[llvm] r358346 - [X86] Update bool_reduction_v8f32 test cases from vector-compare-any_of.ll and vector-compare-all_of.ll to be proper reductions.
Craig Topper via llvm-commits
llvm-commits at lists.llvm.org
Sat Apr 13 21:20:42 PDT 2019
Author: ctopper
Date: Sat Apr 13 21:20:42 2019
New Revision: 358346
URL: http://llvm.org/viewvc/llvm-project?rev=358346&view=rev
Log:
[X86] Update bool_reduction_v8f32 test cases from vector-compare-any_of.ll and vector-compare-all_of.ll to be proper reductions.
One of the shuffles was used twice. While the intended shuffle wasn't connected.
Modified:
llvm/trunk/test/CodeGen/X86/vector-compare-all_of.ll
llvm/trunk/test/CodeGen/X86/vector-compare-any_of.ll
Modified: llvm/trunk/test/CodeGen/X86/vector-compare-all_of.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vector-compare-all_of.ll?rev=358346&r1=358345&r2=358346&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/vector-compare-all_of.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vector-compare-all_of.ll Sat Apr 13 21:20:42 2019
@@ -1043,27 +1043,18 @@ define i1 @bool_reduction_v8f32(<8 x flo
; SSE-NEXT: cmpneqps %xmm3, %xmm1
; SSE-NEXT: cmpneqps %xmm2, %xmm0
; SSE-NEXT: packssdw %xmm1, %xmm0
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE-NEXT: pand %xmm0, %xmm1
-; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
-; SSE-NEXT: pand %xmm0, %xmm1
-; SSE-NEXT: pand %xmm0, %xmm1
-; SSE-NEXT: pextrb $0, %xmm1, %eax
-; SSE-NEXT: # kill: def $al killed $al killed $eax
+; SSE-NEXT: packsswb %xmm0, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: cmpb $-1, %al
+; SSE-NEXT: sete %al
; SSE-NEXT: retq
;
; AVX-LABEL: bool_reduction_v8f32:
; AVX: # %bb.0:
; AVX-NEXT: vcmpneqps %ymm1, %ymm0, %ymm0
-; AVX-NEXT: vextractf128 $1, %ymm0, %xmm1
-; AVX-NEXT: vpackssdw %xmm1, %xmm0, %xmm0
-; AVX-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX-NEXT: vpand %xmm0, %xmm1, %xmm0
-; AVX-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX-NEXT: vpand %xmm1, %xmm0, %xmm0
-; AVX-NEXT: vpand %xmm0, %xmm1, %xmm0
-; AVX-NEXT: vpextrb $0, %xmm0, %eax
-; AVX-NEXT: # kill: def $al killed $al killed $eax
+; AVX-NEXT: vmovmskps %ymm0, %eax
+; AVX-NEXT: cmpb $-1, %al
+; AVX-NEXT: sete %al
; AVX-NEXT: vzeroupper
; AVX-NEXT: retq
;
@@ -1075,10 +1066,13 @@ define i1 @bool_reduction_v8f32(<8 x flo
; AVX512-NEXT: vextracti128 $1, %ymm1, %xmm1
; AVX512-NEXT: vpslld $31, %ymm1, %ymm1
; AVX512-NEXT: vptestmd %ymm1, %ymm1, %k1 {%k1}
+; AVX512-NEXT: vmovdqa32 %ymm0, %ymm1 {%k1} {z}
+; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
+; AVX512-NEXT: vpslld $31, %ymm1, %ymm1
+; AVX512-NEXT: vptestmd %ymm1, %ymm1, %k1 {%k1}
; AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
-; AVX512-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
+; AVX512-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
; AVX512-NEXT: vpslld $31, %ymm0, %ymm0
-; AVX512-NEXT: vptestmd %ymm0, %ymm0, %k1 {%k1}
; AVX512-NEXT: vptestmd %ymm0, %ymm0, %k0 {%k1}
; AVX512-NEXT: kmovd %k0, %eax
; AVX512-NEXT: # kill: def $al killed $al killed $eax
@@ -1090,7 +1084,7 @@ define i1 @bool_reduction_v8f32(<8 x flo
