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

ac/llvm: use mul24 intrinsics

Browse Source 
With the current code in clpeak LLVM ended up generating v_mad_u64_u32
instructions, with this we get nice v_mad_u32_s24 ones instead and an 4x
performance increase in the int24 benchmark.

Suggested-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34630>
This commit is contained in:
Karol Herbst
2025-04-21 13:38:01 +02:00
committed by Marge Bot
parent 3c267535ae
commit e3edc6029b
1 changed files with 5 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
src/amd/llvm/ac_nir_to_llvm.c
View File

@@ -612,9 +612,13 @@ static bool visit_alu(struct ac_nir_context *ctx, const nir_alu_instr *instr)
else
result = LLVMBuildSub(ctx->ac.builder, src[0], src[1], "");
break;
case nir_op_imul:
case nir_op_imul24_relaxed:
result = ac_build_intrinsic(&ctx->ac, "llvm.amdgcn.mul.i24", ctx->ac.i32, src, 2, 0);
break;
case nir_op_umul24_relaxed:
result = ac_build_intrinsic(&ctx->ac, "llvm.amdgcn.mul.u24", ctx->ac.i32, src, 2, 0);
break;
case nir_op_imul:
if (instr->no_unsigned_wrap)
result = LLVMBuildNUWMul(ctx->ac.builder, src[0], src[1], "");
else if (instr->no_signed_wrap)