[llvm] r350646 - [x86] add tests for PR40243; NFC
Sanjay Patel via llvm-commits
llvm-commits at lists.llvm.org
Tue Jan 8 11:15:21 PST 2019
Author: spatel
Date: Tue Jan 8 11:15:21 2019
New Revision: 350646
URL: http://llvm.org/viewvc/llvm-project?rev=350646&view=rev
Log:
[x86] add tests for PR40243; NFC
Modified:
llvm/trunk/test/CodeGen/X86/haddsub-undef.ll
llvm/trunk/test/CodeGen/X86/phaddsub-undef.ll
Modified: llvm/trunk/test/CodeGen/X86/haddsub-undef.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/haddsub-undef.ll?rev=350646&r1=350645&r2=350646&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/haddsub-undef.ll (original)
+++ llvm/trunk/test/CodeGen/X86/haddsub-undef.ll Tue Jan 8 11:15:21 2019
@@ -926,8 +926,6 @@ define <4 x float> @v16f32_inputs_v4f32_
ret <4 x float> %r
}
-; FIXME: Miscompiles with any AVX.
-
define <8 x float> @v16f32_inputs_v8f32_output_4567(<16 x float> %a, <16 x float> %b) {
; SSE-LABEL: v16f32_inputs_v8f32_output_4567:
; SSE: # %bb.0:
@@ -964,6 +962,27 @@ define <8 x float> @v16f32_inputs_v8f32_
%add7 = fadd float %b6, %b7
%r4 = insertelement <8 x float> undef, float %add4, i32 4
%r = insertelement <8 x float> %r4, float %add7, i32 7
+ ret <8 x float> %r
+}
+
+define <8 x float> @PR40243(<8 x float> %a, <8 x float> %b) {
+; SSE-LABEL: PR40243:
+; SSE: # %bb.0:
+; SSE-NEXT: haddps %xmm3, %xmm1
+; SSE-NEXT: retq
+;
+; AVX-LABEL: PR40243:
+; AVX: # %bb.0:
+; AVX-NEXT: vhaddps %ymm0, %ymm0, %ymm0
+; AVX-NEXT: retq
+ %a4 = extractelement <8 x float> %a, i32 4
+ %a5 = extractelement <8 x float> %a, i32 5
+ %add4 = fadd float %a4, %a5
+ %b6 = extractelement <8 x float> %b, i32 6
+ %b7 = extractelement <8 x float> %b, i32 7
+ %add7 = fadd float %b6, %b7
+ %r4 = insertelement <8 x float> undef, float %add4, i32 4
+ %r = insertelement <8 x float> %r4, float %add7, i32 7
ret <8 x float> %r
}
Modified: llvm/trunk/test/CodeGen/X86/phaddsub-undef.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/phaddsub-undef.ll?rev=350646&r1=350645&r2=350646&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/phaddsub-undef.ll (original)
+++ llvm/trunk/test/CodeGen/X86/phaddsub-undef.ll Tue Jan 8 11:15:21 2019
@@ -93,6 +93,36 @@ define <8 x i32> @test15_undef(<8 x i32>
ret <8 x i32> %vecinit5
}
+define <8 x i32> @PR40243_alt(<8 x i32> %a, <8 x i32> %b) {
+; SSE-LABEL: PR40243_alt:
+; SSE: # %bb.0:
+; SSE-NEXT: phaddd %xmm3, %xmm1
+; SSE-NEXT: retq
+;
+; AVX1-LABEL: PR40243_alt:
+; AVX1: # %bb.0:
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: PR40243_alt:
+; AVX2: # %bb.0:
+; AVX2-NEXT: vphaddd %ymm0, %ymm0, %ymm0
+; AVX2-NEXT: retq
+;
+; AVX512-LABEL: PR40243_alt:
+; AVX512: # %bb.0:
+; AVX512-NEXT: vphaddd %ymm0, %ymm0, %ymm0
+; AVX512-NEXT: retq
+ %a4 = extractelement <8 x i32> %a, i32 4
+ %a5 = extractelement <8 x i32> %a, i32 5
+ %add4 = add i32 %a4, %a5
+ %b6 = extractelement <8 x i32> %b, i32 6
+ %b7 = extractelement <8 x i32> %b, i32 7
+ %add7 = add i32 %b6, %b7
+ %r4 = insertelement <8 x i32> undef, i32 %add4, i32 4
+ %r = insertelement <8 x i32> %r4, i32 %add7, i32 7
+ ret <8 x i32> %r
+}
+
define <8 x i32> @test16_undef(<8 x i32> %a, <8 x i32> %b) {
; SSE-LABEL: test16_undef:
; SSE: # %bb.0:
More information about the llvm-commits
mailing list