAY_CONTROL] :
bld.move_to_vgrf(inst->src[RT_LOGICAL_SRC_TRACE_RAY_CONTROL], bld.move_to_vgrf(inst->src[RT_LOGICAL_SRC_TRACE_RAY_CONTROL],
inst->components_read(RT_LOGICAL_SRC_TRACE_RAY_CONTROL)); inst->components_read(RT_LOGICAL_SRC_TRACE_RAY_CONTROL));
const fs_reg synchronous_src = inst->src[RT_LOGICAL_SRC_SYNCHRONOUS]; const brw_reg synchronous_src = inst->src[RT_LOGICAL_SRC_SYNCHRONOUS];
assert(synchronous_src.file == BRW_IMMEDIATE_VALUE); assert(synchronous_src.file == BRW_IMMEDIATE_VALUE);
const bool synchronous = synchronous_src.ud; const bool synchronous = synchronous_src.ud;
const unsigned unit = reg_unit(devinfo); const unsigned unit = reg_unit(devinfo);
const unsigned mlen = unit; const unsigned mlen = unit;
const fs_builder ubld = bld.exec_all(); const fs_builder ubld = bld.exec_all();
fs_reg header = ubld.vgrf(BRW_TYPE_UD); brw_reg header = ubld.vgrf(BRW_TYPE_UD);
ubld.MOV(header, brw_imm_ud(0)); ubld.MOV(header, brw_imm_ud(0));
ubld.group(2, 0).MOV(header, globals_addr); ubld.group(2, 0).MOV(header, globals_addr);
if (synchronous) if (synchronous)
ubld.group(1, 0).MOV(byte_offset(header, 16), brw_imm_ud(synchronous)); ubld.group(1, 0).MOV(byte_offset(header, 16), brw_imm_ud(synchronous));
const unsigned ex_mlen = inst->exec_size / 8; const unsigned ex_mlen = inst->exec_size / 8;
fs_reg payload = bld.vgrf(BRW_TYPE_UD); brw_reg payload = bld.vgrf(BRW_TYPE_UD);
if (bvh_level.file == BRW_IMMEDIATE_VALUE && if (bvh_level.file == BRW_IMMEDIATE_VALUE &&
trace_ray_control.file == BRW_IMMEDIATE_VALUE) { trace_ray_control.file == BRW_IMMEDIATE_VALUE) {
bld.MOV(payload, brw_imm_ud(SET_BITS(trace_ray_control.ud, 9, 8) | bld.MOV(payload, brw_imm_ud(SET_BITS(trace_ray_control.ud, 9, 8) |
@ -2720,9 +2720,9 @@ lower_get_buffer_size(const fs_builder &bld, fs_inst *inst)
*/ */
assert(inst->exec_size == (devinfo->ver < 20 ? 8 : 16)); assert(inst->exec_size == (devinfo->ver < 20 ? 8 : 16));
fs_reg surface = inst->src[GET_BUFFER_SIZE_SRC_SURFACE]; brw_reg surface = inst->src[GET_BUFFER_SIZE_SRC_SURFACE];
fs_reg surface_handle = inst->src[GET_BUFFER_SIZE_SRC_SURFACE_HANDLE]; brw_reg surface_handle = inst->src[GET_BUFFER_SIZE_SRC_SURFACE_HANDLE];
fs_reg lod = inst->src[GET_BUFFER_SIZE_SRC_LOD]; brw_reg lod = inst->src[GET_BUFFER_SIZE_SRC_LOD];
inst->opcode = SHADER_OPCODE_SEND; inst->opcode = SHADER_OPCODE_SEND;
inst->mlen = inst->exec_size / 8; inst->mlen = inst->exec_size / 8;
@ -2916,10 +2916,10 @@ brw_fs_lower_uniform_pull_constant_loads(fs_visitor &s)
if (inst->opcode != FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD) if (inst->opcode != FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD)
continue; continue;
const fs_reg surface = inst->src[PULL_UNIFORM_CONSTANT_SRC_SURFACE]; const brw_reg surface = inst->src[PULL_UNIFORM_CONSTANT_SRC_SURFACE];
const fs_reg surface_handle = inst->src[PULL_UNIFORM_CONSTANT_SRC_SURFACE_HANDLE]; const brw_reg surface_handle = inst->src[PULL_UNIFORM_CONSTANT_SRC_SURFACE_HANDLE];
const fs_reg offset_B = inst->src[PULL_UNIFORM_CONSTANT_SRC_OFFSET]; const brw_reg offset_B = inst->src[PULL_UNIFORM_CONSTANT_SRC_OFFSET];
const fs_reg size_B = inst->src[PULL_UNIFORM_CONSTANT_SRC_SIZE]; const brw_reg size_B = inst->src[PULL_UNIFORM_CONSTANT_SRC_SIZE];
assert(surface.file == BAD_FILE || surface_handle.file == BAD_FILE); assert(surface.file == BAD_FILE || surface_handle.file == BAD_FILE);
assert(offset_B.file == IMM); assert(offset_B.file == IMM);
assert(size_B.file == IMM); assert(size_B.file == IMM);
@ -2928,7 +2928,7 @@ brw_fs_lower_uniform_pull_constant_loads(fs_visitor &s)
const fs_builder ubld = const fs_builder ubld =
fs_builder(&s, block, inst).group(8, 0).exec_all(); fs_builder(&s, block, inst).group(8, 0).exec_all();
const fs_reg payload = ubld.vgrf(BRW_TYPE_UD); const brw_reg payload = ubld.vgrf(BRW_TYPE_UD);
ubld.MOV(payload, offset_B); ubld.MOV(payload, offset_B);
inst->sfid = GFX12_SFID_UGM; inst->sfid = GFX12_SFID_UGM;
@ -2964,7 +2964,7 @@ brw_fs_lower_uniform_pull_constant_loads(fs_visitor &s)
s.invalidate_analysis(DEPENDENCY_INSTRUCTIONS | DEPENDENCY_VARIABLES); s.invalidate_analysis(DEPENDENCY_INSTRUCTIONS | DEPENDENCY_VARIABLES);
} else { } else {
const fs_builder ubld = fs_builder(&s, block, inst).exec_all(); const fs_builder ubld = fs_builder(&s, block, inst).exec_all();
fs_reg header = fs_builder(&s, 8).exec_all().vgrf(BRW_TYPE_UD); brw_reg header = fs_builder(&s, 8).exec_all().vgrf(BRW_TYPE_UD);
ubld.group(8, 0).MOV(header, ubld.group(8, 0).MOV(header,
retype(brw_vec8_grf(0, 0), BRW_TYPE_UD)); retype(brw_vec8_grf(0, 0), BRW_TYPE_UD));
@ -2985,7 +2985,7 @@ brw_fs_lower_uniform_pull_constant_loads(fs_visitor &s)
setup_surface_descriptors(ubld, inst, desc, surface, surface_handle); setup_surface_descriptors(ubld, inst, desc, surface, surface_handle);
inst->src[2] = header; inst->src[2] = header;
inst->src[3] = fs_reg(); /* unused for reads */ inst->src[3] = brw_reg(); /* unused for reads */
s.invalidate_analysis(DEPENDENCY_INSTRUCTIONS | DEPENDENCY_VARIABLES); s.invalidate_analysis(DEPENDENCY_INSTRUCTIONS | DEPENDENCY_VARIABLES);
} }

8
src/intel/compiler/brw_schedule_instructions.cpp
View file

@ -589,7 +589,7 @@ public:
void calculate_deps(); void calculate_deps();
bool is_compressed(const fs_inst *inst); bool is_compressed(const fs_inst *inst);
bool register_needs_barrier(const fs_reg &reg); bool register_needs_barrier(const brw_reg &reg);
schedule_node *choose_instruction_to_schedule(); schedule_node *choose_instruction_to_schedule();
int calculate_issue_time(const fs_inst *inst); int calculate_issue_time(const fs_inst *inst);
@ -602,7 +602,7 @@ public:
void schedule_instructions(); void schedule_instructions();
void run(instruction_scheduler_mode mode); void run(instruction_scheduler_mode mode);
int grf_index(const fs_reg &reg); int grf_index(const brw_reg &reg);
void *mem_ctx; void *mem_ctx;
linear_ctx *lin_ctx; linear_ctx *lin_ctx;
@ -1071,7 +1071,7 @@ has_cross_lane_access(const fs_inst *inst)
* Some register access need dependencies on other instructions. * Some register access need dependencies on other instructions.
*/ */
bool bool
instruction_scheduler::register_needs_barrier(const fs_reg &reg) instruction_scheduler::register_needs_barrier(const brw_reg &reg)
{ {
if (reg.file != ARF || reg.is_null()) if (reg.file != ARF || reg.is_null())
return false; return false;
@ -1175,7 +1175,7 @@ instruction_scheduler::clear_last_grf_write()
} }
int int
instruction_scheduler::grf_index(const fs_reg &reg) instruction_scheduler::grf_index(const brw_reg &reg)
{ {
if (post_reg_alloc) if (post_reg_alloc)
return reg.nr; return reg.nr;

8
src/intel/compiler/brw_shader.cpp
View file

@ -30,7 +30,7 @@
#include "util/macros.h" #include "util/macros.h"
bool bool
fs_reg_saturate_immediate(fs_reg *reg) fs_reg_saturate_immediate(brw_reg *reg)
{ {
union { union {
unsigned ud; unsigned ud;
@ -93,7 +93,7 @@ fs_reg_saturate_immediate(fs_reg *reg)
} }
bool bool
fs_reg_negate_immediate(fs_reg *reg) fs_reg_negate_immediate(brw_reg *reg)
{ {
switch (reg->type) { switch (reg->type) {
case BRW_TYPE_D: case BRW_TYPE_D:
@ -136,7 +136,7 @@ fs_reg_negate_immediate(fs_reg *reg)
} }
bool bool
fs_reg_abs_immediate(fs_reg *reg) fs_reg_abs_immediate(brw_reg *reg)
{ {
switch (reg->type) { switch (reg->type) {
case BRW_TYPE_D: case BRW_TYPE_D:
@ -579,7 +579,7 @@ fs_inst::remove(bblock_t *block, bool defer_later_block_ip_updates)
if (exec_list_is_singular(&block->instructions)) { if (exec_list_is_singular(&block->instructions)) {
this->opcode = BRW_OPCODE_NOP; this->opcode = BRW_OPCODE_NOP;
this->resize_sources(0); this->resize_sources(0);
this->dst = fs_reg(); this->dst = brw_reg();
this->size_written = 0; this->size_written = 0;
return; return;
} }

