maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/X86/vec-copysign.ll

170 lines
5.4 KiB
LLVM
Raw Normal View History

[x86, SSE/AVX] allow 128/256-bit lowering for copysign vector intrinsics (PR30433) This should fix: https://llvm.org/bugs/show_bug.cgi?id=30433 There are a couple of open questions about the codegen: 1. Should we let scalar ops be scalars and avoid vector constant loads/splats? 2. Should we have a pass to combine constants such as the inverted pair that we have here? Differential Revision: https://reviews.llvm.org/D25165 llvm-svn: 283119
2016-10-04 00:38:27 +08:00
; RUN: llc < %s -mtriple=x86_64-apple-macosx10.10.0 -mattr=+sse2 | FileCheck %s --check-prefix=SSE2 --check-prefix=CHECK
; RUN: llc < %s -mtriple=x86_64-apple-macosx10.10.0 -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=CHECK
[x86] add test to show unnecessary scalarization of copysign intrinsics (PR30433) llvm-svn: 283071
2016-10-03 00:31:35 +08:00
[x86, SSE/AVX] allow 128/256-bit lowering for copysign vector intrinsics (PR30433) This should fix: https://llvm.org/bugs/show_bug.cgi?id=30433 There are a couple of open questions about the codegen: 1. Should we let scalar ops be scalars and avoid vector constant loads/splats? 2. Should we have a pass to combine constants such as the inverted pair that we have here? Differential Revision: https://reviews.llvm.org/D25165 llvm-svn: 283119
2016-10-04 00:38:27 +08:00
; Assertions have been enhanced from utils/update_test_checks.py to show the constant pool values.
; Use a macosx triple to make sure the format of those constant strings is exact.
; CHECK: [[SIGNMASK1:L.+]]:
; CHECK-NEXT: .long 2147483648
; CHECK-NEXT: .long 2147483648
; CHECK-NEXT: .long 2147483648
; CHECK-NEXT: .long 2147483648
; CHECK: [[MAGMASK1:L.+]]:
; CHECK-NEXT: .long 2147483647
; CHECK-NEXT: .long 2147483647
; CHECK-NEXT: .long 2147483647
; CHECK-NEXT: .long 2147483647
[x86] add test to show unnecessary scalarization of copysign intrinsics (PR30433) llvm-svn: 283071
2016-10-03 00:31:35 +08:00
define <4 x float> @v4f32(<4 x float> %a, <4 x float> %b) nounwind {
; SSE2-LABEL: v4f32:
; SSE2: # BB#0:
[x86, SSE/AVX] allow 128/256-bit lowering for copysign vector intrinsics (PR30433) This should fix: https://llvm.org/bugs/show_bug.cgi?id=30433 There are a couple of open questions about the codegen: 1. Should we let scalar ops be scalars and avoid vector constant loads/splats? 2. Should we have a pass to combine constants such as the inverted pair that we have here? Differential Revision: https://reviews.llvm.org/D25165 llvm-svn: 283119
2016-10-04 00:38:27 +08:00
; SSE2-NEXT: andps [[SIGNMASK1]](%rip), %xmm1
; SSE2-NEXT: andps [[MAGMASK1]](%rip), %xmm0
[x86] add test to show unnecessary scalarization of copysign intrinsics (PR30433) llvm-svn: 283071
2016-10-03 00:31:35 +08:00
; SSE2-NEXT: orps %xmm1, %xmm0
; SSE2-NEXT: retq
;
; AVX-LABEL: v4f32:
; AVX: # BB#0:
[x86, SSE/AVX] allow 128/256-bit lowering for copysign vector intrinsics (PR30433) This should fix: https://llvm.org/bugs/show_bug.cgi?id=30433 There are a couple of open questions about the codegen: 1. Should we let scalar ops be scalars and avoid vector constant loads/splats? 2. Should we have a pass to combine constants such as the inverted pair that we have here? Differential Revision: https://reviews.llvm.org/D25165 llvm-svn: 283119
