From 57b4c5d4730081ea4e8bd21cd747865411ad5cf3 Mon Sep 17 00:00:00 2001 From: Quentin Colombet Date: Wed, 7 May 2014 22:52:58 +0000 Subject: [PATCH] [X86] Add a test case for r208252. Prior to r208252, the FMA 231 family was marked as isCommutable. However the memory variants of this family are not commutable. Therefore, we did not implemented the findCommutedOpIndices for those variants and missed that the default implementation (more or less: commute indices 1 and 2) was firing behind our back. As a result, as demonstrated in the test case before the fix, we were transforming a = b * c + a into a = a * c + b. I.e., before r208252 we were generating for this test case: vmovaps %xmm0, %xmm1 vmoss (%rsi), %xmm0 vfmadd231ss (%rdi), %xmm1, %xmm0 Instead of: vmoss (%rsi), %xmm1 vfmadd231ss (%rdi), %xmm1, %xmm0 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@208260 91177308-0d34-0410-b5e6-96231b3b80d8 --- test/CodeGen/X86/fma-do-not-commute.ll | 30 ++++++++++++++++++++++++++ 1 file changed, 30 insertions(+) create mode 100644 test/CodeGen/X86/fma-do-not-commute.ll diff --git a/test/CodeGen/X86/fma-do-not-commute.ll b/test/CodeGen/X86/fma-do-not-commute.ll new file mode 100644 index 00000000000..4e211721a38 --- /dev/null +++ b/test/CodeGen/X86/fma-do-not-commute.ll @@ -0,0 +1,30 @@ +; RUN: llc -fp-contract=fast -mattr=+fma -disable-cgp < %s -o - | FileCheck %s +; Check that the 2nd and 3rd arguments of fmaXXX231 reg1, reg2, mem3 are not commuted. +; +target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128" +target triple = "x86_64-apple-macosx" + +; CHECK-LABEL: test1: +; %arg lives in xmm0 and it shouldn't be redefined until it is used in the FMA. +; CHECK-NOT {{.*}}, %xmm0 +; %addr lives in rdi. +; %addr2 lives in rsi. +; CHECK: vmovss (%rsi), [[ADDR2:%xmm[0-9]+]] +; The assembly syntax is in the reverse order. +; CHECK: vfmadd231ss (%rdi), [[ADDR2]], %xmm0 +define void @test1(float* %addr, float* %addr2, float %arg) { +entry: + br label %loop + +loop: + %sum0 = phi float [ %fma, %loop ], [ %arg, %entry ] + %addrVal = load float* %addr, align 4 + %addr2Val = load float* %addr2, align 4 + %fmul = fmul float %addrVal, %addr2Val + %fma = fadd float %sum0, %fmul + br i1 true, label %exit, label %loop + +exit: + store float %fma, float* %addr, align 4 + ret void +}