460
src/intel/compiler/test_fs_cmod_propagation.cpp
View file

@ -130,10 +130,10 @@ cmod_propagation(fs_visitor *v)
TEST_F(cmod_propagation_test, basic) TEST_F(cmod_propagation_test, basic)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_F); brw_reg dest = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
fs_reg zero(brw_imm_f(0.0f)); brw_reg zero(brw_imm_f(0.0f));
bld.ADD(dest, src0, src1); bld.ADD(dest, src0, src1);
bld.CMP(bld.null_reg_f(), dest, zero, BRW_CONDITIONAL_GE); bld.CMP(bld.null_reg_f(), dest, zero, BRW_CONDITIONAL_GE);
@ -161,10 +161,10 @@ TEST_F(cmod_propagation_test, basic)
TEST_F(cmod_propagation_test, basic_other_flag) TEST_F(cmod_propagation_test, basic_other_flag)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_F); brw_reg dest = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
fs_reg zero(brw_imm_f(0.0f)); brw_reg zero(brw_imm_f(0.0f));
bld.ADD(dest, src0, src1); bld.ADD(dest, src0, src1);
bld.CMP(bld.null_reg_f(), dest, zero, BRW_CONDITIONAL_GE) bld.CMP(bld.null_reg_f(), dest, zero, BRW_CONDITIONAL_GE)
->flag_subreg = 1; ->flag_subreg = 1;
@ -194,10 +194,10 @@ TEST_F(cmod_propagation_test, basic_other_flag)
TEST_F(cmod_propagation_test, cmp_nonzero) TEST_F(cmod_propagation_test, cmp_nonzero)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_F); brw_reg dest = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
fs_reg nonzero(brw_imm_f(1.0f)); brw_reg nonzero(brw_imm_f(1.0f));
bld.ADD(dest, src0, src1); bld.ADD(dest, src0, src1);
bld.CMP(bld.null_reg_f(), dest, nonzero, BRW_CONDITIONAL_GE); bld.CMP(bld.null_reg_f(), dest, nonzero, BRW_CONDITIONAL_GE);
@ -226,9 +226,9 @@ TEST_F(cmod_propagation_test, cmp_nonzero)
TEST_F(cmod_propagation_test, non_cmod_instruction) TEST_F(cmod_propagation_test, non_cmod_instruction)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_UD); brw_reg dest = bld.vgrf(BRW_TYPE_UD);
fs_reg src0 = bld.vgrf(BRW_TYPE_UD); brw_reg src0 = bld.vgrf(BRW_TYPE_UD);
fs_reg zero(brw_imm_ud(0u)); brw_reg zero(brw_imm_ud(0u));
bld.FBL(dest, src0); bld.FBL(dest, src0);
bld.CMP(bld.null_reg_ud(), dest, zero, BRW_CONDITIONAL_GE); bld.CMP(bld.null_reg_ud(), dest, zero, BRW_CONDITIONAL_GE);
@ -257,8 +257,8 @@ TEST_F(cmod_propagation_test, non_cmod_instruction)
TEST_F(cmod_propagation_test, non_cmod_livechannel) TEST_F(cmod_propagation_test, non_cmod_livechannel)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_UD); brw_reg dest = bld.vgrf(BRW_TYPE_UD);
fs_reg zero(brw_imm_d(0)); brw_reg zero(brw_imm_d(0));
bld.emit(SHADER_OPCODE_FIND_LIVE_CHANNEL, dest)->exec_size = 32; bld.emit(SHADER_OPCODE_FIND_LIVE_CHANNEL, dest)->exec_size = 32;
bld.CMP(bld.null_reg_d(), dest, zero, BRW_CONDITIONAL_Z)->exec_size = 32; bld.CMP(bld.null_reg_d(), dest, zero, BRW_CONDITIONAL_Z)->exec_size = 32;
@ -288,11 +288,11 @@ TEST_F(cmod_propagation_test, non_cmod_livechannel)
TEST_F(cmod_propagation_test, intervening_flag_write) TEST_F(cmod_propagation_test, intervening_flag_write)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_F); brw_reg dest = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
fs_reg src2 = bld.vgrf(BRW_TYPE_F); brw_reg src2 = bld.vgrf(BRW_TYPE_F);
fs_reg zero(brw_imm_f(0.0f)); brw_reg zero(brw_imm_f(0.0f));
bld.ADD(dest, src0, src1); bld.ADD(dest, src0, src1);
bld.CMP(bld.null_reg_f(), src2, zero, BRW_CONDITIONAL_GE); bld.CMP(bld.null_reg_f(), src2, zero, BRW_CONDITIONAL_GE);
bld.CMP(bld.null_reg_f(), dest, zero, BRW_CONDITIONAL_GE); bld.CMP(bld.null_reg_f(), dest, zero, BRW_CONDITIONAL_GE);
@ -325,11 +325,11 @@ TEST_F(cmod_propagation_test, intervening_flag_write)
TEST_F(cmod_propagation_test, intervening_mismatch_flag_write) TEST_F(cmod_propagation_test, intervening_mismatch_flag_write)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_F); brw_reg dest = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
fs_reg src2 = bld.vgrf(BRW_TYPE_F); brw_reg src2 = bld.vgrf(BRW_TYPE_F);
fs_reg zero(brw_imm_f(0.0f)); brw_reg zero(brw_imm_f(0.0f));
bld.ADD(dest, src0, src1); bld.ADD(dest, src0, src1);
bld.CMP(bld.null_reg_f(), src2, zero, BRW_CONDITIONAL_GE) bld.CMP(bld.null_reg_f(), src2, zero, BRW_CONDITIONAL_GE)
->flag_subreg = 1; ->flag_subreg = 1;
@ -365,12 +365,12 @@ TEST_F(cmod_propagation_test, intervening_mismatch_flag_write)
TEST_F(cmod_propagation_test, intervening_flag_read) TEST_F(cmod_propagation_test, intervening_flag_read)
{ {
fs_reg dest0 = bld.vgrf(BRW_TYPE_F); brw_reg dest0 = bld.vgrf(BRW_TYPE_F);
fs_reg dest1 = bld.vgrf(BRW_TYPE_F); brw_reg dest1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
fs_reg src2 = bld.vgrf(BRW_TYPE_F); brw_reg src2 = bld.vgrf(BRW_TYPE_F);
fs_reg zero(brw_imm_f(0.0f)); brw_reg zero(brw_imm_f(0.0f));
bld.ADD(dest0, src0, src1); bld.ADD(dest0, src0, src1);
set_predicate(BRW_PREDICATE_NORMAL, bld.SEL(dest1, src2, zero)); set_predicate(BRW_PREDICATE_NORMAL, bld.SEL(dest1, src2, zero));
bld.CMP(bld.null_reg_f(), dest0, zero, BRW_CONDITIONAL_GE); bld.CMP(bld.null_reg_f(), dest0, zero, BRW_CONDITIONAL_GE);
@ -403,12 +403,12 @@ TEST_F(cmod_propagation_test, intervening_flag_read)
TEST_F(cmod_propagation_test, intervening_mismatch_flag_read) TEST_F(cmod_propagation_test, intervening_mismatch_flag_read)
{ {
fs_reg dest0 = bld.vgrf(BRW_TYPE_F); brw_reg dest0 = bld.vgrf(BRW_TYPE_F);
fs_reg dest1 = bld.vgrf(BRW_TYPE_F); brw_reg dest1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
fs_reg src2 = bld.vgrf(BRW_TYPE_F); brw_reg src2 = bld.vgrf(BRW_TYPE_F);
fs_reg zero(brw_imm_f(0.0f)); brw_reg zero(brw_imm_f(0.0f));
bld.ADD(dest0, src0, src1); bld.ADD(dest0, src0, src1);
set_predicate(BRW_PREDICATE_NORMAL, bld.SEL(dest1, src2, zero)) set_predicate(BRW_PREDICATE_NORMAL, bld.SEL(dest1, src2, zero))
->flag_subreg = 1; ->flag_subreg = 1;
@ -444,13 +444,13 @@ TEST_F(cmod_propagation_test, intervening_mismatch_flag_read)
TEST_F(cmod_propagation_test, intervening_dest_write) TEST_F(cmod_propagation_test, intervening_dest_write)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_F, 4); brw_reg dest = bld.vgrf(BRW_TYPE_F, 4);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
fs_reg src2 = bld.vgrf(BRW_TYPE_F, 2); brw_reg src2 = bld.vgrf(BRW_TYPE_F, 2);
fs_reg zero(brw_imm_f(0.0f)); brw_reg zero(brw_imm_f(0.0f));
fs_reg tex_srcs[TEX_LOGICAL_NUM_SRCS]; brw_reg tex_srcs[TEX_LOGICAL_NUM_SRCS];
tex_srcs[TEX_LOGICAL_SRC_COORDINATE] = src2; tex_srcs[TEX_LOGICAL_SRC_COORDINATE] = src2;
tex_srcs[TEX_LOGICAL_SRC_SURFACE] = brw_imm_ud(0); tex_srcs[TEX_LOGICAL_SRC_SURFACE] = brw_imm_ud(0);
tex_srcs[TEX_LOGICAL_SRC_COORD_COMPONENTS] = brw_imm_ud(2); tex_srcs[TEX_LOGICAL_SRC_COORD_COMPONENTS] = brw_imm_ud(2);
@ -491,12 +491,12 @@ TEST_F(cmod_propagation_test, intervening_dest_write)
TEST_F(cmod_propagation_test, intervening_flag_read_same_value) TEST_F(cmod_propagation_test, intervening_flag_read_same_value)
{ {
fs_reg dest0 = bld.vgrf(BRW_TYPE_F); brw_reg dest0 = bld.vgrf(BRW_TYPE_F);
fs_reg dest1 = bld.vgrf(BRW_TYPE_F); brw_reg dest1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
fs_reg src2 = bld.vgrf(BRW_TYPE_F); brw_reg src2 = bld.vgrf(BRW_TYPE_F);
fs_reg zero(brw_imm_f(0.0f)); brw_reg zero(brw_imm_f(0.0f));
set_condmod(BRW_CONDITIONAL_GE, bld.ADD(dest0, src0, src1)); set_condmod(BRW_CONDITIONAL_GE, bld.ADD(dest0, src0, src1));
set_predicate(BRW_PREDICATE_NORMAL, bld.SEL(dest1, src2, zero)); set_predicate(BRW_PREDICATE_NORMAL, bld.SEL(dest1, src2, zero));
bld.CMP(bld.null_reg_f(), dest0, zero, BRW_CONDITIONAL_GE); bld.CMP(bld.null_reg_f(), dest0, zero, BRW_CONDITIONAL_GE);
@ -529,10 +529,10 @@ TEST_F(cmod_propagation_test, intervening_flag_read_same_value)
TEST_F(cmod_propagation_test, negate) TEST_F(cmod_propagation_test, negate)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_F); brw_reg dest = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
fs_reg zero(brw_imm_f(0.0f)); brw_reg zero(brw_imm_f(0.0f));
bld.ADD(dest, src0, src1); bld.ADD(dest, src0, src1);
dest.negate = true; dest.negate = true;
bld.CMP(bld.null_reg_f(), dest, zero, BRW_CONDITIONAL_GE); bld.CMP(bld.null_reg_f(), dest, zero, BRW_CONDITIONAL_GE);
@ -561,9 +561,9 @@ TEST_F(cmod_propagation_test, negate)
TEST_F(cmod_propagation_test, movnz) TEST_F(cmod_propagation_test, movnz)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_F); brw_reg dest = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
bld.CMP(dest, src0, src1, BRW_CONDITIONAL_GE); bld.CMP(dest, src0, src1, BRW_CONDITIONAL_GE);
set_condmod(BRW_CONDITIONAL_NZ, set_condmod(BRW_CONDITIONAL_NZ,
bld.MOV(bld.null_reg_f(), dest)); bld.MOV(bld.null_reg_f(), dest));
@ -592,10 +592,10 @@ TEST_F(cmod_propagation_test, movnz)
TEST_F(cmod_propagation_test, different_types_cmod_with_zero) TEST_F(cmod_propagation_test, different_types_cmod_with_zero)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_D); brw_reg dest = bld.vgrf(BRW_TYPE_D);
fs_reg src0 = bld.vgrf(BRW_TYPE_D); brw_reg src0 = bld.vgrf(BRW_TYPE_D);
fs_reg src1 = bld.vgrf(BRW_TYPE_D); brw_reg src1 = bld.vgrf(BRW_TYPE_D);
fs_reg zero(brw_imm_f(0.0f)); brw_reg zero(brw_imm_f(0.0f));
bld.ADD(dest, src0, src1); bld.ADD(dest, src0, src1);
bld.CMP(bld.null_reg_f(), retype(dest, BRW_TYPE_F), zero, bld.CMP(bld.null_reg_f(), retype(dest, BRW_TYPE_F), zero,
BRW_CONDITIONAL_GE); BRW_CONDITIONAL_GE);
@ -625,10 +625,10 @@ TEST_F(cmod_propagation_test, different_types_cmod_with_zero)
TEST_F(cmod_propagation_test, andnz_one) TEST_F(cmod_propagation_test, andnz_one)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_D); brw_reg dest = bld.vgrf(BRW_TYPE_D);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg zero(brw_imm_f(0.0f)); brw_reg zero(brw_imm_f(0.0f));
fs_reg one(brw_imm_d(1)); brw_reg one(brw_imm_d(1));
bld.CMP(retype(dest, BRW_TYPE_F), src0, zero, BRW_CONDITIONAL_L); bld.CMP(retype(dest, BRW_TYPE_F), src0, zero, BRW_CONDITIONAL_L);
set_condmod(BRW_CONDITIONAL_NZ, set_condmod(BRW_CONDITIONAL_NZ,
@ -659,10 +659,10 @@ TEST_F(cmod_propagation_test, andnz_one)
TEST_F(cmod_propagation_test, andnz_non_one) TEST_F(cmod_propagation_test, andnz_non_one)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_D); brw_reg dest = bld.vgrf(BRW_TYPE_D);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg zero(brw_imm_f(0.0f)); brw_reg zero(brw_imm_f(0.0f));
fs_reg nonone(brw_imm_d(38)); brw_reg nonone(brw_imm_d(38));
bld.CMP(retype(dest, BRW_TYPE_F), src0, zero, BRW_CONDITIONAL_L); bld.CMP(retype(dest, BRW_TYPE_F), src0, zero, BRW_CONDITIONAL_L);
set_condmod(BRW_CONDITIONAL_NZ, set_condmod(BRW_CONDITIONAL_NZ,
@ -693,9 +693,9 @@ TEST_F(cmod_propagation_test, andnz_non_one)
TEST_F(cmod_propagation_test, cmp_cmpnz) TEST_F(cmod_propagation_test, cmp_cmpnz)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_F); brw_reg dst0 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg zero(brw_imm_f(0)); brw_reg zero(brw_imm_f(0));
bld.CMP(dst0, src0, zero, BRW_CONDITIONAL_NZ); bld.CMP(dst0, src0, zero, BRW_CONDITIONAL_NZ);
bld.CMP(bld.null_reg_f(), dst0, zero, BRW_CONDITIONAL_NZ); bld.CMP(bld.null_reg_f(), dst0, zero, BRW_CONDITIONAL_NZ);
@ -720,9 +720,9 @@ TEST_F(cmod_propagation_test, cmp_cmpnz)
TEST_F(cmod_propagation_test, cmp_cmpg) TEST_F(cmod_propagation_test, cmp_cmpg)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_F); brw_reg dst0 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg zero(brw_imm_f(0)); brw_reg zero(brw_imm_f(0));
bld.CMP(dst0, src0, zero, BRW_CONDITIONAL_NZ); bld.CMP(dst0, src0, zero, BRW_CONDITIONAL_NZ);
bld.CMP(bld.null_reg_f(), dst0, zero, BRW_CONDITIONAL_G); bld.CMP(bld.null_reg_f(), dst0, zero, BRW_CONDITIONAL_G);
@ -749,9 +749,9 @@ TEST_F(cmod_propagation_test, cmp_cmpg)
TEST_F(cmod_propagation_test, plnnz_cmpnz) TEST_F(cmod_propagation_test, plnnz_cmpnz)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_F); brw_reg dst0 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg zero(brw_imm_f(0)); brw_reg zero(brw_imm_f(0));
set_condmod(BRW_CONDITIONAL_NZ, bld.PLN(dst0, src0, zero)); set_condmod(BRW_CONDITIONAL_NZ, bld.PLN(dst0, src0, zero));
bld.CMP(bld.null_reg_f(), dst0, zero, BRW_CONDITIONAL_NZ); bld.CMP(bld.null_reg_f(), dst0, zero, BRW_CONDITIONAL_NZ);
@ -776,9 +776,9 @@ TEST_F(cmod_propagation_test, plnnz_cmpnz)
TEST_F(cmod_propagation_test, plnnz_cmpz) TEST_F(cmod_propagation_test, plnnz_cmpz)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_F); brw_reg dst0 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg zero(brw_imm_f(0)); brw_reg zero(brw_imm_f(0));
set_condmod(BRW_CONDITIONAL_NZ, bld.PLN(dst0, src0, zero)); set_condmod(BRW_CONDITIONAL_NZ, bld.PLN(dst0, src0, zero));
bld.CMP(bld.null_reg_f(), dst0, zero, BRW_CONDITIONAL_Z); bld.CMP(bld.null_reg_f(), dst0, zero, BRW_CONDITIONAL_Z);
@ -803,10 +803,10 @@ TEST_F(cmod_propagation_test, plnnz_cmpz)
TEST_F(cmod_propagation_test, plnnz_sel_cmpz) TEST_F(cmod_propagation_test, plnnz_sel_cmpz)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_F); brw_reg dst0 = bld.vgrf(BRW_TYPE_F);
fs_reg dst1 = bld.vgrf(BRW_TYPE_F); brw_reg dst1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg zero(brw_imm_f(0)); brw_reg zero(brw_imm_f(0));
