KonstantinSeurer
/
mesa
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

nir_to_tgsi: Add a flag for lowering fabs, and use it in r300/i915. 

Saves instructions if the same fabs value is used multiple times.

i915g:
total instructions in shared programs: 397005 -> 396525 (-0.12%)
instructions in affected programs: 11061 -> 10581 (-4.34%)
LOST:   0
GAINED: 22

r300 (not r500):
total instructions in shared programs: 180286 -> 179767 (-0.29%)
instructions in affected programs: 27102 -> 26583 (-1.91%)
total temps in shared programs: 29692 -> 29638 (-0.18%)
temps in affected programs: 356 -> 302 (-15.17%)

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14938>
This commit is contained in:
Emma Anholt 2022-02-08 10:13:59 -08:00
parent f4ee7146f9
commit 2883e8f33d
5 changed files with 34 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
src/gallium/auxiliary/nir/nir_to_tgsi.c
View File

@ -1009,7 +1009,10 @@ ntt_emit_alu(struct ntt_compile *c, nir_alu_instr *instr)
break;
case nir_op_fabs:
ureg_MOV(c->ureg, dst, ureg_abs(src[0]));
if (c->options->lower_fabs)
ureg_MAX(c->ureg, dst, src[0], ureg_negate(src[0]));
else
ureg_MOV(c->ureg, dst, ureg_abs(src[0]));
break;
case nir_op_fsat:
@ -3162,8 +3165,15 @@ const void *nir_to_tgsi_options(struct nir_shader *s,
/* Only lower 32-bit floats. The only other modifier type officially
* supported by TGSI is 32-bit integer negates, but even those are broken on
* virglrenderer, so skip lowering all integer and f64 float mods.
*
* The options->lower_fabs requests that we not have native source modifiers
* for fabs, and instead emit MAX(a,-a) for nir_op_fabs.
*/
NIR_PASS_V(s, nir_lower_to_source_mods, nir_lower_float_source_mods);
nir_lower_to_source_mods_flags source_mods = nir_lower_fneg_source_mods;
if (!options->lower_fabs)
source_mods |= nir_lower_fabs_source_mods;
NIR_PASS_V(s, nir_lower_to_source_mods, source_mods);
NIR_PASS_V(s, nir_convert_from_ssa, true);
NIR_PASS_V(s, nir_lower_vec_to_movs, NULL, NULL);

2
src/gallium/auxiliary/nir/nir_to_tgsi.h
View File

@ -31,6 +31,8 @@ struct pipe_screen;
struct nir_to_tgsi_options {
bool lower_cmp;
/* Emit MAX(a,-a) instead of abs src modifier) */
bool lower_fabs;
};
const void *nir_to_tgsi(struct nir_shader *s,

