OpenHarmony/third_party_mesa3d
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

nir/fold_16bit_tex_image: Add type granularity for dst folding

Browse Source 
Some HW may be able to fold only some of dst types, e.g.
for Adreno folding i32 -> i16 could cause a different result since
folded variant clamps the result instead of masking it.

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20396>
This commit is contained in:
Danylo Piliaiev
2022-12-20 16:23:15 +01:00
parent c5231025be
commit 1c9ee30838
6 changed files with 12 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/amd/vulkan/radv_pipeline.c
View File

@@ -3745,7 +3745,7 @@ radv_postprocess_nir(struct radv_pipeline *pipeline,
}; };
struct nir_fold_16bit_tex_image_options fold_16bit_options = { struct nir_fold_16bit_tex_image_options fold_16bit_options = {
.rounding_mode = nir_rounding_mode_rtne, .rounding_mode = nir_rounding_mode_rtne,
.fold_tex_dest = true, .fold_tex_dest_types = nir_type_float | nir_type_uint | nir_type_int,
.fold_image_load_store_data = true, .fold_image_load_store_data = true,
.fold_image_srcs = !radv_use_llvm_for_stage(device, stage->stage), .fold_image_srcs = !radv_use_llvm_for_stage(device, stage->stage),
.fold_srcs_options_count = separate_g16 ? 2 : 1, .fold_srcs_options_count = separate_g16 ? 2 : 1,

2
src/compiler/nir/nir.h
View File

@@ -5491,7 +5491,7 @@ struct nir_fold_tex_srcs_options {
struct nir_fold_16bit_tex_image_options { struct nir_fold_16bit_tex_image_options {
nir_rounding_mode rounding_mode; nir_rounding_mode rounding_mode;
bool fold_tex_dest; nir_alu_type fold_tex_dest_types;
bool fold_image_load_store_data; bool fold_image_load_store_data;
bool fold_image_srcs; bool fold_image_srcs;
unsigned fold_srcs_options_count; unsigned fold_srcs_options_count;

10
src/compiler/nir/nir_lower_mediump.c
View File

@@ -901,7 +901,7 @@ fold_16bit_load_data(nir_builder *b, nir_intrinsic_instr *instr,
static bool static bool
fold_16bit_tex_dest(nir_tex_instr *tex, unsigned exec_mode, fold_16bit_tex_dest(nir_tex_instr *tex, unsigned exec_mode,
nir_rounding_mode rdm) nir_alu_type allowed_types, nir_rounding_mode rdm)
{ {
/* Skip sparse residency */ /* Skip sparse residency */
if (tex->is_sparse) if (tex->is_sparse)
@@ -918,6 +918,9 @@ fold_16bit_tex_dest(nir_tex_instr *tex, unsigned exec_mode,
tex->op != nir_texop_fragment_fetch_amd) tex->op != nir_texop_fragment_fetch_amd)
return false; return false;
if (!(nir_alu_type_get_base_type(tex->dest_type) & allowed_types))
return false;
if (!fold_16bit_destination(&tex->dest.ssa, tex->dest_type, exec_mode, rdm)) if (!fold_16bit_destination(&tex->dest.ssa, tex->dest_type, exec_mode, rdm))
return false; return false;
@@ -1086,8 +1089,9 @@ fold_16bit_tex_image(nir_builder *b, nir_instr *instr, void *params)
} else if (instr->type == nir_instr_type_tex) { } else if (instr->type == nir_instr_type_tex) {
nir_tex_instr *tex = nir_instr_as_tex(instr); nir_tex_instr *tex = nir_instr_as_tex(instr);
if (options->fold_tex_dest) if (options->fold_tex_dest_types)
progress |= fold_16bit_tex_dest(tex, exec_mode, options->rounding_mode); progress |= fold_16bit_tex_dest(tex, exec_mode, options->fold_tex_dest_types,
options->rounding_mode);
for (unsigned i = 0; i < options->fold_srcs_options_count; i++) { for (unsigned i = 0; i < options->fold_srcs_options_count; i++) {
progress |= fold_16bit_tex_srcs(b, tex, &options->fold_srcs_options[i]); progress |= fold_16bit_tex_srcs(b, tex, &options->fold_srcs_options[i]);

2
src/freedreno/ir3/ir3_nir.c
View File

@@ -769,7 +769,7 @@ ir3_nir_lower_variant(struct ir3_shader_variant *so, nir_shader *s)
}; };
struct nir_fold_16bit_tex_image_options fold_16bit_options = { struct nir_fold_16bit_tex_image_options fold_16bit_options = {
.rounding_mode = nir_rounding_mode_rtz, .rounding_mode = nir_rounding_mode_rtz,
.fold_tex_dest = true, .fold_tex_dest_types = nir_type_float | nir_type_uint | nir_type_int,
/* blob dumps have no half regs on pixel 2's ldib or stib, so only enable for a6xx+. */ /* blob dumps have no half regs on pixel 2's ldib or stib, so only enable for a6xx+. */
.fold_image_load_store_data = so->compiler->gen >= 6, .fold_image_load_store_data = so->compiler->gen >= 6,
.fold_srcs_options_count = 1, .fold_srcs_options_count = 1,

2
src/gallium/drivers/radeonsi/si_shader_nir.c
View File

@@ -196,7 +196,7 @@ static void si_late_optimize_16bit_samplers(struct si_screen *sscreen, nir_shade
}; };
struct nir_fold_16bit_tex_image_options fold_16bit_options = { struct nir_fold_16bit_tex_image_options fold_16bit_options = {
.rounding_mode = nir_rounding_mode_rtne, .rounding_mode = nir_rounding_mode_rtne,
.fold_tex_dest = true, .fold_tex_dest_types = nir_type_float | nir_type_uint | nir_type_int,
.fold_image_load_store_data = true, .fold_image_load_store_data = true,
.fold_srcs_options_count = has_g16 ? 2 : 1, .fold_srcs_options_count = has_g16 ? 2 : 1,
.fold_srcs_options = fold_srcs_options, .fold_srcs_options = fold_srcs_options,

2
src/gallium/frontends/lavapipe/lvp_pipeline.c
View File

@@ -489,7 +489,7 @@ lvp_shader_compile_to_ir(struct lvp_pipeline *pipeline,
/* Skip if there are potentially conflicting rounding modes */ /* Skip if there are potentially conflicting rounding modes */
struct nir_fold_16bit_tex_image_options fold_16bit_options = { struct nir_fold_16bit_tex_image_options fold_16bit_options = {
.rounding_mode = nir_rounding_mode_undef, .rounding_mode = nir_rounding_mode_undef,
.fold_tex_dest = true, .fold_tex_dest_types = nir_type_float | nir_type_uint | nir_type_int,
}; };
NIR_PASS_V(nir, nir_fold_16bit_tex_image, &fold_16bit_options); NIR_PASS_V(nir, nir_fold_16bit_tex_image, &fold_16bit_options);