set_condmod(BRW_CONDITIONAL_NZ, bld.PLN(dst0, src0, zero)); set_condmod(BRW_CONDITIONAL_NZ, bld.PLN(dst0, src0, zero));
set_predicate(BRW_PREDICATE_NORMAL, bld.SEL(dst1, src0, zero)); set_predicate(BRW_PREDICATE_NORMAL, bld.SEL(dst1, src0, zero));
@ -837,9 +837,9 @@ TEST_F(cmod_propagation_test, plnnz_sel_cmpz)
TEST_F(cmod_propagation_test, cmp_cmpg_D) TEST_F(cmod_propagation_test, cmp_cmpg_D)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_D); brw_reg dst0 = bld.vgrf(BRW_TYPE_D);
fs_reg src0 = bld.vgrf(BRW_TYPE_D); brw_reg src0 = bld.vgrf(BRW_TYPE_D);
fs_reg zero(brw_imm_d(0)); brw_reg zero(brw_imm_d(0));
bld.CMP(dst0, src0, zero, BRW_CONDITIONAL_NZ); bld.CMP(dst0, src0, zero, BRW_CONDITIONAL_NZ);
bld.CMP(bld.null_reg_d(), dst0, zero, BRW_CONDITIONAL_G); bld.CMP(bld.null_reg_d(), dst0, zero, BRW_CONDITIONAL_G);
@ -866,9 +866,9 @@ TEST_F(cmod_propagation_test, cmp_cmpg_D)
TEST_F(cmod_propagation_test, cmp_cmpg_UD) TEST_F(cmod_propagation_test, cmp_cmpg_UD)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_UD); brw_reg dst0 = bld.vgrf(BRW_TYPE_UD);
fs_reg src0 = bld.vgrf(BRW_TYPE_UD); brw_reg src0 = bld.vgrf(BRW_TYPE_UD);
fs_reg zero(brw_imm_ud(0)); brw_reg zero(brw_imm_ud(0));
bld.CMP(dst0, src0, zero, BRW_CONDITIONAL_NZ); bld.CMP(dst0, src0, zero, BRW_CONDITIONAL_NZ);
bld.CMP(bld.null_reg_ud(), dst0, zero, BRW_CONDITIONAL_G); bld.CMP(bld.null_reg_ud(), dst0, zero, BRW_CONDITIONAL_G);
@ -893,9 +893,9 @@ TEST_F(cmod_propagation_test, cmp_cmpg_UD)
TEST_F(cmod_propagation_test, cmp_cmpl_D) TEST_F(cmod_propagation_test, cmp_cmpl_D)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_D); brw_reg dst0 = bld.vgrf(BRW_TYPE_D);
fs_reg src0 = bld.vgrf(BRW_TYPE_D); brw_reg src0 = bld.vgrf(BRW_TYPE_D);
fs_reg zero(brw_imm_d(0)); brw_reg zero(brw_imm_d(0));
bld.CMP(dst0, src0, zero, BRW_CONDITIONAL_NZ); bld.CMP(dst0, src0, zero, BRW_CONDITIONAL_NZ);
bld.CMP(bld.null_reg_d(), dst0, zero, BRW_CONDITIONAL_L); bld.CMP(bld.null_reg_d(), dst0, zero, BRW_CONDITIONAL_L);
@ -920,9 +920,9 @@ TEST_F(cmod_propagation_test, cmp_cmpl_D)
TEST_F(cmod_propagation_test, cmp_cmpl_UD) TEST_F(cmod_propagation_test, cmp_cmpl_UD)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_UD); brw_reg dst0 = bld.vgrf(BRW_TYPE_UD);
fs_reg src0 = bld.vgrf(BRW_TYPE_UD); brw_reg src0 = bld.vgrf(BRW_TYPE_UD);
fs_reg zero(brw_imm_ud(0)); brw_reg zero(brw_imm_ud(0));
bld.CMP(dst0, src0, zero, BRW_CONDITIONAL_NZ); bld.CMP(dst0, src0, zero, BRW_CONDITIONAL_NZ);
bld.CMP(bld.null_reg_ud(), dst0, zero, BRW_CONDITIONAL_L); bld.CMP(bld.null_reg_ud(), dst0, zero, BRW_CONDITIONAL_L);
@ -949,10 +949,10 @@ TEST_F(cmod_propagation_test, cmp_cmpl_UD)
TEST_F(cmod_propagation_test, andz_one) TEST_F(cmod_propagation_test, andz_one)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_D); brw_reg dest = bld.vgrf(BRW_TYPE_D);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg zero(brw_imm_f(0.0f)); brw_reg zero(brw_imm_f(0.0f));
fs_reg one(brw_imm_d(1)); brw_reg one(brw_imm_d(1));
bld.CMP(retype(dest, BRW_TYPE_F), src0, zero, BRW_CONDITIONAL_L); bld.CMP(retype(dest, BRW_TYPE_F), src0, zero, BRW_CONDITIONAL_L);
set_condmod(BRW_CONDITIONAL_Z, set_condmod(BRW_CONDITIONAL_Z,
@ -983,9 +983,9 @@ TEST_F(cmod_propagation_test, andz_one)
TEST_F(cmod_propagation_test, add_not_merge_with_compare) TEST_F(cmod_propagation_test, add_not_merge_with_compare)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_F); brw_reg dest = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
bld.ADD(dest, src0, src1); bld.ADD(dest, src0, src1);
bld.CMP(bld.null_reg_f(), src0, src1, BRW_CONDITIONAL_L); bld.CMP(bld.null_reg_f(), src0, src1, BRW_CONDITIONAL_L);
@ -1016,9 +1016,9 @@ TEST_F(cmod_propagation_test, add_not_merge_with_compare)
TEST_F(cmod_propagation_test, subtract_merge_with_compare) TEST_F(cmod_propagation_test, subtract_merge_with_compare)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_F); brw_reg dest = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
bld.ADD(dest, src0, negate(src1)); bld.ADD(dest, src0, negate(src1));
bld.CMP(bld.null_reg_f(), src0, src1, BRW_CONDITIONAL_L); bld.CMP(bld.null_reg_f(), src0, src1, BRW_CONDITIONAL_L);
@ -1044,10 +1044,10 @@ TEST_F(cmod_propagation_test, subtract_merge_with_compare)
TEST_F(cmod_propagation_test, subtract_immediate_merge_with_compare) TEST_F(cmod_propagation_test, subtract_immediate_merge_with_compare)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_F); brw_reg dest = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg one(brw_imm_f(1.0f)); brw_reg one(brw_imm_f(1.0f));
fs_reg negative_one(brw_imm_f(-1.0f)); brw_reg negative_one(brw_imm_f(-1.0f));
bld.ADD(dest, src0, negative_one); bld.ADD(dest, src0, negative_one);
bld.CMP(bld.null_reg_f(), src0, one, BRW_CONDITIONAL_NZ); bld.CMP(bld.null_reg_f(), src0, one, BRW_CONDITIONAL_NZ);
@ -1074,10 +1074,10 @@ TEST_F(cmod_propagation_test, subtract_immediate_merge_with_compare)
TEST_F(cmod_propagation_test, subtract_merge_with_compare_intervening_add) TEST_F(cmod_propagation_test, subtract_merge_with_compare_intervening_add)
{ {
fs_reg dest0 = bld.vgrf(BRW_TYPE_F); brw_reg dest0 = bld.vgrf(BRW_TYPE_F);
fs_reg dest1 = bld.vgrf(BRW_TYPE_F); brw_reg dest1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
bld.ADD(dest0, src0, negate(src1)); bld.ADD(dest0, src0, negate(src1));
bld.ADD(dest1, src0, src1); bld.ADD(dest1, src0, src1);
bld.CMP(bld.null_reg_f(), src0, src1, BRW_CONDITIONAL_L); bld.CMP(bld.null_reg_f(), src0, src1, BRW_CONDITIONAL_L);
@ -1108,10 +1108,10 @@ TEST_F(cmod_propagation_test, subtract_merge_with_compare_intervening_add)
TEST_F(cmod_propagation_test, subtract_not_merge_with_compare_intervening_partial_write) TEST_F(cmod_propagation_test, subtract_not_merge_with_compare_intervening_partial_write)
{ {
fs_reg dest0 = bld.vgrf(BRW_TYPE_F); brw_reg dest0 = bld.vgrf(BRW_TYPE_F);
fs_reg dest1 = bld.vgrf(BRW_TYPE_F); brw_reg dest1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
bld.ADD(dest0, src0, negate(src1)); bld.ADD(dest0, src0, negate(src1));
set_predicate(BRW_PREDICATE_NORMAL, bld.ADD(dest1, src0, negate(src1))); set_predicate(BRW_PREDICATE_NORMAL, bld.ADD(dest1, src0, negate(src1)));
bld.CMP(bld.null_reg_f(), src0, src1, BRW_CONDITIONAL_L); bld.CMP(bld.null_reg_f(), src0, src1, BRW_CONDITIONAL_L);
@ -1143,10 +1143,10 @@ TEST_F(cmod_propagation_test, subtract_not_merge_with_compare_intervening_partia
TEST_F(cmod_propagation_test, subtract_not_merge_with_compare_intervening_add) TEST_F(cmod_propagation_test, subtract_not_merge_with_compare_intervening_add)
{ {
fs_reg dest0 = bld.vgrf(BRW_TYPE_F); brw_reg dest0 = bld.vgrf(BRW_TYPE_F);
fs_reg dest1 = bld.vgrf(BRW_TYPE_F); brw_reg dest1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
bld.ADD(dest0, src0, negate(src1)); bld.ADD(dest0, src0, negate(src1));
set_condmod(BRW_CONDITIONAL_EQ, bld.ADD(dest1, src0, src1)); set_condmod(BRW_CONDITIONAL_EQ, bld.ADD(dest1, src0, src1));
bld.CMP(bld.null_reg_f(), src0, src1, BRW_CONDITIONAL_L); bld.CMP(bld.null_reg_f(), src0, src1, BRW_CONDITIONAL_L);
@ -1178,9 +1178,9 @@ TEST_F(cmod_propagation_test, subtract_not_merge_with_compare_intervening_add)
TEST_F(cmod_propagation_test, add_merge_with_compare) TEST_F(cmod_propagation_test, add_merge_with_compare)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_F); brw_reg dest = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
bld.ADD(dest, src0, src1); bld.ADD(dest, src0, src1);
bld.CMP(bld.null_reg_f(), src0, negate(src1), BRW_CONDITIONAL_L); bld.CMP(bld.null_reg_f(), src0, negate(src1), BRW_CONDITIONAL_L);
@ -1206,9 +1206,9 @@ TEST_F(cmod_propagation_test, add_merge_with_compare)
TEST_F(cmod_propagation_test, negative_subtract_merge_with_compare) TEST_F(cmod_propagation_test, negative_subtract_merge_with_compare)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_F); brw_reg dest = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
bld.ADD(dest, src1, negate(src0)); bld.ADD(dest, src1, negate(src0));
bld.CMP(bld.null_reg_f(), src0, src1, BRW_CONDITIONAL_L); bld.CMP(bld.null_reg_f(), src0, src1, BRW_CONDITIONAL_L);
@ -1237,11 +1237,11 @@ TEST_F(cmod_propagation_test, negative_subtract_merge_with_compare)
TEST_F(cmod_propagation_test, subtract_delete_compare) TEST_F(cmod_propagation_test, subtract_delete_compare)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_F); brw_reg dest = bld.vgrf(BRW_TYPE_F);
fs_reg dest1 = bld.vgrf(BRW_TYPE_F); brw_reg dest1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
fs_reg src2 = bld.vgrf(BRW_TYPE_F); brw_reg src2 = bld.vgrf(BRW_TYPE_F);
set_condmod(BRW_CONDITIONAL_L, bld.ADD(dest, src0, negate(src1))); set_condmod(BRW_CONDITIONAL_L, bld.ADD(dest, src0, negate(src1)));
set_predicate(BRW_PREDICATE_NORMAL, bld.MOV(dest1, src2)); set_predicate(BRW_PREDICATE_NORMAL, bld.MOV(dest1, src2));
@ -1276,11 +1276,11 @@ TEST_F(cmod_propagation_test, subtract_delete_compare_other_flag)
/* This test is the same as subtract_delete_compare but it explicitly used /* This test is the same as subtract_delete_compare but it explicitly used
* flag f0.1 for the subtraction and the comparison. * flag f0.1 for the subtraction and the comparison.
*/ */
fs_reg dest = bld.vgrf(BRW_TYPE_F); brw_reg dest = bld.vgrf(BRW_TYPE_F);
fs_reg dest1 = bld.vgrf(BRW_TYPE_F); brw_reg dest1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
fs_reg src2 = bld.vgrf(BRW_TYPE_F); brw_reg src2 = bld.vgrf(BRW_TYPE_F);
set_condmod(BRW_CONDITIONAL_L, bld.ADD(dest, src0, negate(src1))) set_condmod(BRW_CONDITIONAL_L, bld.ADD(dest, src0, negate(src1)))
->flag_subreg = 1; ->flag_subreg = 1;
@ -1315,9 +1315,9 @@ TEST_F(cmod_propagation_test, subtract_delete_compare_other_flag)
TEST_F(cmod_propagation_test, subtract_to_mismatch_flag) TEST_F(cmod_propagation_test, subtract_to_mismatch_flag)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_F); brw_reg dest = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
set_condmod(BRW_CONDITIONAL_L, bld.ADD(dest, src0, negate(src1))); set_condmod(BRW_CONDITIONAL_L, bld.ADD(dest, src0, negate(src1)));
bld.CMP(bld.null_reg_f(), src0, src1, BRW_CONDITIONAL_L) bld.CMP(bld.null_reg_f(), src0, src1, BRW_CONDITIONAL_L)
@ -1350,9 +1350,9 @@ TEST_F(cmod_propagation_test, subtract_to_mismatch_flag)
TEST_F(cmod_propagation_test, TEST_F(cmod_propagation_test,
subtract_merge_with_compare_intervening_mismatch_flag_write) subtract_merge_with_compare_intervening_mismatch_flag_write)
{ {
fs_reg dest0 = bld.vgrf(BRW_TYPE_F); brw_reg dest0 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
bld.ADD(dest0, src0, negate(src1)); bld.ADD(dest0, src0, negate(src1));
bld.CMP(bld.null_reg_f(), src0, src1, BRW_CONDITIONAL_L) bld.CMP(bld.null_reg_f(), src0, src1, BRW_CONDITIONAL_L)
@ -1397,12 +1397,12 @@ TEST_F(cmod_propagation_test,
TEST_F(cmod_propagation_test, TEST_F(cmod_propagation_test,
subtract_merge_with_compare_intervening_mismatch_flag_read) subtract_merge_with_compare_intervening_mismatch_flag_read)
{ {
fs_reg dest0 = bld.vgrf(BRW_TYPE_F); brw_reg dest0 = bld.vgrf(BRW_TYPE_F);
fs_reg dest1 = bld.vgrf(BRW_TYPE_F); brw_reg dest1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
fs_reg src2 = bld.vgrf(BRW_TYPE_F); brw_reg src2 = bld.vgrf(BRW_TYPE_F);
fs_reg zero(brw_imm_f(0.0f)); brw_reg zero(brw_imm_f(0.0f));
bld.ADD(dest0, src0, negate(src1)); bld.ADD(dest0, src0, negate(src1));
set_predicate(BRW_PREDICATE_NORMAL, bld.SEL(dest1, src2, zero)) set_predicate(BRW_PREDICATE_NORMAL, bld.SEL(dest1, src2, zero))
@ -1437,10 +1437,10 @@ TEST_F(cmod_propagation_test,
TEST_F(cmod_propagation_test, subtract_delete_compare_derp) TEST_F(cmod_propagation_test, subtract_delete_compare_derp)
{ {
fs_reg dest0 = bld.vgrf(BRW_TYPE_F); brw_reg dest0 = bld.vgrf(BRW_TYPE_F);
fs_reg dest1 = bld.vgrf(BRW_TYPE_F); brw_reg dest1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
set_condmod(BRW_CONDITIONAL_L, bld.ADD(dest0, src0, negate(src1))); set_condmod(BRW_CONDITIONAL_L, bld.ADD(dest0, src0, negate(src1)));
set_predicate(BRW_PREDICATE_NORMAL, bld.ADD(dest1, negate(src0), src1)); set_predicate(BRW_PREDICATE_NORMAL, bld.ADD(dest1, negate(src0), src1));
@ -1472,8 +1472,8 @@ TEST_F(cmod_propagation_test, subtract_delete_compare_derp)
TEST_F(cmod_propagation_test, signed_unsigned_comparison_mismatch) TEST_F(cmod_propagation_test, signed_unsigned_comparison_mismatch)
{ {
fs_reg dest0 = bld.vgrf(BRW_TYPE_D); brw_reg dest0 = bld.vgrf(BRW_TYPE_D);
fs_reg src0 = bld.vgrf(BRW_TYPE_D); brw_reg src0 = bld.vgrf(BRW_TYPE_D);
src0.type = BRW_TYPE_W; src0.type = BRW_TYPE_W;
bld.ASR(dest0, negate(src0), brw_imm_d(15)); bld.ASR(dest0, negate(src0), brw_imm_d(15));
@ -1503,9 +1503,9 @@ TEST_F(cmod_propagation_test, signed_unsigned_comparison_mismatch)
TEST_F(cmod_propagation_test, ior_f2i_nz) TEST_F(cmod_propagation_test, ior_f2i_nz)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_D); brw_reg dest = bld.vgrf(BRW_TYPE_D);
fs_reg src0 = bld.vgrf(BRW_TYPE_D); brw_reg src0 = bld.vgrf(BRW_TYPE_D);
