From c97d8c24823b4bb60b05fea808b173c998efeece Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Tue, 20 Aug 2019 13:25:55 +0000
Subject: [PATCH] [X86][FMA] Add FMA 'negated expression' combine tests for
 D63141

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369384 91177308-0d34-0410-b5e6-96231b3b80d8
---
 test/CodeGen/X86/fma-fneg-combine-2.ll | 89 ++++++++++++++++++++++++++
 1 file changed, 89 insertions(+)
 create mode 100644 test/CodeGen/X86/fma-fneg-combine-2.ll

diff --git a/test/CodeGen/X86/fma-fneg-combine-2.ll b/test/CodeGen/X86/fma-fneg-combine-2.ll
new file mode 100644
index 00000000000..1f520f9a398
--- /dev/null
+++ b/test/CodeGen/X86/fma-fneg-combine-2.ll
@@ -0,0 +1,89 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+avx,fma | FileCheck %s --check-prefixes=CHECK,FMA3
+; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+avx,fma4 | FileCheck %s --check-prefixes=CHECK,FMA4
+
+define float @test_fneg_fma_subx_y_negz_f32(float %w, float %x, float %y, float %z)  {
+; FMA3-LABEL: test_fneg_fma_subx_y_negz_f32:
+; FMA3:       # %bb.0: # %entry
+; FMA3-NEXT:    vsubss %xmm1, %xmm0, %xmm0
+; FMA3-NEXT:    vfnmadd213ss {{.*#+}} xmm0 = -(xmm2 * xmm0) + xmm3
+; FMA3-NEXT:    retq
+;
+; FMA4-LABEL: test_fneg_fma_subx_y_negz_f32:
+; FMA4:       # %bb.0: # %entry
+; FMA4-NEXT:    vsubss %xmm1, %xmm0, %xmm0
+; FMA4-NEXT:    vfnmaddss %xmm3, %xmm2, %xmm0, %xmm0
+; FMA4-NEXT:    retq
+entry:
+  %subx = fsub nsz float %w, %x
+  %negz = fsub float -0.000000e+00, %z
+  %0 = tail call nsz float @llvm.fma.f32(float %subx, float %y, float %negz)
+  %1 = fsub float -0.000000e+00, %0
+  ret float %1
+}
+
+define float @test_fneg_fma_x_suby_negz_f32(float %w, float %x, float %y, float %z)  {
+; FMA3-LABEL: test_fneg_fma_x_suby_negz_f32:
+; FMA3:       # %bb.0: # %entry
+; FMA3-NEXT:    vsubss %xmm2, %xmm0, %xmm0
+; FMA3-NEXT:    vfnmadd213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm3
+; FMA3-NEXT:    retq
+;
+; FMA4-LABEL: test_fneg_fma_x_suby_negz_f32:
+; FMA4:       # %bb.0: # %entry
+; FMA4-NEXT:    vsubss %xmm2, %xmm0, %xmm0
+; FMA4-NEXT:    vfnmaddss %xmm3, %xmm0, %xmm1, %xmm0
+; FMA4-NEXT:    retq
+entry:
+  %suby = fsub nsz float %w, %y
+  %negz = fsub float -0.000000e+00, %z
+  %0 = tail call nsz float @llvm.fma.f32(float %x, float %suby, float %negz)
+  %1 = fsub float -0.000000e+00, %0
+  ret float %1
+}
+
+define float @test_fneg_fma_subx_suby_negz_f32(float %w, float %x, float %y, float %z)  {
+; FMA3-LABEL: test_fneg_fma_subx_suby_negz_f32:
+; FMA3:       # %bb.0: # %entry
+; FMA3-NEXT:    vsubss %xmm1, %xmm0, %xmm1
+; FMA3-NEXT:    vsubss %xmm2, %xmm0, %xmm0
+; FMA3-NEXT:    vfnmadd213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm3
+; FMA3-NEXT:    retq
+;
+; FMA4-LABEL: test_fneg_fma_subx_suby_negz_f32:
+; FMA4:       # %bb.0: # %entry
+; FMA4-NEXT:    vsubss %xmm1, %xmm0, %xmm1
+; FMA4-NEXT:    vsubss %xmm2, %xmm0, %xmm0
+; FMA4-NEXT:    vfnmaddss %xmm3, %xmm0, %xmm1, %xmm0
+; FMA4-NEXT:    retq
+entry:
+  %subx = fsub nsz float %w, %x
+  %suby = fsub nsz float %w, %y
+  %negz = fsub float -0.000000e+00, %z
+  %0 = tail call nsz float @llvm.fma.f32(float %subx, float %suby, float %negz)
+  %1 = fsub float -0.000000e+00, %0
+  ret float %1
+}
+
+define float @test_fneg_fma_subx_negy_negz_f32(float %w, float %x, float %y, float %z)  {
+; FMA3-LABEL: test_fneg_fma_subx_negy_negz_f32:
+; FMA3:       # %bb.0: # %entry
+; FMA3-NEXT:    vsubss %xmm1, %xmm0, %xmm0
+; FMA3-NEXT:    vfmadd213ss {{.*#+}} xmm0 = (xmm2 * xmm0) + xmm3
+; FMA3-NEXT:    retq
+;
+; FMA4-LABEL: test_fneg_fma_subx_negy_negz_f32:
+; FMA4:       # %bb.0: # %entry
+; FMA4-NEXT:    vsubss %xmm1, %xmm0, %xmm0
+; FMA4-NEXT:    vfmaddss %xmm3, %xmm2, %xmm0, %xmm0
+; FMA4-NEXT:    retq
+entry:
+  %subx = fsub nsz float %w, %x
+  %negy = fsub float -0.000000e+00, %y
+  %negz = fsub float -0.000000e+00, %z
+  %0 = tail call nsz float @llvm.fma.f32(float %subx, float %negy, float %negz)
+  %1 = fsub float -0.000000e+00, %0
+  ret float %1
+}
+
+declare float @llvm.fma.f32(float, float, float)
-- 
2.40.0