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

radv: use nir_lower_conv64

Browse Source 
This has a more accurate i2f/u2f implementation.

fossil-db (navi21):
Totals from 342 (0.26% of 133461) affected shaders:
MaxWaves: 10480 -> 10472 (-0.08%)
Instrs: 273455 -> 274019 (+0.21%); split: -0.02%, +0.22%
CodeSize: 1545020 -> 1546852 (+0.12%); split: -0.11%, +0.23%
VGPRs: 9528 -> 9552 (+0.25%)
SpillSGPRs: 553 -> 592 (+7.05%); split: -2.71%, +9.76%
SpillVGPRs: 821 -> 811 (-1.22%); split: -1.71%, +0.49%
Latency: 3837590 -> 3822989 (-0.38%); split: -0.52%, +0.14%
InvThroughput: 1284512 -> 1277012 (-0.58%); split: -0.77%, +0.19%
VClause: 6480 -> 6486 (+0.09%); split: -0.15%, +0.25%
SClause: 6640 -> 6645 (+0.08%); split: -0.02%, +0.09%
Copies: 45858 -> 46160 (+0.66%); split: -0.10%, +0.76%
Branches: 8736 -> 8741 (+0.06%); split: -0.01%, +0.07%
PreSGPRs: 14320 -> 14323 (+0.02%)
PreVGPRs: 8545 -> 8551 (+0.07%)

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/9275
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23926>
This commit is contained in:
Rhys Perry
2023-06-27 19:22:42 +01:00
committed by Marge Bot
parent 3d0e997e99
commit 38cff03e58
1 changed files with 1 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/amd/vulkan/radv_shader.c
View File

@@ -111,10 +111,8 @@ get_nir_options_for_stage(struct radv_physical_device *device, gl_shader_stage s
.max_unroll_iterations_aggressive = 128,
.use_interpolated_input_intrinsics = true,
.vectorize_vec2_16bit = true,
/* nir_lower_int64() isn't actually called for the LLVM backend,
* but this helps the loop unrolling heuristics. */
.lower_int64_options = nir_lower_imul64 | nir_lower_imul_high64 | nir_lower_imul_2x32_64 | nir_lower_divmod64 |
nir_lower_minmax64 | nir_lower_iabs64 | nir_lower_iadd_sat64,
nir_lower_minmax64 | nir_lower_iabs64 | nir_lower_iadd_sat64 | nir_lower_conv64,
.lower_doubles_options = nir_lower_drcp | nir_lower_dsqrt | nir_lower_drsq | nir_lower_ddiv,
.divergence_analysis_options = nir_divergence_view_index_uniform,
.optimize_quad_vote_to_reduce = true,