fs_reg src1 = bld.vgrf(BRW_TYPE_D); brw_reg src1 = bld.vgrf(BRW_TYPE_D);
bld.OR(dest, src0, src1); bld.OR(dest, src0, src1);
bld.MOV(bld.null_reg_d(), retype(dest, BRW_TYPE_F)) bld.MOV(bld.null_reg_d(), retype(dest, BRW_TYPE_F))
@ -1549,9 +1549,9 @@ cmod_propagation_test::test_mov_prop(enum brw_conditional_mod cmod,
enum brw_reg_type mov_dst_type, enum brw_reg_type mov_dst_type,
bool expected_cmod_prop_progress) bool expected_cmod_prop_progress)
{ {
fs_reg dest = bld.vgrf(add_type); brw_reg dest = bld.vgrf(add_type);
fs_reg src0 = bld.vgrf(add_type); brw_reg src0 = bld.vgrf(add_type);
fs_reg src1 = bld.vgrf(add_type); brw_reg src1 = bld.vgrf(add_type);
bld.ADD(dest, src0, src1); bld.ADD(dest, src0, src1);
bld.MOV(retype(bld.null_reg_ud(), mov_dst_type), dest) bld.MOV(retype(bld.null_reg_ud(), mov_dst_type), dest)
@ -2126,10 +2126,10 @@ cmod_propagation_test::test_saturate_prop(enum brw_conditional_mod before,
enum brw_reg_type op_type, enum brw_reg_type op_type,
bool expected_cmod_prop_progress) bool expected_cmod_prop_progress)
{ {
fs_reg dest = bld.vgrf(add_type); brw_reg dest = bld.vgrf(add_type);
fs_reg src0 = bld.vgrf(add_type); brw_reg src0 = bld.vgrf(add_type);
fs_reg src1 = bld.vgrf(add_type); brw_reg src1 = bld.vgrf(add_type);
fs_reg zero(brw_imm_ud(0)); brw_reg zero(brw_imm_ud(0));
bld.ADD(dest, src0, src1)->saturate = true; bld.ADD(dest, src0, src1)->saturate = true;
@ -2611,9 +2611,9 @@ TEST_F(cmod_propagation_test, not_to_or)
/* Exercise propagation of conditional modifier from a NOT instruction to /* Exercise propagation of conditional modifier from a NOT instruction to
* another ALU instruction as performed by cmod_propagate_not. * another ALU instruction as performed by cmod_propagate_not.
*/ */
fs_reg dest = bld.vgrf(BRW_TYPE_UD); brw_reg dest = bld.vgrf(BRW_TYPE_UD);
fs_reg src0 = bld.vgrf(BRW_TYPE_UD); brw_reg src0 = bld.vgrf(BRW_TYPE_UD);
fs_reg src1 = bld.vgrf(BRW_TYPE_UD); brw_reg src1 = bld.vgrf(BRW_TYPE_UD);
bld.OR(dest, src0, src1); bld.OR(dest, src0, src1);
set_condmod(BRW_CONDITIONAL_NZ, bld.NOT(bld.null_reg_ud(), dest)); set_condmod(BRW_CONDITIONAL_NZ, bld.NOT(bld.null_reg_ud(), dest));
@ -2644,9 +2644,9 @@ TEST_F(cmod_propagation_test, not_to_and)
/* Exercise propagation of conditional modifier from a NOT instruction to /* Exercise propagation of conditional modifier from a NOT instruction to
* another ALU instruction as performed by cmod_propagate_not. * another ALU instruction as performed by cmod_propagate_not.
*/ */
fs_reg dest = bld.vgrf(BRW_TYPE_UD); brw_reg dest = bld.vgrf(BRW_TYPE_UD);
fs_reg src0 = bld.vgrf(BRW_TYPE_UD); brw_reg src0 = bld.vgrf(BRW_TYPE_UD);
fs_reg src1 = bld.vgrf(BRW_TYPE_UD); brw_reg src1 = bld.vgrf(BRW_TYPE_UD);
bld.AND(dest, src0, src1); bld.AND(dest, src0, src1);
set_condmod(BRW_CONDITIONAL_NZ, bld.NOT(bld.null_reg_ud(), dest)); set_condmod(BRW_CONDITIONAL_NZ, bld.NOT(bld.null_reg_ud(), dest));
@ -2682,9 +2682,9 @@ TEST_F(cmod_propagation_test, not_to_uadd)
* restriction is just the the destination type of the ALU instruction is * restriction is just the the destination type of the ALU instruction is
* the same as the source type of the NOT instruction. * the same as the source type of the NOT instruction.
*/ */
fs_reg dest = bld.vgrf(BRW_TYPE_UD); brw_reg dest = bld.vgrf(BRW_TYPE_UD);
fs_reg src0 = bld.vgrf(BRW_TYPE_UD); brw_reg src0 = bld.vgrf(BRW_TYPE_UD);
fs_reg src1 = bld.vgrf(BRW_TYPE_UD); brw_reg src1 = bld.vgrf(BRW_TYPE_UD);
bld.ADD(dest, src0, src1); bld.ADD(dest, src0, src1);
set_condmod(BRW_CONDITIONAL_NZ, bld.NOT(bld.null_reg_ud(), dest)); set_condmod(BRW_CONDITIONAL_NZ, bld.NOT(bld.null_reg_ud(), dest));
@ -2722,9 +2722,9 @@ TEST_F(cmod_propagation_test, not_to_fadd_to_ud)
* restriction is just the the destination type of the ALU instruction is * restriction is just the the destination type of the ALU instruction is
* the same as the source type of the NOT instruction. * the same as the source type of the NOT instruction.
*/ */
fs_reg dest = bld.vgrf(BRW_TYPE_UD); brw_reg dest = bld.vgrf(BRW_TYPE_UD);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
bld.ADD(dest, src0, src1); bld.ADD(dest, src0, src1);
set_condmod(BRW_CONDITIONAL_NZ, bld.NOT(bld.null_reg_ud(), dest)); set_condmod(BRW_CONDITIONAL_NZ, bld.NOT(bld.null_reg_ud(), dest));
@ -2762,9 +2762,9 @@ TEST_F(cmod_propagation_test, not_to_fadd)
* restriction is just the the destination type of the ALU instruction is * restriction is just the the destination type of the ALU instruction is
* the same as the source type of the NOT instruction. * the same as the source type of the NOT instruction.
*/ */
fs_reg dest = bld.vgrf(BRW_TYPE_F); brw_reg dest = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
bld.ADD(dest, src0, src1); bld.ADD(dest, src0, src1);
set_condmod(BRW_CONDITIONAL_NZ, set_condmod(BRW_CONDITIONAL_NZ,
bld.NOT(bld.null_reg_ud(), bld.NOT(bld.null_reg_ud(),
@ -2799,12 +2799,12 @@ TEST_F(cmod_propagation_test, not_to_or_intervening_flag_read_compatible_value)
/* Exercise propagation of conditional modifier from a NOT instruction to /* Exercise propagation of conditional modifier from a NOT instruction to
* another ALU instruction as performed by cmod_propagate_not. * another ALU instruction as performed by cmod_propagate_not.
*/ */
fs_reg dest0 = bld.vgrf(BRW_TYPE_UD); brw_reg dest0 = bld.vgrf(BRW_TYPE_UD);
fs_reg dest1 = bld.vgrf(BRW_TYPE_F); brw_reg dest1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_UD); brw_reg src0 = bld.vgrf(BRW_TYPE_UD);
fs_reg src1 = bld.vgrf(BRW_TYPE_UD); brw_reg src1 = bld.vgrf(BRW_TYPE_UD);
fs_reg src2 = bld.vgrf(BRW_TYPE_F); brw_reg src2 = bld.vgrf(BRW_TYPE_F);
fs_reg zero(brw_imm_f(0.0f)); brw_reg zero(brw_imm_f(0.0f));
set_condmod(BRW_CONDITIONAL_Z, bld.OR(dest0, src0, src1)); set_condmod(BRW_CONDITIONAL_Z, bld.OR(dest0, src0, src1));
set_predicate(BRW_PREDICATE_NORMAL, bld.SEL(dest1, src2, zero)); set_predicate(BRW_PREDICATE_NORMAL, bld.SEL(dest1, src2, zero));
set_condmod(BRW_CONDITIONAL_NZ, bld.NOT(bld.null_reg_ud(), dest0)); set_condmod(BRW_CONDITIONAL_NZ, bld.NOT(bld.null_reg_ud(), dest0));
@ -2841,12 +2841,12 @@ TEST_F(cmod_propagation_test,
/* Exercise propagation of conditional modifier from a NOT instruction to /* Exercise propagation of conditional modifier from a NOT instruction to
* another ALU instruction as performed by cmod_propagate_not. * another ALU instruction as performed by cmod_propagate_not.
*/ */
fs_reg dest0 = bld.vgrf(BRW_TYPE_UD); brw_reg dest0 = bld.vgrf(BRW_TYPE_UD);
fs_reg dest1 = bld.vgrf(BRW_TYPE_F); brw_reg dest1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_UD); brw_reg src0 = bld.vgrf(BRW_TYPE_UD);
fs_reg src1 = bld.vgrf(BRW_TYPE_UD); brw_reg src1 = bld.vgrf(BRW_TYPE_UD);
fs_reg src2 = bld.vgrf(BRW_TYPE_F); brw_reg src2 = bld.vgrf(BRW_TYPE_F);
fs_reg zero(brw_imm_f(0.0f)); brw_reg zero(brw_imm_f(0.0f));
set_condmod(BRW_CONDITIONAL_Z, bld.OR(dest0, src0, src1)) set_condmod(BRW_CONDITIONAL_Z, bld.OR(dest0, src0, src1))
->flag_subreg = 1; ->flag_subreg = 1;
set_predicate(BRW_PREDICATE_NORMAL, bld.SEL(dest1, src2, zero)); set_predicate(BRW_PREDICATE_NORMAL, bld.SEL(dest1, src2, zero));
@ -2886,12 +2886,12 @@ TEST_F(cmod_propagation_test, not_to_or_intervening_flag_read_incompatible_value
/* Exercise propagation of conditional modifier from a NOT instruction to /* Exercise propagation of conditional modifier from a NOT instruction to
* another ALU instruction as performed by cmod_propagate_not. * another ALU instruction as performed by cmod_propagate_not.
*/ */
fs_reg dest0 = bld.vgrf(BRW_TYPE_UD); brw_reg dest0 = bld.vgrf(BRW_TYPE_UD);
fs_reg dest1 = bld.vgrf(BRW_TYPE_F); brw_reg dest1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_UD); brw_reg src0 = bld.vgrf(BRW_TYPE_UD);
fs_reg src1 = bld.vgrf(BRW_TYPE_UD); brw_reg src1 = bld.vgrf(BRW_TYPE_UD);
fs_reg src2 = bld.vgrf(BRW_TYPE_F); brw_reg src2 = bld.vgrf(BRW_TYPE_F);
fs_reg zero(brw_imm_f(0.0f)); brw_reg zero(brw_imm_f(0.0f));
set_condmod(BRW_CONDITIONAL_NZ, bld.OR(dest0, src0, src1)); set_condmod(BRW_CONDITIONAL_NZ, bld.OR(dest0, src0, src1));
set_predicate(BRW_PREDICATE_NORMAL, bld.SEL(dest1, src2, zero)); set_predicate(BRW_PREDICATE_NORMAL, bld.SEL(dest1, src2, zero));
set_condmod(BRW_CONDITIONAL_NZ, bld.NOT(bld.null_reg_ud(), dest0)); set_condmod(BRW_CONDITIONAL_NZ, bld.NOT(bld.null_reg_ud(), dest0));
@ -2928,10 +2928,10 @@ TEST_F(cmod_propagation_test, not_to_or_intervening_mismatch_flag_write)
/* Exercise propagation of conditional modifier from a NOT instruction to /* Exercise propagation of conditional modifier from a NOT instruction to
* another ALU instruction as performed by cmod_propagate_not. * another ALU instruction as performed by cmod_propagate_not.
*/ */
fs_reg dest0 = bld.vgrf(BRW_TYPE_UD); brw_reg dest0 = bld.vgrf(BRW_TYPE_UD);
fs_reg dest1 = bld.vgrf(BRW_TYPE_UD); brw_reg dest1 = bld.vgrf(BRW_TYPE_UD);
fs_reg src0 = bld.vgrf(BRW_TYPE_UD); brw_reg src0 = bld.vgrf(BRW_TYPE_UD);
fs_reg src1 = bld.vgrf(BRW_TYPE_UD); brw_reg src1 = bld.vgrf(BRW_TYPE_UD);
bld.OR(dest0, src0, src1); bld.OR(dest0, src0, src1);
set_condmod(BRW_CONDITIONAL_Z, bld.OR(dest1, src0, src1)) set_condmod(BRW_CONDITIONAL_Z, bld.OR(dest1, src0, src1))
@ -2971,12 +2971,12 @@ TEST_F(cmod_propagation_test, not_to_or_intervening_mismatch_flag_read)
/* Exercise propagation of conditional modifier from a NOT instruction to /* Exercise propagation of conditional modifier from a NOT instruction to
* another ALU instruction as performed by cmod_propagate_not. * another ALU instruction as performed by cmod_propagate_not.
*/ */
fs_reg dest0 = bld.vgrf(BRW_TYPE_UD); brw_reg dest0 = bld.vgrf(BRW_TYPE_UD);
fs_reg dest1 = bld.vgrf(BRW_TYPE_F); brw_reg dest1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_UD); brw_reg src0 = bld.vgrf(BRW_TYPE_UD);
fs_reg src1 = bld.vgrf(BRW_TYPE_UD); brw_reg src1 = bld.vgrf(BRW_TYPE_UD);
fs_reg src2 = bld.vgrf(BRW_TYPE_F); brw_reg src2 = bld.vgrf(BRW_TYPE_F);
fs_reg zero(brw_imm_f(0.0f)); brw_reg zero(brw_imm_f(0.0f));
bld.OR(dest0, src0, src1); bld.OR(dest0, src0, src1);
set_predicate(BRW_PREDICATE_NORMAL, bld.SEL(dest1, src2, zero)) set_predicate(BRW_PREDICATE_NORMAL, bld.SEL(dest1, src2, zero))
@ -3013,10 +3013,10 @@ TEST_F(cmod_propagation_test, not_to_or_intervening_mismatch_flag_read)
TEST_F(cmod_propagation_test, cmp_to_add_float_e) TEST_F(cmod_propagation_test, cmp_to_add_float_e)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_F); brw_reg dest = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg neg10(brw_imm_f(-10.0f)); brw_reg neg10(brw_imm_f(-10.0f));
fs_reg pos10(brw_imm_f(10.0f)); brw_reg pos10(brw_imm_f(10.0f));
bld.ADD(dest, src0, neg10)->saturate = true; bld.ADD(dest, src0, neg10)->saturate = true;
bld.CMP(bld.null_reg_f(), src0, pos10, BRW_CONDITIONAL_EQ); bld.CMP(bld.null_reg_f(), src0, pos10, BRW_CONDITIONAL_EQ);
@ -3043,10 +3043,10 @@ TEST_F(cmod_propagation_test, cmp_to_add_float_e)
TEST_F(cmod_propagation_test, cmp_to_add_float_g) TEST_F(cmod_propagation_test, cmp_to_add_float_g)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_F); brw_reg dest = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg neg10(brw_imm_f(-10.0f)); brw_reg neg10(brw_imm_f(-10.0f));
fs_reg pos10(brw_imm_f(10.0f)); brw_reg pos10(brw_imm_f(10.0f));
bld.ADD(dest, src0, neg10)->saturate = true; bld.ADD(dest, src0, neg10)->saturate = true;
bld.CMP(bld.null_reg_f(), src0, pos10, BRW_CONDITIONAL_G); bld.CMP(bld.null_reg_f(), src0, pos10, BRW_CONDITIONAL_G);
@ -3071,10 +3071,10 @@ TEST_F(cmod_propagation_test, cmp_to_add_float_g)
TEST_F(cmod_propagation_test, cmp_to_add_float_le) TEST_F(cmod_propagation_test, cmp_to_add_float_le)
{ {
fs_reg dest = bld.vgrf(BRW_TYPE_F); brw_reg dest = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg neg10(brw_imm_f(-10.0f)); brw_reg neg10(brw_imm_f(-10.0f));
fs_reg pos10(brw_imm_f(10.0f)); brw_reg pos10(brw_imm_f(10.0f));
bld.ADD(dest, src0, neg10)->saturate = true; bld.ADD(dest, src0, neg10)->saturate = true;
bld.CMP(bld.null_reg_f(), src0, pos10, BRW_CONDITIONAL_LE); bld.CMP(bld.null_reg_f(), src0, pos10, BRW_CONDITIONAL_LE);
@ -3099,13 +3099,13 @@ TEST_F(cmod_propagation_test, cmp_to_add_float_le)
TEST_F(cmod_propagation_test, prop_across_sel) TEST_F(cmod_propagation_test, prop_across_sel)
{ {
fs_reg dest1 = bld.vgrf(BRW_TYPE_F); brw_reg dest1 = bld.vgrf(BRW_TYPE_F);
fs_reg dest2 = bld.vgrf(BRW_TYPE_F); brw_reg dest2 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
fs_reg src2 = bld.vgrf(BRW_TYPE_F); brw_reg src2 = bld.vgrf(BRW_TYPE_F);
fs_reg src3 = bld.vgrf(BRW_TYPE_F); brw_reg src3 = bld.vgrf(BRW_TYPE_F);
fs_reg zero(brw_imm_f(0.0f)); brw_reg zero(brw_imm_f(0.0f));
bld.ADD(dest1, src0, src1); bld.ADD(dest1, src0, src1);
bld.emit_minmax(dest2, src2, src3, BRW_CONDITIONAL_GE); bld.emit_minmax(dest2, src2, src3, BRW_CONDITIONAL_GE);
bld.CMP(bld.null_reg_f(), dest1, zero, BRW_CONDITIONAL_GE); bld.CMP(bld.null_reg_f(), dest1, zero, BRW_CONDITIONAL_GE);