12
src/gallium/drivers/i915/ci/i915-g33-fails.txt
View File

@ -47,19 +47,14 @@ dEQP-GLES2.functional.shaders.loops.for_constant_iterations.double_continue_frag
dEQP-GLES2.functional.shaders.loops.for_constant_iterations.mixed_break_continue_fragment,Fail
dEQP-GLES2.functional.shaders.random.all_features.fragment.22,Fail
dEQP-GLES2.functional.shaders.random.all_features.fragment.32,Fail
dEQP-GLES2.functional.shaders.random.all_features.fragment.34,Fail
dEQP-GLES2.functional.shaders.random.all_features.fragment.38,Fail
dEQP-GLES2.functional.shaders.random.all_features.fragment.5,Fail
dEQP-GLES2.functional.shaders.random.all_features.fragment.79,Fail
dEQP-GLES2.functional.shaders.random.all_features.fragment.80,Fail
dEQP-GLES2.functional.shaders.random.all_features.fragment.86,Fail
dEQP-GLES2.functional.shaders.random.all_features.fragment.89,Fail
dEQP-GLES2.functional.shaders.random.all_features.fragment.93,Fail
dEQP-GLES2.functional.shaders.random.all_features.fragment.97,Fail
dEQP-GLES2.functional.shaders.random.trigonometric.fragment.15,Fail
dEQP-GLES2.functional.shaders.random.trigonometric.fragment.26,Fail
dEQP-GLES2.functional.shaders.random.trigonometric.fragment.4,Fail
dEQP-GLES2.functional.shaders.random.trigonometric.fragment.42,Fail
dEQP-GLES2.functional.shaders.random.trigonometric.fragment.45,Fail
dEQP-GLES2.functional.shaders.random.trigonometric.fragment.52,Fail
@ -109,11 +104,9 @@ dEQP-GLES2.functional.uniform_api.random.20,Fail
dEQP-GLES2.functional.uniform_api.random.54,Fail
dEQP-GLES2.functional.uniform_api.random.71,Fail
dEQP-GLES2.functional.uniform_api.random.74,Fail
dEQP-GLES2.functional.uniform_api.random.80,Fail
dEQP-GLES2.functional.uniform_api.random.81,Fail
dEQP-GLES2.functional.uniform_api.value.assigned.by_pointer.render.array_in_struct.mat4_mat2_both,Fail
dEQP-GLES2.functional.uniform_api.value.assigned.by_pointer.render.array_in_struct.mat4_mat2_fragment,Fail
dEQP-GLES2.functional.uniform_api.value.assigned.by_pointer.render.array_in_struct.sampler2D_samplerCube_fragment,Fail
dEQP-GLES2.functional.uniform_api.value.assigned.by_pointer.render.basic_array.mat4_both,Fail
dEQP-GLES2.functional.uniform_api.value.assigned.by_pointer.render.basic_array.mat4_fragment,Fail
dEQP-GLES2.functional.uniform_api.value.assigned.by_pointer.render.multiple_nested_structs_arrays.both,Fail
@ -121,13 +114,10 @@ dEQP-GLES2.functional.uniform_api.value.assigned.by_pointer.render.multiple_nest
dEQP-GLES2.functional.uniform_api.value.assigned.by_pointer.render.nested_structs_arrays.mat4_mat2_fragment,Fail
dEQP-GLES2.functional.uniform_api.value.assigned.by_pointer.render.struct_in_array.mat4_mat2_both,Fail
dEQP-GLES2.functional.uniform_api.value.assigned.by_pointer.render.struct_in_array.mat4_mat2_fragment,Fail
dEQP-GLES2.functional.uniform_api.value.assigned.by_value.render.array_in_struct.sampler2D_samplerCube_fragment,Fail
dEQP-GLES2.functional.uniform_api.value.assigned.by_value.render.multiple_nested_structs_arrays.both,Fail
dEQP-GLES2.functional.uniform_api.value.assigned.by_value.render.multiple_nested_structs_arrays.fragment,Fail
dEQP-GLES2.functional.uniform_api.value.initial.render.array_in_struct.mat4_mat2_both,Fail
dEQP-GLES2.functional.uniform_api.value.initial.render.array_in_struct.mat4_mat2_fragment,Fail
dEQP-GLES2.functional.uniform_api.value.initial.render.basic_array.mat4_both,Fail
dEQP-GLES2.functional.uniform_api.value.initial.render.basic_array.mat4_fragment,Fail
dEQP-GLES2.functional.uniform_api.value.initial.render.multiple_nested_structs_arrays.both,Fail
dEQP-GLES2.functional.uniform_api.value.initial.render.multiple_nested_structs_arrays.fragment,Fail
dEQP-GLES2.functional.uniform_api.value.initial.render.nested_structs_arrays.mat4_mat2_fragment,Fail
@ -900,7 +890,7 @@ spec@glsl-1.10@execution@samplers@glsl-fs-shadow2d-bias,Fail
spec@glsl-1.10@execution@samplers@glsl-fs-shadow2d-clamp-z,Fail
spec@glsl-1.10@execution@samplers@glsl-fs-shadow2dproj,Fail
spec@glsl-1.10@execution@samplers@glsl-fs-shadow2dproj-bias,Fail
spec@glsl-1.10@execution@samplers@glsl-fs-texture2d-dependent-5,Fail
spec@glsl-1.10@execution@temp-array-indexing@glsl-fs-vec4-indexing-temp-dst-in-loop,Fail
spec@glsl-1.10@execution@temp-array-indexing@glsl-fs-vec4-indexing-temp-dst-in-nested-loop-combined,Fail
spec@glsl-1.10@execution@temp-array-indexing@glsl-fs-vec4-indexing-temp-src-in-loop,Fail

5
src/gallium/drivers/i915/i915_state.c
View File

@ -550,7 +550,10 @@ i915_create_fs_state(struct pipe_context *pipe,
NIR_PASS_V(s, i915_nir_lower_sincos);
ifs->state.tokens = nir_to_tgsi(s, pipe->screen);
static const struct nir_to_tgsi_options ntt_options = {
.lower_fabs = true,
};
ifs->state.tokens = nir_to_tgsi_options(s, pipe->screen, &ntt_options);
} else {
assert(templ->type == PIPE_SHADER_IR_TGSI);
/* we need to keep a local copy of the tokens */

18
src/gallium/drivers/r300/r300_state.c
View File

@ -1944,12 +1944,24 @@ static void* r300_create_vs_state(struct pipe_context* pipe,
if (vs->state.type == PIPE_SHADER_IR_NIR) {
static const struct nir_to_tgsi_options swtcl_options = {0};
static const struct nir_to_tgsi_options hwtcl_options = {
static const struct nir_to_tgsi_options hwtcl_r300_options = {
.lower_cmp = true,
.lower_fabs = true,
};
static const struct nir_to_tgsi_options hwtcl_r500_options = {
.lower_cmp = true,
};
const struct nir_to_tgsi_options *ntt_options;
if (r300->screen->caps.has_tcl) {
if (r300->screen->caps.is_r500)
ntt_options = &hwtcl_r500_options;
else
ntt_options = &hwtcl_r300_options;
} else {
ntt_options = &swtcl_options;
}
vs->state.tokens = nir_to_tgsi_options(shader->ir.nir, pipe->screen,
r300->screen->caps.has_tcl ?
&hwtcl_options : &swtcl_options);
ntt_options);
} else {
assert(vs->state.type == PIPE_SHADER_IR_TGSI);
/* we need to keep a local copy of the tokens */