2016-10-04 00:38:27 +08:00
; AVX-NEXT: vandps [[SIGNMASK1]](%rip), %xmm1, %xmm1
; AVX-NEXT: vandps [[MAGMASK1]](%rip), %xmm0, %xmm0
[x86] add test to show unnecessary scalarization of copysign intrinsics (PR30433) llvm-svn: 283071
2016-10-03 00:31:35 +08:00
; AVX-NEXT: vorps %xmm1, %xmm0, %xmm0
; AVX-NEXT: retq
;
%tmp = tail call <4 x float> @llvm.copysign.v4f32( <4 x float> %a, <4 x float> %b )
ret <4 x float> %tmp
}
[x86, SSE/AVX] allow 128/256-bit lowering for copysign vector intrinsics (PR30433) This should fix: https://llvm.org/bugs/show_bug.cgi?id=30433 There are a couple of open questions about the codegen: 1. Should we let scalar ops be scalars and avoid vector constant loads/splats? 2. Should we have a pass to combine constants such as the inverted pair that we have here? Differential Revision: https://reviews.llvm.org/D25165 llvm-svn: 283119
2016-10-04 00:38:27 +08:00
; SSE2: [[SIGNMASK2:L.+]]:
; SSE2-NEXT: .long 2147483648
; SSE2-NEXT: .long 2147483648
; SSE2-NEXT: .long 2147483648
; SSE2-NEXT: .long 2147483648
; SSE2: [[MAGMASK2:L.+]]:
; SSE2-NEXT: .long 2147483647
; SSE2-NEXT: .long 2147483647
; SSE2-NEXT: .long 2147483647
; SSE2-NEXT: .long 2147483647
; AVX: [[SIGNMASK2:L.+]]:
; AVX-NEXT: .long 2147483648
; AVX-NEXT: .long 2147483648
; AVX-NEXT: .long 2147483648
; AVX-NEXT: .long 2147483648
; AVX-NEXT: .long 2147483648
; AVX-NEXT: .long 2147483648
; AVX-NEXT: .long 2147483648
; AVX-NEXT: .long 2147483648
; AVX: [[MAGMASK2:L.+]]:
; AVX-NEXT: .long 2147483647
; AVX-NEXT: .long 2147483647
; AVX-NEXT: .long 2147483647
; AVX-NEXT: .long 2147483647
; AVX-NEXT: .long 2147483647
; AVX-NEXT: .long 2147483647
; AVX-NEXT: .long 2147483647
; AVX-NEXT: .long 2147483647
[x86] add test to show unnecessary scalarization of copysign intrinsics (PR30433) llvm-svn: 283071
2016-10-03 00:31:35 +08:00
define <8 x float> @v8f32(<8 x float> %a, <8 x float> %b) nounwind {
; SSE2-LABEL: v8f32:
; SSE2: # BB#0:
[x86, SSE/AVX] allow 128/256-bit lowering for copysign vector intrinsics (PR30433) This should fix: https://llvm.org/bugs/show_bug.cgi?id=30433 There are a couple of open questions about the codegen: 1. Should we let scalar ops be scalars and avoid vector constant loads/splats? 2. Should we have a pass to combine constants such as the inverted pair that we have here? Differential Revision: https://reviews.llvm.org/D25165 llvm-svn: 283119
2016-10-04 00:38:27 +08:00
; SSE2-NEXT: movaps [[SIGNMASK2]](%rip), %xmm4
; SSE2-NEXT: andps %xmm4, %xmm2
; SSE2-NEXT: movaps [[MAGMASK2]](%rip), %xmm5
; SSE2-NEXT: andps %xmm5, %xmm0
; SSE2-NEXT: orps %xmm2, %xmm0
; SSE2-NEXT: andps %xmm4, %xmm3
; SSE2-NEXT: andps %xmm5, %xmm1
[x86] add test to show unnecessary scalarization of copysign intrinsics (PR30433) llvm-svn: 283071
2016-10-03 00:31:35 +08:00
; SSE2-NEXT: orps %xmm3, %xmm1
; SSE2-NEXT: retq
;
; AVX-LABEL: v8f32:
; AVX: # BB#0:
[x86, SSE/AVX] allow 128/256-bit lowering for copysign vector intrinsics (PR30433) This should fix: https://llvm.org/bugs/show_bug.cgi?id=30433 There are a couple of open questions about the codegen: 1. Should we let scalar ops be scalars and avoid vector constant loads/splats? 2. Should we have a pass to combine constants such as the inverted pair that we have here? Differential Revision: https://reviews.llvm.org/D25165 llvm-svn: 283119