14
src/intel/compiler/test_fs_combine_constants.cpp
View file

@ -77,9 +77,9 @@ TEST_F(FSCombineConstantsTest, Simple)
{ {
fs_builder bld = make_builder(shader); fs_builder bld = make_builder(shader);
fs_reg r = brw_vec8_grf(1, 0); brw_reg r = brw_vec8_grf(1, 0);
fs_reg imm_a = brw_imm_ud(1); brw_reg imm_a = brw_imm_ud(1);
fs_reg imm_b = brw_imm_ud(2); brw_reg imm_b = brw_imm_ud(2);
bld.SEL(r, imm_a, imm_b); bld.SEL(r, imm_a, imm_b);
shader->calculate_cfg(); shader->calculate_cfg();
@ -102,10 +102,10 @@ TEST_F(FSCombineConstantsTest, DoContainingDo)
{ {
fs_builder bld = make_builder(shader); fs_builder bld = make_builder(shader);
fs_reg r1 = brw_vec8_grf(1, 0); brw_reg r1 = brw_vec8_grf(1, 0);
fs_reg r2 = brw_vec8_grf(2, 0); brw_reg r2 = brw_vec8_grf(2, 0);
fs_reg imm_a = brw_imm_ud(1); brw_reg imm_a = brw_imm_ud(1);
fs_reg imm_b = brw_imm_ud(2); brw_reg imm_b = brw_imm_ud(2);
bld.DO(); bld.DO();
bld.DO(); bld.DO();

14
src/intel/compiler/test_fs_copy_propagation.cpp
View file

@ -119,10 +119,10 @@ copy_propagation(fs_visitor *v)
TEST_F(copy_propagation_test, basic) TEST_F(copy_propagation_test, basic)
{ {
fs_reg vgrf0 = bld.vgrf(BRW_TYPE_F); brw_reg vgrf0 = bld.vgrf(BRW_TYPE_F);
fs_reg vgrf1 = bld.vgrf(BRW_TYPE_F); brw_reg vgrf1 = bld.vgrf(BRW_TYPE_F);
fs_reg vgrf2 = bld.vgrf(BRW_TYPE_F); brw_reg vgrf2 = bld.vgrf(BRW_TYPE_F);
fs_reg vgrf3 = bld.vgrf(BRW_TYPE_F); brw_reg vgrf3 = bld.vgrf(BRW_TYPE_F);
bld.MOV(vgrf0, vgrf2); bld.MOV(vgrf0, vgrf2);
bld.ADD(vgrf1, vgrf0, vgrf3); bld.ADD(vgrf1, vgrf0, vgrf3);
@ -160,9 +160,9 @@ TEST_F(copy_propagation_test, basic)
TEST_F(copy_propagation_test, maxmax_sat_imm) TEST_F(copy_propagation_test, maxmax_sat_imm)
{ {
fs_reg vgrf0 = bld.vgrf(BRW_TYPE_F); brw_reg vgrf0 = bld.vgrf(BRW_TYPE_F);
fs_reg vgrf1 = bld.vgrf(BRW_TYPE_F); brw_reg vgrf1 = bld.vgrf(BRW_TYPE_F);
fs_reg vgrf2 = bld.vgrf(BRW_TYPE_F); brw_reg vgrf2 = bld.vgrf(BRW_TYPE_F);
static const struct { static const struct {
enum brw_conditional_mod conditional_mod; enum brw_conditional_mod conditional_mod;

10
src/intel/compiler/test_fs_cse.cpp
View file

@ -102,11 +102,11 @@ cse(fs_visitor *v)
TEST_F(cse_test, add3_invalid) TEST_F(cse_test, add3_invalid)
{ {
fs_reg dst0 = bld.null_reg_d(); brw_reg dst0 = bld.null_reg_d();
fs_reg src0 = bld.vgrf(BRW_TYPE_D); brw_reg src0 = bld.vgrf(BRW_TYPE_D);
fs_reg src1 = bld.vgrf(BRW_TYPE_D); brw_reg src1 = bld.vgrf(BRW_TYPE_D);
fs_reg src2 = bld.vgrf(BRW_TYPE_D); brw_reg src2 = bld.vgrf(BRW_TYPE_D);
fs_reg src3 = bld.vgrf(BRW_TYPE_D); brw_reg src3 = bld.vgrf(BRW_TYPE_D);
bld.ADD3(dst0, src0, src1, src2) bld.ADD3(dst0, src0, src1, src2)
->conditional_mod = BRW_CONDITIONAL_NZ; ->conditional_mod = BRW_CONDITIONAL_NZ;

