mirror of
https://gitlab.freedesktop.org/mesa/mesa.git
synced 2026-05-02 05:48:07 +02:00
mesa: Return SSA defs from PTN ALU helpers
Mostly a big simplifcation. Some noise on Haswell shader-db: total instructions in shared programs: 2978203 -> 2978161 (<.01%) instructions in affected programs: 9812 -> 9770 (-0.43%) helped: 61 HURT: 39 helped stats (abs) min: 1 max: 5 x̄: 1.44 x̃: 1 helped stats (rel) min: 0.27% max: 7.69% x̄: 1.76% x̃: 1.18% HURT stats (abs) min: 1 max: 4 x̄: 1.18 x̃: 1 HURT stats (rel) min: 0.55% max: 16.67% x̄: 4.49% x̃: 3.45% 95% mean confidence interval for instructions value: -0.71 -0.13 95% mean confidence interval for instructions %-change: -0.11% 1.46% Inconclusive result (%-change mean confidence interval includes 0). total cycles in shared programs: 45346214 -> 45346684 (<.01%) cycles in affected programs: 519970 -> 520440 (0.09%) helped: 157 HURT: 157 helped stats (abs) min: 2 max: 2970 x̄: 166.80 x̃: 6 helped stats (rel) min: 0.05% max: 40.38% x̄: 5.01% x̃: 1.42% HURT stats (abs) min: 2 max: 1922 x̄: 169.80 x̃: 10 HURT stats (rel) min: 0.04% max: 44.00% x̄: 6.28% x̃: 2.46% 95% mean confidence interval for cycles value: -49.93 52.92 95% mean confidence interval for cycles %-change: -0.49% 1.76% Inconclusive result (value mean confidence interval includes 0). Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io> Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com> Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23089>
This commit is contained in:
Alyssa Rosenzweig
Marge Bot
parent
5b835f1a08
commit
b55836a74d
1 changed files with 113 additions and 245 deletions
|
|
@ -67,58 +67,6 @@ struct ptn_compile {
|
|||
(unsigned[4]){ SWIZZLE_##X, SWIZZLE_##Y, SWIZZLE_##Z, SWIZZLE_##W }
|
||||
#define ptn_channel(b, src, ch) nir_channel(b, src, SWIZZLE_##ch)
|
||||
|
||||
static nir_ssa_def *
|
||||
ptn_src_for_dest(struct ptn_compile *c, nir_alu_dest *dest)
|
||||
{
|
||||
nir_builder *b = &c->build;
|
||||
|
||||
nir_alu_src src;
|
||||
memset(&src, 0, sizeof(src));
|
||||
|
||||
if (dest->dest.is_ssa)
|
||||
src.src = nir_src_for_ssa(&dest->dest.ssa);
|
||||
else {
|
||||
assert(!dest->dest.reg.indirect);
|
||||
src.src = nir_src_for_reg(dest->dest.reg.reg);
|
||||
src.src.reg.base_offset = dest->dest.reg.base_offset;
|
||||
}
|
||||
|
||||
for (int i = 0; i < 4; i++)
|
||||
src.swizzle[i] = i;
|
||||
|
||||
return nir_mov_alu(b, src, 4);
|
||||
}
|
||||
|
||||
static nir_alu_dest
|
||||
ptn_get_dest(struct ptn_compile *c, const struct prog_dst_register *prog_dst)
|
||||
{
|
||||
nir_alu_dest dest;
|
||||
|
||||
memset(&dest, 0, sizeof(dest));
|
||||
|
||||
switch (prog_dst->File) {
|
||||
case PROGRAM_TEMPORARY:
|
||||
dest.dest.reg.reg = c->temp_regs[prog_dst->Index];
|
||||
break;
|
||||
case PROGRAM_OUTPUT:
|
||||
