fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-17 20:28:05 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

radeonsi: remove dead get_frag_coord_from_pixel_coord optimization

Browse source 
It's applied unconditionally by ac_nir_lower_ps_early. It was probably
replaced by that at some point.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/41226>
This commit is contained in:
Marek Olšák 2026-04-25 14:25:04 -04:00 • committed by Marge Bot
parent c9c0dce948
commit b63f866c84
11 changed files with 1 additions and 81 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
src/amd/compiler/aco_shader_info.h
View file

@ -93,8 +93,6 @@ struct aco_ps_prolog_info {
bool force_linear_center_interp;
unsigned samplemask_log_ps_iter;
bool get_frag_coord_from_pixel_coord;
bool pixel_center_integer;
bool force_samplemask_to_helper_invocation;
unsigned num_interp_inputs;
unsigned colors_read;

27
src/amd/compiler/instruction_selection/aco_select_ps_prolog.cpp
View file

@ -167,32 +167,6 @@ overwrite_samplemask_arg(isel_context* ctx, const struct aco_ps_prolog_info* fin
is_helper_invoc);
}
}
void
overwrite_pos_xy_args(isel_context* ctx, const struct aco_ps_prolog_info* finfo)
{
if (!finfo->get_frag_coord_from_pixel_coord)
return;
Builder bld(ctx->program, ctx->block);
Temp pos_fixed_pt = get_arg(ctx, ctx->args->pos_fixed_pt);
for (unsigned i = 0; i < 2; i++) {
if (!ctx->args->frag_pos[i].used)
continue;
Temp t;
if (i)
t = bld.vop2(aco_opcode::v_lshrrev_b32, bld.def(v1), Operand::c32(16), pos_fixed_pt);
else
t = bld.vop2(aco_opcode::v_and_b32, bld.def(v1), Operand::c32(0xffff), pos_fixed_pt);
t = bld.vop1(aco_opcode::v_cvt_f32_u32, bld.def(v1), t);
if (!finfo->pixel_center_integer)
t = bld.vop2(aco_opcode::v_add_f32, bld.def(v1), Operand::c32(0x3f000000 /*0.5*/), t);
ctx->arg_temps[ctx->args->frag_pos[i].arg_index] = t;
}
}
void
passthrough_all_args(isel_context* ctx, std::vector<Operand>& regs)
@ -297,7 +271,6 @@ select_ps_prolog(Program* program, void* pinfo, ac_shader_config* config,
overwrite_interp_args(&ctx, finfo);
overwrite_samplemask_arg(&ctx, finfo);
overwrite_pos_xy_args(&ctx, finfo);
std::vector<Operand> regs;
passthrough_all_args(&ctx, regs);

7
src/gallium/drivers/radeonsi/gfx/si_shader.c
View file

@ -856,8 +856,7 @@ static void si_preprocess_nir(struct si_nir_shader_ctx *ctx)
.ps_iter_samples = key->ps.part.prolog.samplemask_log_ps_iter ?
(1 << key->ps.part.prolog.samplemask_log_ps_iter) :
(key->ps.part.prolog.force_persp_sample_interp ||
key->ps.part.prolog.force_linear_sample_interp ? 2 :
(key->ps.part.prolog.get_frag_coord_from_pixel_coord ? 1 : 0)),
key->ps.part.prolog.force_linear_sample_interp ? 2 : 0),
.fbfetch_is_1D = key->ps.mono.fbfetch_is_1D,
.fbfetch_layered = key->ps.mono.fbfetch_layered,
@ -1742,15 +1741,12 @@ static void si_get_ps_prolog_key(struct si_shader *shader, union si_shader_part_
key->ps_prolog.states.force_linear_center_interp ||
key->ps_prolog.states.bc_optimize_for_persp || key->ps_prolog.states.bc_optimize_for_linear ||
key->ps_prolog.states.samplemask_log_ps_iter ||
key->ps_prolog.states.get_frag_coord_from_pixel_coord ||
key->ps_prolog.states.force_samplemask_to_helper_invocation);
key->ps_prolog.fragcoord_usage_mask =
G_0286CC_POS_X_FLOAT_ENA(shader->config.spi_ps_input_ena) |
(G_0286CC_POS_Y_FLOAT_ENA(shader->config.spi_ps_input_ena) << 1) |
(G_0286CC_POS_Z_FLOAT_ENA(shader->config.spi_ps_input_ena) << 2) |
(G_0286CC_POS_W_FLOAT_ENA(shader->config.spi_ps_input_ena) << 3);
key->ps_prolog.pixel_center_integer = key->ps_prolog.fragcoord_usage_mask &&
shader->selector->info.base.fs.pixel_center_integer;
if (shader->key.ps.part.prolog.poly_stipple)
shader->info.uses_vmem_load_other = true;
@ -1852,7 +1848,6 @@ static bool si_need_ps_prolog(const union si_shader_part_key *key)
key->ps_prolog.states.bc_optimize_for_persp ||
key->ps_prolog.states.bc_optimize_for_linear || key->ps_prolog.states.poly_stipple ||
key->ps_prolog.states.samplemask_log_ps_iter ||
key->ps_prolog.states.get_frag_coord_from_pixel_coord ||
key->ps_prolog.states.force_samplemask_to_helper_invocation;
}