172
src/intel/compiler/test_fs_saturate_propagation.cpp
View file

@ -120,10 +120,10 @@ saturate_propagation(fs_visitor *v)
TEST_F(saturate_propagation_test, basic) TEST_F(saturate_propagation_test, basic)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_F); brw_reg dst0 = bld.vgrf(BRW_TYPE_F);
fs_reg dst1 = bld.vgrf(BRW_TYPE_F); brw_reg dst1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
bld.ADD(dst0, src0, src1); bld.ADD(dst0, src0, src1);
set_saturate(true, bld.MOV(dst1, dst0)); set_saturate(true, bld.MOV(dst1, dst0));
@ -154,11 +154,11 @@ TEST_F(saturate_propagation_test, basic)
TEST_F(saturate_propagation_test, other_non_saturated_use) TEST_F(saturate_propagation_test, other_non_saturated_use)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_F); brw_reg dst0 = bld.vgrf(BRW_TYPE_F);
fs_reg dst1 = bld.vgrf(BRW_TYPE_F); brw_reg dst1 = bld.vgrf(BRW_TYPE_F);
fs_reg dst2 = bld.vgrf(BRW_TYPE_F); brw_reg dst2 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
bld.ADD(dst0, src0, src1); bld.ADD(dst0, src0, src1);
set_saturate(true, bld.MOV(dst1, dst0)); set_saturate(true, bld.MOV(dst1, dst0));
bld.ADD(dst2, dst0, src0); bld.ADD(dst2, dst0, src0);
@ -191,10 +191,10 @@ TEST_F(saturate_propagation_test, other_non_saturated_use)
TEST_F(saturate_propagation_test, predicated_instruction) TEST_F(saturate_propagation_test, predicated_instruction)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_F); brw_reg dst0 = bld.vgrf(BRW_TYPE_F);
fs_reg dst1 = bld.vgrf(BRW_TYPE_F); brw_reg dst1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
bld.ADD(dst0, src0, src1) bld.ADD(dst0, src0, src1)
->predicate = BRW_PREDICATE_NORMAL; ->predicate = BRW_PREDICATE_NORMAL;
set_saturate(true, bld.MOV(dst1, dst0)); set_saturate(true, bld.MOV(dst1, dst0));
@ -225,9 +225,9 @@ TEST_F(saturate_propagation_test, predicated_instruction)
TEST_F(saturate_propagation_test, neg_mov_sat) TEST_F(saturate_propagation_test, neg_mov_sat)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_F); brw_reg dst0 = bld.vgrf(BRW_TYPE_F);
fs_reg dst1 = bld.vgrf(BRW_TYPE_F); brw_reg dst1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
bld.RNDU(dst0, src0); bld.RNDU(dst0, src0);
dst0.negate = true; dst0.negate = true;
set_saturate(true, bld.MOV(dst1, dst0)); set_saturate(true, bld.MOV(dst1, dst0));
@ -258,10 +258,10 @@ TEST_F(saturate_propagation_test, neg_mov_sat)
TEST_F(saturate_propagation_test, add_neg_mov_sat) TEST_F(saturate_propagation_test, add_neg_mov_sat)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_F); brw_reg dst0 = bld.vgrf(BRW_TYPE_F);
fs_reg dst1 = bld.vgrf(BRW_TYPE_F); brw_reg dst1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
bld.ADD(dst0, src0, src1); bld.ADD(dst0, src0, src1);
dst0.negate = true; dst0.negate = true;
set_saturate(true, bld.MOV(dst1, dst0)); set_saturate(true, bld.MOV(dst1, dst0));
@ -295,10 +295,10 @@ TEST_F(saturate_propagation_test, add_neg_mov_sat)
TEST_F(saturate_propagation_test, add_imm_float_neg_mov_sat) TEST_F(saturate_propagation_test, add_imm_float_neg_mov_sat)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_F); brw_reg dst0 = bld.vgrf(BRW_TYPE_F);
fs_reg dst1 = bld.vgrf(BRW_TYPE_F); brw_reg dst1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = brw_imm_f(1.0f); brw_reg src1 = brw_imm_f(1.0f);
bld.ADD(dst0, src0, src1); bld.ADD(dst0, src0, src1);
dst0.negate = true; dst0.negate = true;
set_saturate(true, bld.MOV(dst1, dst0)); set_saturate(true, bld.MOV(dst1, dst0));
@ -332,10 +332,10 @@ TEST_F(saturate_propagation_test, add_imm_float_neg_mov_sat)
TEST_F(saturate_propagation_test, mul_neg_mov_sat) TEST_F(saturate_propagation_test, mul_neg_mov_sat)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_F); brw_reg dst0 = bld.vgrf(BRW_TYPE_F);
fs_reg dst1 = bld.vgrf(BRW_TYPE_F); brw_reg dst1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
bld.MUL(dst0, src0, src1); bld.MUL(dst0, src0, src1);
dst0.negate = true; dst0.negate = true;
set_saturate(true, bld.MOV(dst1, dst0)); set_saturate(true, bld.MOV(dst1, dst0));
@ -369,11 +369,11 @@ TEST_F(saturate_propagation_test, mul_neg_mov_sat)
TEST_F(saturate_propagation_test, mad_neg_mov_sat) TEST_F(saturate_propagation_test, mad_neg_mov_sat)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_F); brw_reg dst0 = bld.vgrf(BRW_TYPE_F);
fs_reg dst1 = bld.vgrf(BRW_TYPE_F); brw_reg dst1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
fs_reg src2 = bld.vgrf(BRW_TYPE_F); brw_reg src2 = bld.vgrf(BRW_TYPE_F);
bld.MAD(dst0, src0, src1, src2); bld.MAD(dst0, src0, src1, src2);
dst0.negate = true; dst0.negate = true;
set_saturate(true, bld.MOV(dst1, dst0)); set_saturate(true, bld.MOV(dst1, dst0));
@ -409,11 +409,11 @@ TEST_F(saturate_propagation_test, mad_neg_mov_sat)
TEST_F(saturate_propagation_test, mad_imm_float_neg_mov_sat) TEST_F(saturate_propagation_test, mad_imm_float_neg_mov_sat)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_F); brw_reg dst0 = bld.vgrf(BRW_TYPE_F);
fs_reg dst1 = bld.vgrf(BRW_TYPE_F); brw_reg dst1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = brw_imm_f(1.0f); brw_reg src0 = brw_imm_f(1.0f);
fs_reg src1 = brw_imm_f(-2.0f); brw_reg src1 = brw_imm_f(-2.0f);
fs_reg src2 = bld.vgrf(BRW_TYPE_F); brw_reg src2 = bld.vgrf(BRW_TYPE_F);
/* The builder for MAD tries to be helpful and not put immediates as direct /* The builder for MAD tries to be helpful and not put immediates as direct
* sources. We want to test specifically that case. * sources. We want to test specifically that case.
*/ */
@ -453,11 +453,11 @@ TEST_F(saturate_propagation_test, mad_imm_float_neg_mov_sat)
TEST_F(saturate_propagation_test, mul_mov_sat_neg_mov_sat) TEST_F(saturate_propagation_test, mul_mov_sat_neg_mov_sat)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_F); brw_reg dst0 = bld.vgrf(BRW_TYPE_F);
fs_reg dst1 = bld.vgrf(BRW_TYPE_F); brw_reg dst1 = bld.vgrf(BRW_TYPE_F);
fs_reg dst2 = bld.vgrf(BRW_TYPE_F); brw_reg dst2 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
bld.MUL(dst0, src0, src1); bld.MUL(dst0, src0, src1);
set_saturate(true, bld.MOV(dst1, dst0)); set_saturate(true, bld.MOV(dst1, dst0));
dst0.negate = true; dst0.negate = true;
@ -494,11 +494,11 @@ TEST_F(saturate_propagation_test, mul_mov_sat_neg_mov_sat)
TEST_F(saturate_propagation_test, mul_neg_mov_sat_neg_mov_sat) TEST_F(saturate_propagation_test, mul_neg_mov_sat_neg_mov_sat)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_F); brw_reg dst0 = bld.vgrf(BRW_TYPE_F);
fs_reg dst1 = bld.vgrf(BRW_TYPE_F); brw_reg dst1 = bld.vgrf(BRW_TYPE_F);
fs_reg dst2 = bld.vgrf(BRW_TYPE_F); brw_reg dst2 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
bld.MUL(dst0, src0, src1); bld.MUL(dst0, src0, src1);
dst0.negate = true; dst0.negate = true;
set_saturate(true, bld.MOV(dst1, dst0)); set_saturate(true, bld.MOV(dst1, dst0));
@ -536,10 +536,10 @@ TEST_F(saturate_propagation_test, mul_neg_mov_sat_neg_mov_sat)
TEST_F(saturate_propagation_test, abs_mov_sat) TEST_F(saturate_propagation_test, abs_mov_sat)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_F); brw_reg dst0 = bld.vgrf(BRW_TYPE_F);
fs_reg dst1 = bld.vgrf(BRW_TYPE_F); brw_reg dst1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
bld.ADD(dst0, src0, src1); bld.ADD(dst0, src0, src1);
dst0.abs = true; dst0.abs = true;
set_saturate(true, bld.MOV(dst1, dst0)); set_saturate(true, bld.MOV(dst1, dst0));
@ -570,11 +570,11 @@ TEST_F(saturate_propagation_test, abs_mov_sat)
TEST_F(saturate_propagation_test, producer_saturates) TEST_F(saturate_propagation_test, producer_saturates)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_F); brw_reg dst0 = bld.vgrf(BRW_TYPE_F);
fs_reg dst1 = bld.vgrf(BRW_TYPE_F); brw_reg dst1 = bld.vgrf(BRW_TYPE_F);
fs_reg dst2 = bld.vgrf(BRW_TYPE_F); brw_reg dst2 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
set_saturate(true, bld.ADD(dst0, src0, src1)); set_saturate(true, bld.ADD(dst0, src0, src1));
set_saturate(true, bld.MOV(dst1, dst0)); set_saturate(true, bld.MOV(dst1, dst0));
bld.MOV(dst2, dst0); bld.MOV(dst2, dst0);
@ -608,11 +608,11 @@ TEST_F(saturate_propagation_test, producer_saturates)
TEST_F(saturate_propagation_test, intervening_saturating_copy) TEST_F(saturate_propagation_test, intervening_saturating_copy)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_F); brw_reg dst0 = bld.vgrf(BRW_TYPE_F);
fs_reg dst1 = bld.vgrf(BRW_TYPE_F); brw_reg dst1 = bld.vgrf(BRW_TYPE_F);
fs_reg dst2 = bld.vgrf(BRW_TYPE_F); brw_reg dst2 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
bld.ADD(dst0, src0, src1); bld.ADD(dst0, src0, src1);
set_saturate(true, bld.MOV(dst1, dst0)); set_saturate(true, bld.MOV(dst1, dst0));
set_saturate(true, bld.MOV(dst2, dst0)); set_saturate(true, bld.MOV(dst2, dst0));
@ -648,13 +648,13 @@ TEST_F(saturate_propagation_test, intervening_saturating_copy)
TEST_F(saturate_propagation_test, intervening_dest_write) TEST_F(saturate_propagation_test, intervening_dest_write)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_F, 4); brw_reg dst0 = bld.vgrf(BRW_TYPE_F, 4);
fs_reg dst1 = bld.vgrf(BRW_TYPE_F); brw_reg dst1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
fs_reg src2 = bld.vgrf(BRW_TYPE_F, 2); brw_reg src2 = bld.vgrf(BRW_TYPE_F, 2);
fs_reg tex_srcs[TEX_LOGICAL_NUM_SRCS]; brw_reg tex_srcs[TEX_LOGICAL_NUM_SRCS];
tex_srcs[TEX_LOGICAL_SRC_COORDINATE] = src2; tex_srcs[TEX_LOGICAL_SRC_COORDINATE] = src2;
tex_srcs[TEX_LOGICAL_SRC_SURFACE] = brw_imm_ud(0); tex_srcs[TEX_LOGICAL_SRC_SURFACE] = brw_imm_ud(0);
tex_srcs[TEX_LOGICAL_SRC_COORD_COMPONENTS] = brw_imm_ud(2); tex_srcs[TEX_LOGICAL_SRC_COORD_COMPONENTS] = brw_imm_ud(2);
@ -695,11 +695,11 @@ TEST_F(saturate_propagation_test, intervening_dest_write)
TEST_F(saturate_propagation_test, mul_neg_mov_sat_mov_sat) TEST_F(saturate_propagation_test, mul_neg_mov_sat_mov_sat)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_F); brw_reg dst0 = bld.vgrf(BRW_TYPE_F);
fs_reg dst1 = bld.vgrf(BRW_TYPE_F); brw_reg dst1 = bld.vgrf(BRW_TYPE_F);
fs_reg dst2 = bld.vgrf(BRW_TYPE_F); brw_reg dst2 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
bld.MUL(dst0, src0, src1); bld.MUL(dst0, src0, src1);
dst0.negate = true; dst0.negate = true;
set_saturate(true, bld.MOV(dst1, dst0)); set_saturate(true, bld.MOV(dst1, dst0));
@ -737,10 +737,10 @@ TEST_F(saturate_propagation_test, mul_neg_mov_sat_mov_sat)
TEST_F(saturate_propagation_test, smaller_exec_size_consumer) TEST_F(saturate_propagation_test, smaller_exec_size_consumer)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_F); brw_reg dst0 = bld.vgrf(BRW_TYPE_F);
fs_reg dst1 = bld.vgrf(BRW_TYPE_F); brw_reg dst1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
bld.ADD(dst0, src0, src1); bld.ADD(dst0, src0, src1);
set_saturate(true, bld.group(8, 0).MOV(dst1, dst0)); set_saturate(true, bld.group(8, 0).MOV(dst1, dst0));
@ -770,10 +770,10 @@ TEST_F(saturate_propagation_test, smaller_exec_size_consumer)
TEST_F(saturate_propagation_test, larger_exec_size_consumer) TEST_F(saturate_propagation_test, larger_exec_size_consumer)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_F); brw_reg dst0 = bld.vgrf(BRW_TYPE_F);
fs_reg dst1 = bld.vgrf(BRW_TYPE_F); brw_reg dst1 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
bld.group(8, 0).ADD(dst0, src0, src1); bld.group(8, 0).ADD(dst0, src0, src1);
set_saturate(true, bld.MOV(dst1, dst0)); set_saturate(true, bld.MOV(dst1, dst0));
@ -803,11 +803,11 @@ TEST_F(saturate_propagation_test, larger_exec_size_consumer)
TEST_F(saturate_propagation_test, offset_source_barrier) TEST_F(saturate_propagation_test, offset_source_barrier)
{ {
fs_reg dst0 = bld.vgrf(BRW_TYPE_F); brw_reg dst0 = bld.vgrf(BRW_TYPE_F);
fs_reg dst1 = bld.vgrf(BRW_TYPE_F); brw_reg dst1 = bld.vgrf(BRW_TYPE_F);
fs_reg dst2 = bld.vgrf(BRW_TYPE_F); brw_reg dst2 = bld.vgrf(BRW_TYPE_F);
fs_reg src0 = bld.vgrf(BRW_TYPE_F); brw_reg src0 = bld.vgrf(BRW_TYPE_F);
fs_reg src1 = bld.vgrf(BRW_TYPE_F); brw_reg src1 = bld.vgrf(BRW_TYPE_F);
bld.group(16, 0).ADD(dst0, src0, src1); bld.group(16, 0).ADD(dst0, src0, src1);
bld.group(1, 0).ADD(dst1, component(dst0, 8), brw_imm_f(1.0f)); bld.group(1, 0).ADD(dst1, component(dst0, 8), brw_imm_f(1.0f));
set_saturate(true, bld.group(16, 0).MOV(dst2, dst0)); set_saturate(true, bld.group(16, 0).MOV(dst2, dst0));