dest.dest.reg.reg = c->output_regs[prog_dst->Index];
|
||||
break;
|
||||
case PROGRAM_ADDRESS:
|
||||
assert(prog_dst->Index == 0);
|
||||
dest.dest.reg.reg = c->addr_reg;
|
||||
break;
|
||||
case PROGRAM_UNDEFINED:
|
||||
break;
|
||||
}
|
||||
|
||||
dest.write_mask = prog_dst->WriteMask;
|
||||
dest.saturate = false;
|
||||
|
||||
assert(!prog_dst->RelAddr);
|
||||
|
||||
return dest;
|
||||
}
|
||||
|
||||
static nir_ssa_def *
|
||||
ptn_get_src(struct ptn_compile *c, const struct prog_src_register *prog_src)
|
||||
{
|
||||
|
|
@ -244,66 +192,21 @@ ptn_get_src(struct ptn_compile *c, const struct prog_src_register *prog_src)
|
|||
return def;
|
||||
}
|
||||
|
||||
static void
|
||||
ptn_alu(nir_builder *b, nir_op op, nir_alu_dest dest, nir_ssa_def **src)
|
||||
{
|
||||
unsigned num_srcs = nir_op_infos[op].num_inputs;
|
||||
nir_alu_instr *instr = nir_alu_instr_create(b->shader, op);
|
||||
unsigned i;
|
||||
|
||||
for (i = 0; i < num_srcs; i++)
|
||||
instr->src[i].src = nir_src_for_ssa(src[i]);
|
||||
|
||||
instr->dest = dest;
|
||||
nir_builder_instr_insert(b, &instr->instr);
|
||||
}
|
||||
|
||||
static void
|
||||
ptn_move_dest_masked(nir_builder *b, nir_alu_dest dest,
|
||||
nir_ssa_def *def, unsigned write_mask)
|
||||
{
|
||||
if (!(dest.write_mask & write_mask))
|
||||
return;
|
||||
|
||||
nir_alu_instr *mov = nir_alu_instr_create(b->shader, nir_op_mov);
|
||||
if (!mov)
|
||||
return;
|
||||
|
||||
mov->dest = dest;
|
||||
mov->dest.write_mask &= write_mask;
|
||||
mov->src[0].src = nir_src_for_ssa(def);
|
||||
for (unsigned i = def->num_components; i < 4; i++)
|
||||
mov->src[0].swizzle[i] = def->num_components - 1;
|
||||
nir_builder_instr_insert(b, &mov->instr);
|
||||
}
|
||||
|
||||
static void
|
||||
ptn_move_dest(nir_builder *b, nir_alu_dest dest, nir_ssa_def *def)
|
||||
{
|
||||
ptn_move_dest_masked(b, dest, def, WRITEMASK_XYZW);
|
||||
}
|
||||
|
||||
static void
|
||||
ptn_arl(nir_builder *b, nir_alu_dest dest, nir_ssa_def **src)
|
||||
{
|
||||
ptn_move_dest(b, dest, nir_f2i32(b, nir_ffloor(b, src[0])));
|
||||
}
|
||||
|
||||
/* EXP - Approximate Exponential Base 2
|
||||
* dst.x = 2^{\lfloor src.x\rfloor}
|
||||
* dst.y = src.x - \lfloor src.x\rfloor
|
||||
* dst.z = 2^{src.x}
|
||||
* dst.w = 1.0
|
||||
*/
|
||||
static void
|
||||
ptn_exp(nir_builder *b, nir_alu_dest dest, nir_ssa_def **src)
|
||||
static nir_ssa_def *
|
||||
ptn_exp(nir_builder *b, nir_ssa_def **src)
|
||||
{
|
||||
nir_ssa_def *srcx = ptn_channel(b, src[0], X);
|
||||
|
||||
ptn_move_dest_masked(b, dest, nir_fexp2(b, nir_ffloor(b, srcx)), WRITEMASK_X);
|
||||
ptn_move_dest_masked(b, dest, nir_fsub(b, srcx, nir_ffloor(b, srcx)), WRITEMASK_Y);
|
||||
ptn_move_dest_masked(b, dest, nir_fexp2(b, srcx), WRITEMASK_Z);
|
||||
ptn_move_dest_masked(b, dest, nir_imm_float(b, 1.0), WRITEMASK_W);