%s2 = shufflevector <8 x i1> %b, <8 x i1> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%c = and <8 x i1> %s2, %b
%s3 = shufflevector <8 x i1> %c, <8 x i1> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
- %d = and <8 x i1> %s2, %c
+ %d = and <8 x i1> %s3, %c
%e = extractelement <8 x i1> %d, i32 0
ret i1 %e
}
Modified: llvm/trunk/test/CodeGen/X86/vector-compare-any_of.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vector-compare-any_of.ll?rev=358346&r1=358345&r2=358346&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/vector-compare-any_of.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vector-compare-any_of.ll Sat Apr 13 21:20:42 2019
@@ -968,27 +968,18 @@ define i1 @bool_reduction_v8f32(<8 x flo
; SSE-NEXT: cmpneqps %xmm3, %xmm1
; SSE-NEXT: cmpneqps %xmm2, %xmm0
; SSE-NEXT: packssdw %xmm1, %xmm0
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE-NEXT: por %xmm0, %xmm1
-; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
-; SSE-NEXT: por %xmm0, %xmm1
-; SSE-NEXT: por %xmm0, %xmm1
-; SSE-NEXT: pextrb $0, %xmm1, %eax
-; SSE-NEXT: # kill: def $al killed $al killed $eax
+; SSE-NEXT: packsswb %xmm0, %xmm0
+; SSE-NEXT: pmovmskb %xmm0, %eax
+; SSE-NEXT: testb %al, %al
+; SSE-NEXT: setne %al
; SSE-NEXT: retq
;
; AVX-LABEL: bool_reduction_v8f32:
; AVX: # %bb.0:
; AVX-NEXT: vcmpneqps %ymm1, %ymm0, %ymm0
-; AVX-NEXT: vextractf128 $1, %ymm0, %xmm1
-; AVX-NEXT: vpackssdw %xmm1, %xmm0, %xmm0
-; AVX-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX-NEXT: vpor %xmm0, %xmm1, %xmm0
-; AVX-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX-NEXT: vpor %xmm1, %xmm0, %xmm0
-; AVX-NEXT: vpor %xmm0, %xmm1, %xmm0
-; AVX-NEXT: vpextrb $0, %xmm0, %eax
-; AVX-NEXT: # kill: def $al killed $al killed $eax
+; AVX-NEXT: vmovmskps %ymm0, %eax
+; AVX-NEXT: testb %al, %al
+; AVX-NEXT: setne %al
; AVX-NEXT: vzeroupper
; AVX-NEXT: retq
;
@@ -1001,11 +992,15 @@ define i1 @bool_reduction_v8f32(<8 x flo
; AVX512-NEXT: vpslld $31, %ymm1, %ymm1
; AVX512-NEXT: vptestmd %ymm1, %ymm1, %k0
; AVX512-NEXT: korw %k1, %k0, %k1
+; AVX512-NEXT: vmovdqa32 %ymm0, %ymm1 {%k1} {z}
+; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
+; AVX512-NEXT: vpslld $31, %ymm1, %ymm1
+; AVX512-NEXT: vptestmd %ymm1, %ymm1, %k0
+; AVX512-NEXT: korw %k1, %k0, %k1
; AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
-; AVX512-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
+; AVX512-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
; AVX512-NEXT: vpslld $31, %ymm0, %ymm0
; AVX512-NEXT: vptestmd %ymm0, %ymm0, %k0
-; AVX512-NEXT: korw %k1, %k0, %k1
; AVX512-NEXT: korw %k1, %k0, %k0
; AVX512-NEXT: kmovd %k0, %eax
; AVX512-NEXT: # kill: def $al killed $al killed $eax
@@ -1017,7 +1012,7 @@ define i1 @bool_reduction_v8f32(<8 x flo
%s2 = shufflevector <8 x i1> %b, <8 x i1> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%c = or <8 x i1> %s2, %b
%s3 = shufflevector <8 x i1> %c, <8 x i1> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
- %d = or <8 x i1> %s2, %c
+ %d = or <8 x i1> %s3, %c
%e = extractelement <8 x i1> %d, i32 0
ret i1 %e
}
More information about the llvm-commits
mailing list