108
src/intel/compiler/test_fs_scoreboard.cpp
View file

@ -106,8 +106,8 @@ lower_scoreboard(fs_visitor *v)
} }
fs_inst * fs_inst *
emit_SEND(const fs_builder &bld, const fs_reg &dst, emit_SEND(const fs_builder &bld, const brw_reg &dst,
const fs_reg &desc, const fs_reg &payload) const brw_reg &desc, const brw_reg &payload)
{ {
fs_inst *inst = bld.emit(SHADER_OPCODE_SEND, dst, desc, desc, payload); fs_inst *inst = bld.emit(SHADER_OPCODE_SEND, dst, desc, desc, payload);
inst->mlen = 1; inst->mlen = 1;
@ -150,12 +150,12 @@ std::ostream &operator<<(std::ostream &os, const tgl_swsb &swsb) {
TEST_F(scoreboard_test, RAW_inorder_inorder) TEST_F(scoreboard_test, RAW_inorder_inorder)
{ {
fs_reg g[16]; brw_reg g[16];
for (unsigned i = 0; i < ARRAY_SIZE(g); i++) for (unsigned i = 0; i < ARRAY_SIZE(g); i++)
g[i] = bld.vgrf(BRW_TYPE_D); g[i] = bld.vgrf(BRW_TYPE_D);
fs_reg x = bld.vgrf(BRW_TYPE_D); brw_reg x = bld.vgrf(BRW_TYPE_D);
fs_reg y = bld.vgrf(BRW_TYPE_D); brw_reg y = bld.vgrf(BRW_TYPE_D);
bld.ADD( x, g[1], g[2]); bld.ADD( x, g[1], g[2]);
bld.MUL( y, g[3], g[4]); bld.MUL( y, g[3], g[4]);
bld.AND(g[5], x, y); bld.AND(g[5], x, y);
@ -176,11 +176,11 @@ TEST_F(scoreboard_test, RAW_inorder_inorder)
TEST_F(scoreboard_test, RAW_inorder_outoforder) TEST_F(scoreboard_test, RAW_inorder_outoforder)
{ {
fs_reg g[16]; brw_reg g[16];
for (unsigned i = 0; i < ARRAY_SIZE(g); i++) for (unsigned i = 0; i < ARRAY_SIZE(g); i++)
g[i] = bld.vgrf(BRW_TYPE_D); g[i] = bld.vgrf(BRW_TYPE_D);
fs_reg x = bld.vgrf(BRW_TYPE_D); brw_reg x = bld.vgrf(BRW_TYPE_D);
bld.ADD( x, g[1], g[2]); bld.ADD( x, g[1], g[2]);
bld.MUL( g[3], g[4], g[5]); bld.MUL( g[3], g[4], g[5]);
emit_SEND(bld, g[6], g[7], x); emit_SEND(bld, g[6], g[7], x);
@ -201,12 +201,12 @@ TEST_F(scoreboard_test, RAW_inorder_outoforder)
TEST_F(scoreboard_test, RAW_outoforder_inorder) TEST_F(scoreboard_test, RAW_outoforder_inorder)
{ {
fs_reg g[16]; brw_reg g[16];
for (unsigned i = 0; i < ARRAY_SIZE(g); i++) for (unsigned i = 0; i < ARRAY_SIZE(g); i++)
g[i] = bld.vgrf(BRW_TYPE_D); g[i] = bld.vgrf(BRW_TYPE_D);
fs_reg x = bld.vgrf(BRW_TYPE_D); brw_reg x = bld.vgrf(BRW_TYPE_D);
fs_reg y = bld.vgrf(BRW_TYPE_D); brw_reg y = bld.vgrf(BRW_TYPE_D);
emit_SEND(bld, x, g[1], g[2]); emit_SEND(bld, x, g[1], g[2]);
bld.MUL( y, g[3], g[4]); bld.MUL( y, g[3], g[4]);
bld.AND( g[5], x, y); bld.AND( g[5], x, y);
@ -227,7 +227,7 @@ TEST_F(scoreboard_test, RAW_outoforder_inorder)
TEST_F(scoreboard_test, RAW_outoforder_outoforder) TEST_F(scoreboard_test, RAW_outoforder_outoforder)
{ {
fs_reg g[16]; brw_reg g[16];
for (unsigned i = 0; i < ARRAY_SIZE(g); i++) for (unsigned i = 0; i < ARRAY_SIZE(g); i++)
g[i] = bld.vgrf(BRW_TYPE_D); g[i] = bld.vgrf(BRW_TYPE_D);
@ -235,7 +235,7 @@ TEST_F(scoreboard_test, RAW_outoforder_outoforder)
* SBIDs. Since it is not possible we expect a SYNC instruction to be * SBIDs. Since it is not possible we expect a SYNC instruction to be
* added. * added.
*/ */
fs_reg x = bld.vgrf(BRW_TYPE_D); brw_reg x = bld.vgrf(BRW_TYPE_D);
emit_SEND(bld, x, g[1], g[2]); emit_SEND(bld, x, g[1], g[2]);
emit_SEND(bld, g[3], x, g[4])->sfid++; emit_SEND(bld, g[3], x, g[4])->sfid++;
@ -259,11 +259,11 @@ TEST_F(scoreboard_test, RAW_outoforder_outoforder)
TEST_F(scoreboard_test, WAR_inorder_inorder) TEST_F(scoreboard_test, WAR_inorder_inorder)
{ {
fs_reg g[16]; brw_reg g[16];
for (unsigned i = 0; i < ARRAY_SIZE(g); i++) for (unsigned i = 0; i < ARRAY_SIZE(g); i++)
g[i] = bld.vgrf(BRW_TYPE_D); g[i] = bld.vgrf(BRW_TYPE_D);
fs_reg x = bld.vgrf(BRW_TYPE_D); brw_reg x = bld.vgrf(BRW_TYPE_D);
bld.ADD(g[1], x, g[2]); bld.ADD(g[1], x, g[2]);
bld.MUL(g[3], g[4], g[5]); bld.MUL(g[3], g[4], g[5]);
bld.AND( x, g[6], g[7]); bld.AND( x, g[6], g[7]);
@ -284,11 +284,11 @@ TEST_F(scoreboard_test, WAR_inorder_inorder)
TEST_F(scoreboard_test, WAR_inorder_outoforder) TEST_F(scoreboard_test, WAR_inorder_outoforder)
{ {
fs_reg g[16]; brw_reg g[16];
for (unsigned i = 0; i < ARRAY_SIZE(g); i++) for (unsigned i = 0; i < ARRAY_SIZE(g); i++)
g[i] = bld.vgrf(BRW_TYPE_D); g[i] = bld.vgrf(BRW_TYPE_D);
fs_reg x = bld.vgrf(BRW_TYPE_D); brw_reg x = bld.vgrf(BRW_TYPE_D);
bld.ADD( g[1], x, g[2]); bld.ADD( g[1], x, g[2]);
bld.MUL( g[3], g[4], g[5]); bld.MUL( g[3], g[4], g[5]);
emit_SEND(bld, x, g[6], g[7]); emit_SEND(bld, x, g[6], g[7]);
@ -309,11 +309,11 @@ TEST_F(scoreboard_test, WAR_inorder_outoforder)
TEST_F(scoreboard_test, WAR_outoforder_inorder) TEST_F(scoreboard_test, WAR_outoforder_inorder)
{ {
fs_reg g[16]; brw_reg g[16];
for (unsigned i = 0; i < ARRAY_SIZE(g); i++) for (unsigned i = 0; i < ARRAY_SIZE(g); i++)
g[i] = bld.vgrf(BRW_TYPE_D); g[i] = bld.vgrf(BRW_TYPE_D);
fs_reg x = bld.vgrf(BRW_TYPE_D); brw_reg x = bld.vgrf(BRW_TYPE_D);
emit_SEND(bld, g[1], g[2], x); emit_SEND(bld, g[1], g[2], x);
bld.MUL( g[4], g[5], g[6]); bld.MUL( g[4], g[5], g[6]);
bld.AND( x, g[7], g[8]); bld.AND( x, g[7], g[8]);
@ -334,11 +334,11 @@ TEST_F(scoreboard_test, WAR_outoforder_inorder)
TEST_F(scoreboard_test, WAR_outoforder_outoforder) TEST_F(scoreboard_test, WAR_outoforder_outoforder)
{ {
fs_reg g[16]; brw_reg g[16];
for (unsigned i = 0; i < ARRAY_SIZE(g); i++) for (unsigned i = 0; i < ARRAY_SIZE(g); i++)
g[i] = bld.vgrf(BRW_TYPE_D); g[i] = bld.vgrf(BRW_TYPE_D);
fs_reg x = bld.vgrf(BRW_TYPE_D); brw_reg x = bld.vgrf(BRW_TYPE_D);
emit_SEND(bld, g[1], g[2], x); emit_SEND(bld, g[1], g[2], x);
emit_SEND(bld, x, g[3], g[4])->sfid++; emit_SEND(bld, x, g[3], g[4])->sfid++;
@ -362,11 +362,11 @@ TEST_F(scoreboard_test, WAR_outoforder_outoforder)
TEST_F(scoreboard_test, WAW_inorder_inorder) TEST_F(scoreboard_test, WAW_inorder_inorder)
{ {
fs_reg g[16]; brw_reg g[16];
for (unsigned i = 0; i < ARRAY_SIZE(g); i++) for (unsigned i = 0; i < ARRAY_SIZE(g); i++)
g[i] = bld.vgrf(BRW_TYPE_D); g[i] = bld.vgrf(BRW_TYPE_D);
fs_reg x = bld.vgrf(BRW_TYPE_D); brw_reg x = bld.vgrf(BRW_TYPE_D);
bld.ADD( x, g[1], g[2]); bld.ADD( x, g[1], g[2]);
bld.MUL(g[3], g[4], g[5]); bld.MUL(g[3], g[4], g[5]);
bld.AND( x, g[6], g[7]); bld.AND( x, g[6], g[7]);
@ -392,11 +392,11 @@ TEST_F(scoreboard_test, WAW_inorder_inorder)
TEST_F(scoreboard_test, WAW_inorder_outoforder) TEST_F(scoreboard_test, WAW_inorder_outoforder)
{ {
fs_reg g[16]; brw_reg g[16];
for (unsigned i = 0; i < ARRAY_SIZE(g); i++) for (unsigned i = 0; i < ARRAY_SIZE(g); i++)
g[i] = bld.vgrf(BRW_TYPE_D); g[i] = bld.vgrf(BRW_TYPE_D);
fs_reg x = bld.vgrf(BRW_TYPE_D); brw_reg x = bld.vgrf(BRW_TYPE_D);
bld.ADD( x, g[1], g[2]); bld.ADD( x, g[1], g[2]);
bld.MUL( g[3], g[4], g[5]); bld.MUL( g[3], g[4], g[5]);
emit_SEND(bld, x, g[6], g[7]); emit_SEND(bld, x, g[6], g[7]);
@ -417,11 +417,11 @@ TEST_F(scoreboard_test, WAW_inorder_outoforder)
TEST_F(scoreboard_test, WAW_outoforder_inorder) TEST_F(scoreboard_test, WAW_outoforder_inorder)
{ {
fs_reg g[16]; brw_reg g[16];
for (unsigned i = 0; i < ARRAY_SIZE(g); i++) for (unsigned i = 0; i < ARRAY_SIZE(g); i++)
g[i] = bld.vgrf(BRW_TYPE_D); g[i] = bld.vgrf(BRW_TYPE_D);
fs_reg x = bld.vgrf(BRW_TYPE_D); brw_reg x = bld.vgrf(BRW_TYPE_D);
emit_SEND(bld, x, g[1], g[2]); emit_SEND(bld, x, g[1], g[2]);
bld.MUL( g[3], g[4], g[5]); bld.MUL( g[3], g[4], g[5]);
bld.AND( x, g[6], g[7]); bld.AND( x, g[6], g[7]);
@ -442,11 +442,11 @@ TEST_F(scoreboard_test, WAW_outoforder_inorder)
TEST_F(scoreboard_test, WAW_outoforder_outoforder) TEST_F(scoreboard_test, WAW_outoforder_outoforder)
{ {
fs_reg g[16]; brw_reg g[16];
for (unsigned i = 0; i < ARRAY_SIZE(g); i++) for (unsigned i = 0; i < ARRAY_SIZE(g); i++)
g[i] = bld.vgrf(BRW_TYPE_D); g[i] = bld.vgrf(BRW_TYPE_D);
fs_reg x = bld.vgrf(BRW_TYPE_D); brw_reg x = bld.vgrf(BRW_TYPE_D);
emit_SEND(bld, x, g[1], g[2]); emit_SEND(bld, x, g[1], g[2]);
emit_SEND(bld, x, g[3], g[4])->sfid++; emit_SEND(bld, x, g[3], g[4])->sfid++;
@ -471,11 +471,11 @@ TEST_F(scoreboard_test, WAW_outoforder_outoforder)
TEST_F(scoreboard_test, loop1) TEST_F(scoreboard_test, loop1)
{ {
fs_reg g[16]; brw_reg g[16];
for (unsigned i = 0; i < ARRAY_SIZE(g); i++) for (unsigned i = 0; i < ARRAY_SIZE(g); i++)
g[i] = bld.vgrf(BRW_TYPE_D); g[i] = bld.vgrf(BRW_TYPE_D);
fs_reg x = bld.vgrf(BRW_TYPE_D); brw_reg x = bld.vgrf(BRW_TYPE_D);
bld.XOR( x, g[1], g[2]); bld.XOR( x, g[1], g[2]);
bld.emit(BRW_OPCODE_DO); bld.emit(BRW_OPCODE_DO);
@ -501,11 +501,11 @@ TEST_F(scoreboard_test, loop1)
TEST_F(scoreboard_test, loop2) TEST_F(scoreboard_test, loop2)
{ {
fs_reg g[16]; brw_reg g[16];
for (unsigned i = 0; i < ARRAY_SIZE(g); i++) for (unsigned i = 0; i < ARRAY_SIZE(g); i++)
g[i] = bld.vgrf(BRW_TYPE_D); g[i] = bld.vgrf(BRW_TYPE_D);
fs_reg x = bld.vgrf(BRW_TYPE_D); brw_reg x = bld.vgrf(BRW_TYPE_D);
bld.XOR( x, g[1], g[2]); bld.XOR( x, g[1], g[2]);
bld.XOR(g[3], g[1], g[2]); bld.XOR(g[3], g[1], g[2]);
bld.XOR(g[4], g[1], g[2]); bld.XOR(g[4], g[1], g[2]);
@ -536,11 +536,11 @@ TEST_F(scoreboard_test, loop2)
TEST_F(scoreboard_test, loop3) TEST_F(scoreboard_test, loop3)
{ {
fs_reg g[16]; brw_reg g[16];
for (unsigned i = 0; i < ARRAY_SIZE(g); i++) for (unsigned i = 0; i < ARRAY_SIZE(g); i++)
g[i] = bld.vgrf(BRW_TYPE_D); g[i] = bld.vgrf(BRW_TYPE_D);
fs_reg x = bld.vgrf(BRW_TYPE_D); brw_reg x = bld.vgrf(BRW_TYPE_D);
bld.XOR( x, g[1], g[2]); bld.XOR( x, g[1], g[2]);
bld.emit(BRW_OPCODE_DO); bld.emit(BRW_OPCODE_DO);
@ -573,11 +573,11 @@ TEST_F(scoreboard_test, loop3)
TEST_F(scoreboard_test, conditional1) TEST_F(scoreboard_test, conditional1)
{ {
fs_reg g[16]; brw_reg g[16];
for (unsigned i = 0; i < ARRAY_SIZE(g); i++) for (unsigned i = 0; i < ARRAY_SIZE(g); i++)
g[i] = bld.vgrf(BRW_TYPE_D); g[i] = bld.vgrf(BRW_TYPE_D);
fs_reg x = bld.vgrf(BRW_TYPE_D); brw_reg x = bld.vgrf(BRW_TYPE_D);
bld.XOR( x, g[1], g[2]); bld.XOR( x, g[1], g[2]);
bld.emit(BRW_OPCODE_IF); bld.emit(BRW_OPCODE_IF);
@ -602,11 +602,11 @@ TEST_F(scoreboard_test, conditional1)
TEST_F(scoreboard_test, conditional2) TEST_F(scoreboard_test, conditional2)
{ {
fs_reg g[16]; brw_reg g[16];
for (unsigned i = 0; i < ARRAY_SIZE(g); i++) for (unsigned i = 0; i < ARRAY_SIZE(g); i++)
g[i] = bld.vgrf(BRW_TYPE_D); g[i] = bld.vgrf(BRW_TYPE_D);
fs_reg x = bld.vgrf(BRW_TYPE_D); brw_reg x = bld.vgrf(BRW_TYPE_D);
bld.XOR( x, g[1], g[2]); bld.XOR( x, g[1], g[2]);
bld.XOR(g[3], g[1], g[2]); bld.XOR(g[3], g[1], g[2]);
bld.XOR(g[4], g[1], g[2]); bld.XOR(g[4], g[1], g[2]);
@ -634,11 +634,11 @@ TEST_F(scoreboard_test, conditional2)
TEST_F(scoreboard_test, conditional3) TEST_F(scoreboard_test, conditional3)
{ {
fs_reg g[16]; brw_reg g[16];
for (unsigned i = 0; i < ARRAY_SIZE(g); i++) for (unsigned i = 0; i < ARRAY_SIZE(g); i++)
g[i] = bld.vgrf(BRW_TYPE_D); g[i] = bld.vgrf(BRW_TYPE_D);
fs_reg x = bld.vgrf(BRW_TYPE_D); brw_reg x = bld.vgrf(BRW_TYPE_D);
bld.XOR( x, g[1], g[2]); bld.XOR( x, g[1], g[2]);
bld.emit(BRW_OPCODE_IF); bld.emit(BRW_OPCODE_IF);
@ -666,11 +666,11 @@ TEST_F(scoreboard_test, conditional3)
TEST_F(scoreboard_test, conditional4) TEST_F(scoreboard_test, conditional4)
{ {
fs_reg g[16]; brw_reg g[16];
for (unsigned i = 0; i < ARRAY_SIZE(g); i++) for (unsigned i = 0; i < ARRAY_SIZE(g); i++)
g[i] = bld.vgrf(BRW_TYPE_D); g[i] = bld.vgrf(BRW_TYPE_D);
fs_reg x = bld.vgrf(BRW_TYPE_D); brw_reg x = bld.vgrf(BRW_TYPE_D);
bld.XOR( x, g[1], g[2]); bld.XOR( x, g[1], g[2]);
bld.emit(BRW_OPCODE_IF); bld.emit(BRW_OPCODE_IF);
@ -698,11 +698,11 @@ TEST_F(scoreboard_test, conditional4)
TEST_F(scoreboard_test, conditional5) TEST_F(scoreboard_test, conditional5)
{ {
fs_reg g[16]; brw_reg g[16];
for (unsigned i = 0; i < ARRAY_SIZE(g); i++) for (unsigned i = 0; i < ARRAY_SIZE(g); i++)
g[i] = bld.vgrf(BRW_TYPE_D); g[i] = bld.vgrf(BRW_TYPE_D);
fs_reg x = bld.vgrf(BRW_TYPE_D); brw_reg x = bld.vgrf(BRW_TYPE_D);
bld.XOR( x, g[1], g[2]); bld.XOR( x, g[1], g[2]);
bld.emit(BRW_OPCODE_IF); bld.emit(BRW_OPCODE_IF);
@ -735,11 +735,11 @@ TEST_F(scoreboard_test, conditional5)
TEST_F(scoreboard_test, conditional6) TEST_F(scoreboard_test, conditional6)
{ {
fs_reg g[16]; brw_reg g[16];
for (unsigned i = 0; i < ARRAY_SIZE(g); i++) for (unsigned i = 0; i < ARRAY_SIZE(g); i++)
g[i] = bld.vgrf(BRW_TYPE_D); g[i] = bld.vgrf(BRW_TYPE_D);
fs_reg x = bld.vgrf(BRW_TYPE_D); brw_reg x = bld.vgrf(BRW_TYPE_D);
bld.XOR( x, g[1], g[2]); bld.XOR( x, g[1], g[2]);
bld.emit(BRW_OPCODE_IF); bld.emit(BRW_OPCODE_IF);
@ -779,11 +779,11 @@ TEST_F(scoreboard_test, conditional6)
TEST_F(scoreboard_test, conditional7) TEST_F(scoreboard_test, conditional7)
{ {
fs_reg g[16]; brw_reg g[16];
for (unsigned i = 0; i < ARRAY_SIZE(g); i++) for (unsigned i = 0; i < ARRAY_SIZE(g); i++)
g[i] = bld.vgrf(BRW_TYPE_D); g[i] = bld.vgrf(BRW_TYPE_D);
fs_reg x = bld.vgrf(BRW_TYPE_D); brw_reg x = bld.vgrf(BRW_TYPE_D);
bld.XOR( x, g[1], g[2]); bld.XOR( x, g[1], g[2]);
bld.emit(BRW_OPCODE_IF); bld.emit(BRW_OPCODE_IF);
@ -823,11 +823,11 @@ TEST_F(scoreboard_test, conditional7)
TEST_F(scoreboard_test, conditional8) TEST_F(scoreboard_test, conditional8)
{ {
fs_reg g[16]; brw_reg g[16];
for (unsigned i = 0; i < ARRAY_SIZE(g); i++) for (unsigned i = 0; i < ARRAY_SIZE(g); i++)
g[i] = bld.vgrf(BRW_TYPE_D); g[i] = bld.vgrf(BRW_TYPE_D);
fs_reg x = bld.vgrf(BRW_TYPE_D); brw_reg x = bld.vgrf(BRW_TYPE_D);
bld.XOR( x, g[1], g[2]); bld.XOR( x, g[1], g[2]);
bld.XOR(g[3], g[1], g[2]); bld.XOR(g[3], g[1], g[2]);
bld.XOR(g[4], g[1], g[2]); bld.XOR(g[4], g[1], g[2]);
@ -871,10 +871,10 @@ TEST_F(scoreboard_test, gfx125_RaR_over_different_pipes)
devinfo->verx10 = 125; devinfo->verx10 = 125;
brw_init_isa_info(&compiler->isa, devinfo); brw_init_isa_info(&compiler->isa, devinfo);
fs_reg a = bld.vgrf(BRW_TYPE_D); brw_reg a = bld.vgrf(BRW_TYPE_D);
fs_reg b = bld.vgrf(BRW_TYPE_D); brw_reg b = bld.vgrf(BRW_TYPE_D);
fs_reg f = bld.vgrf(BRW_TYPE_F); brw_reg f = bld.vgrf(BRW_TYPE_F);
fs_reg x = bld.vgrf(BRW_TYPE_D); brw_reg x = bld.vgrf(BRW_TYPE_D);
bld.ADD(f, x, x); bld.ADD(f, x, x);
bld.ADD(a, x, x); bld.ADD(a, x, x);