|
||||
return nir_vec4(b, nir_fexp2(b, nir_ffloor(b, srcx)),
|
||||
nir_fsub(b, srcx, nir_ffloor(b, srcx)),
|
||||
nir_fexp2(b, srcx),
|
||||
nir_imm_float(b, 1.0));
|
||||
}
|
||||
|
||||
/* LOG - Approximate Logarithm Base 2
|
||||
|
|
@ -312,19 +215,16 @@ ptn_exp(nir_builder *b, nir_alu_dest dest, nir_ssa_def **src)
|
|||
* dst.z = \log_2{|src.x|}
|
||||
* dst.w = 1.0
|
||||
*/
|
||||
static void
|
||||
ptn_log(nir_builder *b, nir_alu_dest dest, nir_ssa_def **src)
|
||||
static nir_ssa_def *
|
||||
ptn_log(nir_builder *b, nir_ssa_def **src)
|
||||
{
|
||||
nir_ssa_def *abs_srcx = nir_fabs(b, ptn_channel(b, src[0], X));
|
||||
nir_ssa_def *log2 = nir_flog2(b, abs_srcx);
|
||||
nir_ssa_def *floor_log2 = nir_ffloor(b, log2);
|
||||
|
||||
ptn_move_dest_masked(b, dest, floor_log2, WRITEMASK_X);
|
||||
ptn_move_dest_masked(b, dest,
|
||||
nir_fdiv(b, abs_srcx, nir_fexp2(b, floor_log2)),
|
||||
WRITEMASK_Y);
|
||||
ptn_move_dest_masked(b, dest, log2, WRITEMASK_Z);
|
||||
ptn_move_dest_masked(b, dest, nir_imm_float(b, 1.0), WRITEMASK_W);
|
||||
return nir_vec4(b, nir_ffloor(b, log2),
|
||||
nir_fdiv(b, abs_srcx, nir_fexp2(b, nir_ffloor(b, log2))),
|
||||
nir_flog2(b, abs_srcx),
|
||||
nir_imm_float(b, 1.0));
|
||||
}
|
||||
|
||||
/* DST - Distance Vector
|
||||
|
|
@ -333,13 +233,14 @@ ptn_log(nir_builder *b, nir_alu_dest dest, nir_ssa_def **src)
|
|||
* dst.z = src0.z
|
||||
* dst.w = src1.w
|
||||
*/
|
||||
static void
|
||||
ptn_dst(nir_builder *b, nir_alu_dest dest, nir_ssa_def **src)
|
||||
static nir_ssa_def *
|
||||
ptn_dst(nir_builder *b, nir_ssa_def **src)
|
||||
{
|
||||
ptn_move_dest_masked(b, dest, nir_imm_float(b, 1.0), WRITEMASK_X);
|
||||
ptn_move_dest_masked(b, dest, nir_fmul(b, src[0], src[1]), WRITEMASK_Y);
|
||||
ptn_move_dest_masked(b, dest, nir_mov(b, src[0]), WRITEMASK_Z);
|
||||
ptn_move_dest_masked(b, dest, nir_mov(b, src[1]), WRITEMASK_W);
|
||||
return nir_vec4(b, nir_imm_float(b, 1.0),
|
||||
nir_fmul(b, ptn_channel(b, src[0], Y),
|
||||
ptn_channel(b, src[1], Y)),
|
||||
ptn_channel(b, src[0], Z),
|
||||
ptn_channel(b, src[1], W));
|
||||
}
|
||||
|
||||
/* LIT - Light Coefficients
|
||||
|
|
@ -348,29 +249,24 @@ ptn_dst(nir_builder *b, nir_alu_dest dest, nir_ssa_def **src)
|
|||
* dst.z = (src.x > 0.0) ? max(src.y, 0.0)^{clamp(src.w, -128.0, 128.0))} : 0
|
||||
* dst.w = 1.0
|
||||
*/
|
||||
static void
|
||||
ptn_lit(nir_builder *b, nir_alu_dest dest, nir_ssa_def **src)
|
||||
static nir_ssa_def *
|
||||
ptn_lit(nir_builder *b, nir_ssa_def **src)
|
||||
{
|
||||
ptn_move_dest_masked(b, dest, nir_imm_float(b, 1.0), WRITEMASK_XW);
|
||||
nir_ssa_def *src0_y = ptn_channel(b, src[0], Y);
|
||||
nir_ssa_def *wclamp = nir_fmax(b, nir_fmin(b, ptn_channel(b, src[0], W),
|
||||
nir_imm_float(b, 128.0)),
|
||||
