[llvm] [AMDGPU] Prefer packed minimum/maximum ops for two-input ops (PR #215449)

Shilei Tian via llvm-commits llvm-commits at lists.llvm.org
Mon Aug 10 20:11:26 PDT 2026


================
@@ -0,0 +1,23 @@
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx950 < %s | FileCheck -check-prefix=GFX950 %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 < %s | FileCheck -check-prefix=GFX1250 %s
+
+define <2 x half> @maximum_v2f16(<2 x half> %src0, <2 x half> %src1) {
+; GFX950-LABEL: maximum_v2f16:
+; GFX950:       v_pk_maximum3_f16 v0, v0, v1, v1
+; GFX1250-LABEL: maximum_v2f16:
+; GFX1250:       v_pk_maximum_f16 v0, v0, v1
+  %max = call <2 x half> @llvm.maximum.v2f16(<2 x half> %src0, <2 x half> %src1)
----------------
shiltian wrote:

auto generate check lines pls

https://github.com/llvm/llvm-project/pull/215449


More information about the llvm-commits mailing list