30
src/intel/compiler/test_predicated_break.cpp
View file

@ -140,9 +140,9 @@ TEST_F(PredicatedBreakTest, TopBreakWithoutContinue)
fs_builder a = make_builder(shader_a); fs_builder a = make_builder(shader_a);
fs_builder b = make_builder(shader_b); fs_builder b = make_builder(shader_b);
fs_reg r1 = brw_vec8_grf(1, 0); brw_reg r1 = brw_vec8_grf(1, 0);
fs_reg r2 = brw_vec8_grf(2, 0); brw_reg r2 = brw_vec8_grf(2, 0);
fs_reg r3 = brw_vec8_grf(3, 0); brw_reg r3 = brw_vec8_grf(3, 0);
a.DO(); a.DO();
a.CMP(r1, r2, r3, BRW_CONDITIONAL_NZ); a.CMP(r1, r2, r3, BRW_CONDITIONAL_NZ);
@ -174,9 +174,9 @@ TEST_F(PredicatedBreakTest, TopBreakWithContinue)
fs_builder a = make_builder(shader_a); fs_builder a = make_builder(shader_a);
fs_builder b = make_builder(shader_b); fs_builder b = make_builder(shader_b);
fs_reg r1 = brw_vec8_grf(1, 0); brw_reg r1 = brw_vec8_grf(1, 0);
fs_reg r2 = brw_vec8_grf(2, 0); brw_reg r2 = brw_vec8_grf(2, 0);
fs_reg r3 = brw_vec8_grf(3, 0); brw_reg r3 = brw_vec8_grf(3, 0);
a.DO(); a.DO();
a.CMP(r1, r2, r3, BRW_CONDITIONAL_NZ); a.CMP(r1, r2, r3, BRW_CONDITIONAL_NZ);
@ -218,9 +218,9 @@ TEST_F(PredicatedBreakTest, DISABLED_BottomBreakWithoutContinue)
fs_builder a = make_builder(shader_a); fs_builder a = make_builder(shader_a);
fs_builder b = make_builder(shader_b); fs_builder b = make_builder(shader_b);
fs_reg r1 = brw_vec8_grf(1, 0); brw_reg r1 = brw_vec8_grf(1, 0);
fs_reg r2 = brw_vec8_grf(2, 0); brw_reg r2 = brw_vec8_grf(2, 0);
fs_reg r3 = brw_vec8_grf(3, 0); brw_reg r3 = brw_vec8_grf(3, 0);
a.DO(); a.DO();
a.ADD(r1, r2, r3); a.ADD(r1, r2, r3);
@ -256,9 +256,9 @@ TEST_F(PredicatedBreakTest, BottomBreakWithContinue)
fs_builder a = make_builder(shader_a); fs_builder a = make_builder(shader_a);
fs_builder b = make_builder(shader_b); fs_builder b = make_builder(shader_b);
fs_reg r1 = brw_vec8_grf(1, 0); brw_reg r1 = brw_vec8_grf(1, 0);
fs_reg r2 = brw_vec8_grf(2, 0); brw_reg r2 = brw_vec8_grf(2, 0);
fs_reg r3 = brw_vec8_grf(3, 0); brw_reg r3 = brw_vec8_grf(3, 0);
a.DO(); a.DO();
a.ADD(r1, r2, r3); a.ADD(r1, r2, r3);
@ -300,9 +300,9 @@ TEST_F(PredicatedBreakTest, TwoBreaks)
fs_builder a = make_builder(shader_a); fs_builder a = make_builder(shader_a);
fs_builder b = make_builder(shader_b); fs_builder b = make_builder(shader_b);
fs_reg r1 = brw_vec8_grf(1, 0); brw_reg r1 = brw_vec8_grf(1, 0);
fs_reg r2 = brw_vec8_grf(2, 0); brw_reg r2 = brw_vec8_grf(2, 0);
fs_reg r3 = brw_vec8_grf(3, 0); brw_reg r3 = brw_vec8_grf(3, 0);
a.DO(); a.DO();
a.ADD(r1, r2, r3); a.ADD(r1, r2, r3);