2
src/gallium/drivers/radeonsi/gfx/si_shader.h
View file

@ -536,7 +536,6 @@ struct si_ps_prolog_bits {
unsigned bc_optimize_for_persp : 1;
unsigned bc_optimize_for_linear : 1;
unsigned samplemask_log_ps_iter : 2;
unsigned get_frag_coord_from_pixel_coord : 1;
unsigned force_samplemask_to_helper_invocation : 1;
};
@ -566,7 +565,6 @@ union si_shader_part_key {
unsigned colors_read : 8; /* color input components read */
unsigned num_interp_inputs : 5; /* BCOLOR is at this location */
unsigned fragcoord_usage_mask : 4;
unsigned pixel_center_integer : 1;
unsigned wqm : 1;
uint8_t color_attr_index[2];
uint8_t color_interp[2]; /* AC_COLOR_INTERP_* */

2
src/gallium/drivers/radeonsi/gfx/si_shader_aco.c
View file

@ -255,8 +255,6 @@ si_aco_build_ps_prolog(struct aco_compiler_options *options,
.force_linear_center_interp = key->ps_prolog.states.force_linear_center_interp,
.samplemask_log_ps_iter = key->ps_prolog.states.samplemask_log_ps_iter,
.get_frag_coord_from_pixel_coord = key->ps_prolog.states.get_frag_coord_from_pixel_coord,
.pixel_center_integer = key->ps_prolog.pixel_center_integer,
.force_samplemask_to_helper_invocation = key->ps_prolog.states.force_samplemask_to_helper_invocation,
.num_interp_inputs = key->ps_prolog.num_interp_inputs,
.colors_read = key->ps_prolog.colors_read,

2
src/gallium/drivers/radeonsi/gfx/si_shader_binary.c
View file

@ -607,8 +607,6 @@ static void si_dump_shader_key(const struct si_shader *shader, FILE *f)
key->ps.part.prolog.bc_optimize_for_linear);
fprintf(f, " prolog.samplemask_log_ps_iter = %u\n",
key->ps.part.prolog.samplemask_log_ps_iter);
fprintf(f, " prolog.get_frag_coord_from_pixel_coord = %u\n",
key->ps.part.prolog.get_frag_coord_from_pixel_coord);
fprintf(f, " prolog.force_samplemask_to_helper_invocation = %u\n",
key->ps.part.prolog.force_samplemask_to_helper_invocation);
fprintf(f, " epilog.spi_shader_col_format = 0x%x\n",

4
src/gallium/drivers/radeonsi/gfx/si_shader_info.c
View file

@ -369,9 +369,6 @@ static void gather_instruction(const struct nir_shader *nir, struct si_shader_in
if (intr->intrinsic == nir_intrinsic_load_barycentric_at_sample)
info->uses_interp_at_sample = true;
break;
case nir_intrinsic_load_frag_coord:
info->reads_frag_coord_mask |= nir_def_components_read(&intr->def);
break;
case nir_intrinsic_load_input:
case nir_intrinsic_load_per_vertex_input:
case nir_intrinsic_load_per_primitive_input:
@ -525,7 +522,6 @@ void si_nir_gather_info(struct si_screen *sscreen, struct nir_shader *nir,
info->base.fs.uses_sample_shading = nir->info.fs.uses_sample_shading;
info->base.fs.early_fragment_tests = nir->info.fs.early_fragment_tests;
info->base.fs.post_depth_coverage = nir->info.fs.post_depth_coverage;
info->base.fs.pixel_center_integer = nir->info.fs.pixel_center_integer;
info->base.fs.depth_layout = nir->info.fs.depth_layout;
break;