nir_imm_float(b, -128.0));
|
||||
nir_ssa_def *pow = nir_fpow(b, nir_fmax(b, src0_y, nir_imm_float(b, 0.0)),
|
||||
wclamp);
|
||||
|
||||
ptn_move_dest_masked(b, dest, nir_fmax(b, ptn_channel(b, src[0], X),
|
||||
nir_imm_float(b, 0.0)), WRITEMASK_Y);
|
||||
nir_ssa_def *z = nir_bcsel(b, nir_fle_imm(b, ptn_channel(b, src[0], X), 0.0),
|
||||
nir_imm_float(b, 0.0), pow);
|
||||
|
||||
if (dest.write_mask & WRITEMASK_Z) {
|
||||
nir_ssa_def *src0_y = ptn_channel(b, src[0], Y);
|
||||
nir_ssa_def *wclamp = nir_fmax(b, nir_fmin(b, ptn_channel(b, src[0], W),
|
||||
nir_imm_float(b, 128.0)),
|
||||
nir_imm_float(b, -128.0));
|
||||
nir_ssa_def *pow = nir_fpow(b, nir_fmax(b, src0_y, nir_imm_float(b, 0.0)),
|
||||
wclamp);
|
||||
|
||||
nir_ssa_def *z = nir_bcsel(b,
|
||||
nir_fle_imm(b, ptn_channel(b, src[0], X), 0.0),
|
||||
nir_imm_float(b, 0.0),
|
||||
pow);
|
||||
|
||||
ptn_move_dest_masked(b, dest, z, WRITEMASK_Z);
|
||||
}
|
||||
return nir_vec4(b, nir_imm_float(b, 1.0),
|
||||
nir_fmax(b, ptn_channel(b, src[0], X),
|
||||
nir_imm_float(b, 0.0)),
|
||||
z,
|
||||
nir_imm_float(b, 1.0));
|
||||
}
|
||||
|
||||
/* SCS - Sine Cosine
|
||||
|
|
@ -379,80 +275,28 @@ ptn_lit(nir_builder *b, nir_alu_dest dest, nir_ssa_def **src)
|
|||
* dst.z = 0.0
|
||||
* dst.w = 1.0
|
||||
*/
|
||||
static void
|
||||
ptn_scs(nir_builder *b, nir_alu_dest dest, nir_ssa_def **src)
|
||||
static nir_ssa_def *
|
||||
ptn_scs(nir_builder *b, nir_ssa_def **src)
|
||||
{
|
||||
ptn_move_dest_masked(b, dest, nir_fcos(b, ptn_channel(b, src[0], X)),
|
||||
WRITEMASK_X);
|
||||
ptn_move_dest_masked(b, dest, nir_fsin(b, ptn_channel(b, src[0], X)),
|
||||
WRITEMASK_Y);
|
||||
ptn_move_dest_masked(b, dest, nir_imm_float(b, 0.0), WRITEMASK_Z);
|
||||
ptn_move_dest_masked(b, dest, nir_imm_float(b, 1.0), WRITEMASK_W);
|
||||
return nir_vec4(b, nir_fcos(b, ptn_channel(b, src[0], X)),
|
||||
nir_fsin(b, ptn_channel(b, src[0], X)),
|
||||
nir_imm_float(b, 0.0),
|
||||
nir_imm_float(b, 1.0));
|
||||
}
|
||||
|
||||
static void
|
||||
ptn_slt(nir_builder *b, nir_alu_dest dest, nir_ssa_def **src)
|
||||
static nir_ssa_def *
|
||||
ptn_xpd(nir_builder *b, nir_ssa_def **src)
|
||||
{
|
||||
ptn_move_dest(b, dest, nir_slt(b, src[0], src[1]));
|
||||
}
|
||||
nir_ssa_def *vec =
|
||||
nir_fsub(b, nir_fmul(b, nir_swizzle(b, src[0], SWIZ(Y, Z, X, W), 3),
|
||||
nir_swizzle(b, src[1], SWIZ(Z, X, Y, W), 3)),
|
||||
nir_fmul(b, nir_swizzle(b, src[1], SWIZ(Y, Z, X, W), 3),
|
||||
nir_swizzle(b, src[0], SWIZ(Z, X, Y, W), 3)));
|
||||
|
||||
static void
|
||||
ptn_sge(nir_builder *b, nir_alu_dest dest, nir_ssa_def **src)
|
||||
{
|
||||
ptn_move_dest(b, dest, nir_sge(b, src[0], src[1]));
|
||||
}
|
||||
|
||||
static void
|
||||
ptn_xpd(nir_builder *b, nir_alu_dest dest, nir_ssa_def **src)
|
||||
{
|
||||
