27 lines · plain
1; RUN: llc < %s -mtriple=nvptx64 -mcpu=sm_80 -mattr=+ptx81 | FileCheck %s2 3declare { float, float, float, float } @llvm.nvvm.mma.m16n8k4.row.col.tf32(i32, i32, i32, float, float, float, float) #14 5declare noundef i32 @llvm.nvvm.read.ptx.sreg.laneid() #06 7; COM: llvm.nvvm.mma should not sink to the next block and gets reordered to be after laneid check.8; CHECK-LABEL: no_reorder_mma_and_laneid_check9define dso_local void @no_reorder_mma_and_laneid_check(ptr %arg, ptr %arg1) {10bb:11 ; CHECK: mma.sync.aligned.m16n8k4.row.col.f32.tf32.tf32.f3212 ; CHECK: laneid13 %i = tail call { float, float, float, float } @llvm.nvvm.mma.m16n8k4.row.col.tf32(i32 10, i32 10, i32 8, float 0.0, float 0.0, float 0.0, float 0.0)14 %i3 = tail call i32 @llvm.nvvm.read.ptx.sreg.laneid()15 %i4 = icmp eq i32 %i3, 016 br i1 %i4, label %bb5, label %bb817 18bb5: ; preds = %bb19 %i6 = extractvalue { float, float, float, float } %i, 020 %i7 = getelementptr float, ptr %arg, i64 021 store float %i6, ptr %i7, align 422 br label %bb823 24bb8: ; preds = %bb5, %bb25 ret void26}27