fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-07 22:18:13 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 4

aco: fix shared VGPR allocation on RDNA2

Browse source 
VGPRs are now allocated in blocks of 8 normal
or 16 shared VGPRs, respectively.

Fixes: 14a5021aff ('aco/gfx10: Refactor of GFX10 wave64 bpermute.')

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8921>
(cherry picked from commit bacc3b36f5)
This commit is contained in:
Daniel Schürmann 2021-02-05 18:25:18 +01:00 • committed by Dylan Baker
parent f9a57d53c9
commit 8e8ba3402a
2 changed files with 4 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
.pick_status.json
View file

@ -3226,7 +3226,7 @@
"description": "aco: fix shared VGPR allocation on RDNA2",
"nominated": true,
"nomination_type": 1,
"resolution": 0,
"resolution": 1,
"master_sha": null,
"because_sha": "14a5021aff661a26d76f330fec55d400d35443a8"
},

5
src/amd/compiler/aco_instruction_selection.cpp
View file

@ -197,8 +197,9 @@ static Temp emit_bpermute(isel_context *ctx, Builder &bld, Temp index, Temp data
/* GFX10 wave64 mode: emulate full-wave bpermute */
if (!ctx->has_gfx10_wave64_bpermute) {
ctx->has_gfx10_wave64_bpermute = true;
ctx->program->config->num_shared_vgprs = 8; /* Shared VGPRs are allocated in groups of 8 */
ctx->program->vgpr_limit -= 4; /* We allocate 8 shared VGPRs, so we'll have 4 fewer normal VGPRs */
/* Shared VGPRs are allocated in groups of 8/16 */
ctx->program->config->num_shared_vgprs = ctx->program->chip_class >= GFX10_3 ? 16 : 8;
ctx->program->vgpr_limit -= ctx->program->chip_class >= GFX10_3 ? 8 : 4;
}
Temp index_is_lo = bld.vopc(aco_opcode::v_cmp_ge_u32, bld.def(bld.lm), Operand(31u), index);