2016-10-04 00:38:27 +08:00
; AVX-NEXT: vandps [[SIGNMASK2]](%rip), %ymm1, %ymm1
; AVX-NEXT: vandps [[MAGMASK2]](%rip), %ymm0, %ymm0
; AVX-NEXT: vorps %ymm1, %ymm0, %ymm0
[x86] add test to show unnecessary scalarization of copysign intrinsics (PR30433) llvm-svn: 283071
2016-10-03 00:31:35 +08:00
; AVX-NEXT: retq
;
%tmp = tail call <8 x float> @llvm.copysign.v8f32( <8 x float> %a, <8 x float> %b )
ret <8 x float> %tmp
}
[x86, SSE/AVX] allow 128/256-bit lowering for copysign vector intrinsics (PR30433) This should fix: https://llvm.org/bugs/show_bug.cgi?id=30433 There are a couple of open questions about the codegen: 1. Should we let scalar ops be scalars and avoid vector constant loads/splats? 2. Should we have a pass to combine constants such as the inverted pair that we have here? Differential Revision: https://reviews.llvm.org/D25165 llvm-svn: 283119
2016-10-04 00:38:27 +08:00
; CHECK: [[SIGNMASK3:L.+]]:
; CHECK-NEXT: .quad -9223372036854775808
; CHECK-NEXT: .quad -9223372036854775808
; CHECK: [[MAGMASK3:L.+]]:
; CHECK-NEXT: .quad 9223372036854775807
; CHECK-NEXT: .quad 9223372036854775807
[x86] add test to show unnecessary scalarization of copysign intrinsics (PR30433) llvm-svn: 283071
2016-10-03 00:31:35 +08:00
define <2 x double> @v2f64(<2 x double> %a, <2 x double> %b) nounwind {
; SSE2-LABEL: v2f64:
; SSE2: # BB#0:
[x86, SSE/AVX] allow 128/256-bit lowering for copysign vector intrinsics (PR30433) This should fix: https://llvm.org/bugs/show_bug.cgi?id=30433 There are a couple of open questions about the codegen: 1. Should we let scalar ops be scalars and avoid vector constant loads/splats? 2. Should we have a pass to combine constants such as the inverted pair that we have here? Differential Revision: https://reviews.llvm.org/D25165 llvm-svn: 283119
2016-10-04 00:38:27 +08:00
; SSE2-NEXT: andps [[SIGNMASK3]](%rip), %xmm1
; SSE2-NEXT: andps [[MAGMASK3]](%rip), %xmm0
[x86] add test to show unnecessary scalarization of copysign intrinsics (PR30433) llvm-svn: 283071
2016-10-03 00:31:35 +08:00
; SSE2-NEXT: orps %xmm1, %xmm0
; SSE2-NEXT: retq
;
; AVX-LABEL: v2f64:
; AVX: # BB#0:
[x86, SSE/AVX] allow 128/256-bit lowering for copysign vector intrinsics (PR30433) This should fix: https://llvm.org/bugs/show_bug.cgi?id=30433 There are a couple of open questions about the codegen: 1. Should we let scalar ops be scalars and avoid vector constant loads/splats? 2. Should we have a pass to combine constants such as the inverted pair that we have here? Differential Revision: https://reviews.llvm.org/D25165 llvm-svn: 283119
2016-10-04 00:38:27 +08:00
; AVX-NEXT: vandps [[SIGNMASK3]](%rip), %xmm1, %xmm1
; AVX-NEXT: vandps [[MAGMASK3]](%rip), %xmm0, %xmm0
; AVX-NEXT: vorps %xmm1, %xmm0, %xmm0
[x86] add test to show unnecessary scalarization of copysign intrinsics (PR30433) llvm-svn: 283071
2016-10-03 00:31:35 +08:00
; AVX-NEXT: retq
;
%tmp = tail call <2 x double> @llvm.copysign.v2f64( <2 x double> %a, <2 x double> %b )
ret <2 x double> %tmp
}