ptn_move_dest_masked(b, dest,
|
||||
nir_fsub(b,
|
||||
nir_fmul(b,
|
||||
nir_swizzle(b, src[0], SWIZ(Y, Z, X, W), 3),
|
||||
nir_swizzle(b, src[1], SWIZ(Z, X, Y, W), 3)),
|
||||
nir_fmul(b,
|
||||
nir_swizzle(b, src[1], SWIZ(Y, Z, X, W), 3),
|
||||
nir_swizzle(b, src[0], SWIZ(Z, X, Y, W), 3))),
|
||||
WRITEMASK_XYZ);
|
||||
ptn_move_dest_masked(b, dest, nir_imm_float(b, 1.0), WRITEMASK_W);
|
||||
}
|
||||
|
||||
static void
|
||||
ptn_dp2(nir_builder *b, nir_alu_dest dest, nir_ssa_def **src)
|
||||
{
|
||||
ptn_move_dest(b, dest, nir_fdot2(b, src[0], src[1]));
|
||||
}
|
||||
|
||||
static void
|
||||
ptn_dp3(nir_builder *b, nir_alu_dest dest, nir_ssa_def **src)
|
||||
{
|
||||
ptn_move_dest(b, dest, nir_fdot3(b, src[0], src[1]));
|
||||
}
|
||||
|
||||
static void
|
||||
ptn_dp4(nir_builder *b, nir_alu_dest dest, nir_ssa_def **src)
|
||||
{
|
||||
ptn_move_dest(b, dest, nir_fdot4(b, src[0], src[1]));
|
||||
}
|
||||
|
||||
static void
|
||||
ptn_dph(nir_builder *b, nir_alu_dest dest, nir_ssa_def **src)
|
||||
{
|
||||
ptn_move_dest(b, dest, nir_fdph(b, src[0], src[1]));
|
||||
}
|
||||
|
||||
static void
|
||||
ptn_cmp(nir_builder *b, nir_alu_dest dest, nir_ssa_def **src)
|
||||
{
|
||||
ptn_move_dest(b, dest, nir_bcsel(b,
|
||||
nir_flt_imm(b, src[0], 0.0),
|
||||
src[1], src[2]));
|
||||
}
|
||||
|
||||
static void
|
||||
ptn_lrp(nir_builder *b, nir_alu_dest dest, nir_ssa_def **src)
|
||||
{
|
||||
ptn_move_dest(b, dest, nir_flrp(b, src[2], src[1], src[0]));
|
||||
return nir_vec4(b, nir_channel(b, vec, 0),
|
||||
nir_channel(b, vec, 1),
|
||||
nir_channel(b, vec, 2),
|
||||
nir_imm_float(b, 1.0));
|
||||
}
|
||||
|
||||
static void
|
||||
|
|
@ -506,8 +350,8 @@ _mesa_texture_index_to_sampler_dim(gl_texture_index index, bool *is_array)
|
|||
unreachable("unknown texture target");
|
||||
}
|
||||
|
||||
static void
|
||||
ptn_tex(struct ptn_compile *c, nir_alu_dest dest, nir_ssa_def **src,
|
||||
static nir_ssa_def *
|
||||
ptn_tex(struct ptn_compile *c, nir_ssa_def **src,
|
||||
struct prog_instruction *prog_inst)
|
||||
{
|
||||
nir_builder *b = &c->build;
|
||||
|
|
@ -619,8 +463,7 @@ ptn_tex(struct ptn_compile *c, nir_alu_dest dest, nir_ssa_def **src,
|
|||
nir_ssa_dest_init(&instr->instr, &instr->dest, 4, 32);
|
||||
nir_builder_instr_insert(b, &instr->instr);
|
||||
|
||||
/* Resolve the writemask on the texture op. */
|
||||
ptn_move_dest(b, dest, &instr->dest.ssa);
|
||||
return &instr->dest.ssa;
|
||||
}
|
||||
|
||||
static const nir_op op_trans[MAX_OPCODE] = {
|
||||
|
|
@ -684,88 +527,86 @@ ptn_emit_instruction(struct ptn_compile *c, struct prog_instruction *prog_inst)
|
|||
for (i = 0; i < 3; i++) {
|
||||
src[i] = ptn_get_src(c, &prog_inst->SrcReg[i]);
|
||||
}
|
||||
nir_alu_dest dest = ptn_get_dest(c, &prog_inst->DstReg);
|
||||
|
||||
nir_ssa_def *dst = NULL;