4
src/gallium/drivers/radeonsi/gfx/si_shader_info.h
View file

@ -70,7 +70,6 @@ struct si_shader_info {
bool uses_sample_shading:1;
bool early_fragment_tests:1;
bool post_depth_coverage:1;
bool pixel_center_integer:1;
enum gl_frag_depth_layout depth_layout:3;
} fs;
@ -172,9 +171,6 @@ struct si_shader_info {
*/
uint8_t writes_1_if_tex_is_1;
/* frag coord and sample pos per component read mask. */
uint8_t reads_frag_coord_mask;
unsigned ngg_cull_vert_threshold; /* UINT32_MAX = disabled */
enum mesa_prim rast_prim;

22
src/gallium/drivers/radeonsi/gfx/si_shader_llvm_ps.c
View file

@ -678,28 +678,6 @@ void si_llvm_build_ps_prolog(struct si_shader_context *ctx, union si_shader_part
args->ac.sample_coverage.arg_index);
}
if (key->ps_prolog.states.get_frag_coord_from_pixel_coord) {
LLVMValueRef pixel_coord = ac_get_arg(&ctx->ac, args->ac.pos_fixed_pt);
pixel_coord = LLVMBuildBitCast(ctx->ac.builder, pixel_coord, ctx->ac.v2i16, "");
pixel_coord = LLVMBuildUIToFP(ctx->ac.builder, pixel_coord, ctx->ac.v2f32, "");
if (!key->ps_prolog.pixel_center_integer) {
LLVMValueRef vec2_half = LLVMConstVector((LLVMValueRef[]){LLVMConstReal(ctx->ac.f32, 0.5),
LLVMConstReal(ctx->ac.f32, 0.5)}, 2);
pixel_coord = LLVMBuildFAdd(ctx->ac.builder, pixel_coord, vec2_half, "");
}
for (unsigned i = 0; i < 2; i++) {
if (!args->ac.frag_pos[i].used)
continue;
ret = insert_ret_of_arg(ctx, ret,
LLVMBuildExtractElement(ctx->ac.builder, pixel_coord,
LLVMConstInt(ctx->ac.i32, i, 0), ""),
args->ac.frag_pos[i].arg_index);
}
}
/* Tell LLVM to insert WQM instruction sequence when needed. */
if (key->ps_prolog.wqm) {
LLVMAddTargetDependentFunctionAttr(func, "amdgpu-ps-wqm-outputs", "");

6
src/gallium/drivers/radeonsi/gfx/si_shader_variant_info.c
View file

@ -580,12 +580,6 @@ void si_set_spi_ps_input_config_for_separate_prolog(struct si_shader *shader)
/* The sample mask fixup has an optimization that replaces the sample mask with the sample ID. */
if (key->ps.part.prolog.samplemask_log_ps_iter == 3)
shader->config.spi_ps_input_ena &= C_0286CC_SAMPLE_COVERAGE_ENA;
if (key->ps.part.prolog.get_frag_coord_from_pixel_coord) {
shader->config.spi_ps_input_ena &= C_0286CC_POS_X_FLOAT_ENA;
shader->config.spi_ps_input_ena &= C_0286CC_POS_Y_FLOAT_ENA;
shader->config.spi_ps_input_ena |= S_0286CC_POS_FIXED_PT_ENA(1);
}
}
void si_fixup_spi_ps_input_config(struct si_shader *shader)

4
src/gallium/drivers/radeonsi/gfx/si_state_shaders.cpp
View file

@ -2424,8 +2424,6 @@ void si_ps_key_update_framebuffer_rasterizer_sample_shading(struct si_context *s
uses_persp_center && uses_persp_centroid;
key->ps.part.prolog.bc_optimize_for_linear =
sel->info.uses_sysval_linear_center && sel->info.uses_sysval_linear_centroid;
key->ps.part.prolog.get_frag_coord_from_pixel_coord =
!sel->info.base.fs.uses_sample_shading && sel->info.reads_frag_coord_mask & 0x3;
key->ps.part.prolog.force_samplemask_to_helper_invocation = 0;
key->ps.mono.force_mono = 0;
key->ps.mono.interpolate_at_sample_force_center = 0;
@ -2444,8 +2442,6 @@ void si_ps_key_update_framebuffer_rasterizer_sample_shading(struct si_context *s
sel->info.uses_sysval_linear_sample > 1;
key->ps.part.prolog.bc_optimize_for_persp = 0;
key->ps.part.prolog.bc_optimize_for_linear = 0;
key->ps.part.prolog.get_frag_coord_from_pixel_coord =
!!(sel->info.reads_frag_coord_mask & 0x3);
key->ps.part.prolog.force_samplemask_to_helper_invocation = sel->info.uses_sysval_sample_mask_in;
key->ps.mono.force_mono = 0;
key->ps.mono.interpolate_at_sample_force_center = sel->info.uses_interp_at_sample;