[x86, SSE/AVX] allow 128/256-bit lowering for copysign vector intrinsics (PR30433) This should fix: https://llvm.org/bugs/show_bug.cgi?id=30433 There are a couple of open questions about the codegen: 1. Should we let scalar ops be scalars and avoid vector constant loads/splats? 2. Should we have a pass to combine constants such as the inverted pair that we have here? Differential Revision: https://reviews.llvm.org/D25165 llvm-svn: 283119
2016-10-04 00:38:27 +08:00
; SSE2: [[SIGNMASK4:L.+]]:
; SSE2-NEXT: .quad -9223372036854775808
; SSE2-NEXT: .quad -9223372036854775808
; SSE2: [[MAGMASK4:L.+]]:
; SSE2-NEXT: .quad 9223372036854775807
; SSE2-NEXT: .quad 9223372036854775807
; AVX: [[SIGNMASK4:L.+]]:
; AVX-NEXT: .quad -9223372036854775808
; AVX-NEXT: .quad -9223372036854775808
; AVX-NEXT: .quad -9223372036854775808
; AVX-NEXT: .quad -9223372036854775808
; AVX: [[MAGMASK4:L.+]]:
; AVX-NEXT: .quad 9223372036854775807
; AVX-NEXT: .quad 9223372036854775807
; AVX-NEXT: .quad 9223372036854775807
; AVX-NEXT: .quad 9223372036854775807
[x86] add test to show unnecessary scalarization of copysign intrinsics (PR30433) llvm-svn: 283071
2016-10-03 00:31:35 +08:00
define <4 x double> @v4f64(<4 x double> %a, <4 x double> %b) nounwind {
; SSE2-LABEL: v4f64:
; SSE2: # BB#0:
[x86, SSE/AVX] allow 128/256-bit lowering for copysign vector intrinsics (PR30433) This should fix: https://llvm.org/bugs/show_bug.cgi?id=30433 There are a couple of open questions about the codegen: 1. Should we let scalar ops be scalars and avoid vector constant loads/splats? 2. Should we have a pass to combine constants such as the inverted pair that we have here? Differential Revision: https://reviews.llvm.org/D25165 llvm-svn: 283119
2016-10-04 00:38:27 +08:00
; SSE2-NEXT: movaps [[SIGNMASK4]](%rip), %xmm4
; SSE2-NEXT: andps %xmm4, %xmm2
; SSE2-NEXT: movaps [[MAGMASK4]](%rip), %xmm5
; SSE2-NEXT: andps %xmm5, %xmm0
; SSE2-NEXT: orps %xmm2, %xmm0
; SSE2-NEXT: andps %xmm4, %xmm3
; SSE2-NEXT: andps %xmm5, %xmm1
[x86] add test to show unnecessary scalarization of copysign intrinsics (PR30433) llvm-svn: 283071
2016-10-03 00:31:35 +08:00
; SSE2-NEXT: orps %xmm3, %xmm1
; SSE2-NEXT: retq
;
; AVX-LABEL: v4f64:
; AVX: # BB#0:
[x86, SSE/AVX] allow 128/256-bit lowering for copysign vector intrinsics (PR30433) This should fix: https://llvm.org/bugs/show_bug.cgi?id=30433 There are a couple of open questions about the codegen: 1. Should we let scalar ops be scalars and avoid vector constant loads/splats? 2. Should we have a pass to combine constants such as the inverted pair that we have here? Differential Revision: https://reviews.llvm.org/D25165 llvm-svn: 283119
2016-10-04 00:38:27 +08:00
; AVX-NEXT: vandps [[SIGNMASK4]](%rip), %ymm1, %ymm1
; AVX-NEXT: vandps [[MAGMASK4]](%rip), %ymm0, %ymm0
; AVX-NEXT: vorps %ymm1, %ymm0, %ymm0
[x86] add test to show unnecessary scalarization of copysign intrinsics (PR30433) llvm-svn: 283071
2016-10-03 00:31:35 +08:00
; AVX-NEXT: retq
;
%tmp = tail call <4 x double> @llvm.copysign.v4f64( <4 x double> %a, <4 x double> %b )
ret <4 x double> %tmp
}
declare <4 x float> @llvm.copysign.v4f32(<4 x float> %Mag, <4 x float> %Sgn)
declare <8 x float> @llvm.copysign.v8f32(<8 x float> %Mag, <8 x float> %Sgn)
declare <2 x double> @llvm.copysign.v2f64(<2 x double> %Mag, <2 x double> %Sgn)
declare <4 x double> @llvm.copysign.v4f64(<4 x double> %Mag, <4 x double> %Sgn)