|
||||
if (c->error)
|
||||
return;
|
||||
|
||||
switch (op) {
|
||||
case OPCODE_RSQ:
|
||||
ptn_move_dest(b, dest,
|
||||
nir_frsq(b, nir_fabs(b, ptn_channel(b, src[0], X))));
|
||||
dst = nir_frsq(b, nir_fabs(b, ptn_channel(b, src[0], X)));
|
||||
break;
|
||||
|
||||
case OPCODE_RCP:
|
||||
ptn_move_dest(b, dest, nir_frcp(b, ptn_channel(b, src[0], X)));
|
||||
dst = nir_frcp(b, ptn_channel(b, src[0], X));
|
||||
break;
|
||||
|
||||
case OPCODE_EX2:
|
||||
ptn_move_dest(b, dest, nir_fexp2(b, ptn_channel(b, src[0], X)));
|
||||
dst = nir_fexp2(b, ptn_channel(b, src[0], X));
|
||||
break;
|
||||
|
||||
case OPCODE_LG2:
|
||||
ptn_move_dest(b, dest, nir_flog2(b, ptn_channel(b, src[0], X)));
|
||||
dst = nir_flog2(b, ptn_channel(b, src[0], X));
|
||||
break;
|
||||
|
||||
case OPCODE_POW:
|
||||
ptn_move_dest(b, dest, nir_fpow(b,
|
||||
ptn_channel(b, src[0], X),
|
||||
ptn_channel(b, src[1], X)));
|
||||
dst = nir_fpow(b, ptn_channel(b, src[0], X), ptn_channel(b, src[1], X));
|
||||
break;
|
||||
|
||||
case OPCODE_COS:
|
||||
ptn_move_dest(b, dest, nir_fcos(b, ptn_channel(b, src[0], X)));
|
||||
dst = nir_fcos(b, ptn_channel(b, src[0], X));
|
||||
break;
|
||||
|
||||
case OPCODE_SIN:
|
||||
ptn_move_dest(b, dest, nir_fsin(b, ptn_channel(b, src[0], X)));
|
||||
dst = nir_fsin(b, ptn_channel(b, src[0], X));
|
||||
break;
|
||||
|
||||
case OPCODE_ARL:
|
||||
ptn_arl(b, dest, src);
|
||||
dst = nir_f2i32(b, nir_ffloor(b, src[0]));
|
||||
break;
|
||||
|
||||
case OPCODE_EXP:
|
||||
ptn_exp(b, dest, src);
|
||||
dst = ptn_exp(b, src);
|
||||
break;
|
||||
|
||||
case OPCODE_LOG:
|
||||
ptn_log(b, dest, src);
|
||||
dst = ptn_log(b, src);
|
||||
break;
|
||||
|
||||
case OPCODE_LRP:
|
||||
ptn_lrp(b, dest, src);
|
||||
dst = nir_flrp(b, src[2], src[1], src[0]);
|
||||
break;
|
||||
|
||||
case OPCODE_MAD:
|
||||
ptn_move_dest(b, dest, nir_fadd(b, nir_fmul(b, src[0], src[1]), src[2]));
|
||||
dst = nir_fadd(b, nir_fmul(b, src[0], src[1]), src[2]);
|
||||
break;
|
||||
|
||||
case OPCODE_DST:
|
||||
ptn_dst(b, dest, src);
|
||||
dst = ptn_dst(b, src);
|
||||
break;
|
||||
|
||||
case OPCODE_LIT:
|
||||
ptn_lit(b, dest, src);
|
||||
dst = ptn_lit(b, src);
|
||||
break;
|
||||
|
||||
case OPCODE_XPD:
|
||||
ptn_xpd(b, dest, src);
|
||||
dst = ptn_xpd(b, src);
|
||||
break;
|
||||
|
||||
case OPCODE_DP2:
|
||||
ptn_dp2(b, dest, src);
|
||||
dst = nir_fdot2(b, src[0], src[1]);
|
||||
break;
|
||||
|
||||
case OPCODE_DP3:
|
||||
ptn_dp3(b, dest, src);
|
||||
dst = nir_fdot3(b, src[0], src[1]);
|
||||
break;
|
||||
|
||||
case OPCODE_DP4:
|
||||
ptn_dp4(b, dest, src);
|
||||
dst = nir_fdot4(b, src[0], src[1]);
|
||||
break;
|
||||
|
||||
case OPCODE_DPH:
|
||||
ptn_dph(b, dest, src);
|
||||
dst = nir_fdph(b, src[0], src[1]);
|
||||
break;
|
||||
|
||||
case OPCODE_KIL:
|
||||
|
|
@ -773,19 +614,19 @@ ptn_emit_instruction(struct ptn_compile *c, struct prog_instruction *prog_inst)
|
|||
break;
|
||||
|
||||
case OPCODE_CMP:
|
||||
ptn_cmp(b, dest, src);
|
||||
dst = nir_bcsel(b, nir_flt_imm(b, src[0], 0.0), src[1], src[2]);
|
||||
break;
|
||||
|
||||
case OPCODE_SCS:
|
||||
ptn_scs(b, dest, src);
|
||||
dst = ptn_scs(b, src);
|
||||
break;
|
||||
|
||||
case OPCODE_SLT:
|
||||
ptn_slt(b, dest, src);
|
||||
dst = nir_slt(b, src[0], src[1]);
|
||||
break;
|
||||
|
||||
case OPCODE_SGE:
|
||||
ptn_sge(b, dest, src);
|
||||
dst = nir_sge(b, src[0], src[1]);
|
||||
break;
|
||||
|
||||
case OPCODE_TEX:
|
||||
|
|
@ -793,12 +634,12 @@ ptn_emit_instruction(struct ptn_compile *c, struct prog_instruction *prog_inst)
|
|||
case OPCODE_TXD:
|
||||
case OPCODE_TXL:
|
||||
case OPCODE_TXP:
|
||||
ptn_tex(c, dest, src, prog_inst);
|
||||
dst = ptn_tex(c, src, prog_inst);
|
||||
break;
|
||||
|
||||
case OPCODE_SWZ:
|
||||
/* Extended swizzles were already handled in ptn_get_src(). */
|
||||
ptn_alu(b, nir_op_mov, dest, src);
|
||||
dst = nir_build_alu_src_arr(b, nir_op_mov, src);
|
||||
break;
|
||||
|
||||
case OPCODE_NOP:
|
||||
|
|
@ -806,7 +647,7 @@ ptn_emit_instruction(struct ptn_compile *c, struct prog_instruction *prog_inst)
|
|||
|
||||
default:
|
||||
if (op_trans[op] != 0) {
|
||||
ptn_alu(b, op_trans[op], dest, src);
|
||||
dst = nir_build_alu_src_arr(b, op_trans[op], src);
|
||||
} else {
|
||||
fprintf(stderr, "unknown opcode: %s\n", _mesa_opcode_string(op));
|
||||
abort();
|
||||
|
|
@ -814,11 +655,38 @@ ptn_emit_instruction(struct ptn_compile *c, struct prog_instruction *prog_inst)
|
|||
break;
|
||||
}
|
||||
|
||||
if (prog_inst->Saturate) {
|
||||
assert(prog_inst->Saturate);
|
||||
assert(!dest.dest.is_ssa);
|
||||
ptn_move_dest(b, dest, nir_fsat(b, ptn_src_for_dest(c, &dest)));
|
||||
if (dst == NULL)
|
||||
return;
|
||||
|
||||
if (dst->num_components == 1)
|
||||
dst = nir_replicate(b, dst, 4);
|
||||
|
||||
assert(dst->num_components == 4);
|
||||
|
||||
if (prog_inst->Saturate)
|
||||
dst = nir_fsat(b, dst);
|
||||
|
||||
const struct prog_dst_register *prog_dst = &prog_inst->DstReg;
|
||||
assert(!prog_dst->RelAddr);
|
||||
|
||||
nir_register *reg = NULL;
|
||||
switch (prog_dst->File) {
|
||||
case PROGRAM_TEMPORARY:
|
||||
reg = c->temp_regs[prog_dst->Index];
|
||||
break;
|
||||
case PROGRAM_OUTPUT:
|
||||
reg = c->output_regs[prog_dst->Index];
|
||||
break;
|
||||
case PROGRAM_ADDRESS:
|
||||
assert(prog_dst->Index == 0);
|
||||
reg = c->addr_reg;
|
||||
break;
|
||||
case PROGRAM_UNDEFINED:
|
||||
return;
|
||||
}
|
||||
|
||||
assert(reg != NULL);
|
||||
nir_store_register(b, reg, dst, prog_dst->WriteMask);
|
||||
}
|
||||
|
||||
/**
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue