This is an archive of the discontinued LLVM Phabricator instance.

Differential D38350

[X86][SSE] Added more tests for vector multiplications as utility for D37896
ClosedPublic

Authored by pacxx on Sep 28 2017, 3:12 AM.

Details

Reviewers
RKSimon
Commits
rG2b96841d1d00: [X86][SSE] Added more tests for vector multiplications as utility for D37896
rL314504: [X86][SSE] Added more tests for vector multiplications as utility for D37896
Summary

Added additional tests for vector multiplications with multipliers that are:

  1. powers of 2 displaced by 1,
  2. product of a power of 2 displaced by one with another power of 2.

Diff Detail

Repository
rL LLVM

Event Timeline

pacxx created this revision.Sep 28 2017, 3:12 AM
RKSimon removed a reviewer: llvm-commits.Sep 28 2017, 6:15 AM
RKSimon edited edge metadata.
RKSimon added a subscriber: llvm-commits.

What's going on here? It doesn't look like its been diffed against trunk.

Very minor, but please use [X86][SSE] instead of [Test] in your summary.

pacxx updated this revision to Diff 116982.Sep 28 2017, 7:08 AM
pacxx retitled this revision from [Test] Added more tests for vector multiplications as utility for D37896 to [X86][SSE] Added more tests for vector multiplications as utility for D37896.

Sorry somehow things messed up.

RKSimon accepted this revision.Sep 28 2017, 7:31 AM

LGTM - with one minor correction.

./test/CodeGen/X86/vector-mul.ll
1071 ↗(On Diff #116982)

v2i64 not v2i16

This revision is now accepted and ready to land.Sep 28 2017, 7:31 AM
pacxx updated this revision to Diff 116990.Sep 28 2017, 7:40 AM

fixed wrong type in function names

RKSimon accepted this revision.Sep 28 2017, 8:14 AM

LGTM, thanks

pacxx added a comment.Sep 28 2017, 11:32 PM

please commit the patch, so we can go on with D37896

Closed by commit rL314504: [X86][SSE] Added more tests for vector multiplications as utility for D37896 (authored by RKSimon). · Explain WhySep 29 2017, 3:03 AM
This revision was automatically updated to reflect the committed changes.

Revision Contents

PathSize
llvm/
trunk/
test/
CodeGen/
X86/
302 lines

Diff 117114

llvm/trunk/test/CodeGen/X86/vector-mul.ll

Show First 20 Lines • Show All 695 Lines • ▼ Show 20 Lines
; X64-AVX-NEXT: vpmuludq %xmm1, %xmm0, %xmm0 ; X64-AVX-NEXT: vpmuludq %xmm1, %xmm0, %xmm0
; X64-AVX-NEXT: vpsllq $32, %xmm0, %xmm0 ; X64-AVX-NEXT: vpsllq $32, %xmm0, %xmm0
; X64-AVX-NEXT: vpaddq %xmm0, %xmm2, %xmm0 ; X64-AVX-NEXT: vpaddq %xmm0, %xmm2, %xmm0
; X64-AVX-NEXT: retq ; X64-AVX-NEXT: retq
%1 = mul <2 x i64> %a0, <i64 15, i64 63> %1 = mul <2 x i64> %a0, <i64 15, i64 63>
ret <2 x i64> %1 ret <2 x i64> %1
} }
define <2 x i64> @mul_v2i64_neg_15_63(<2 x i64> %a0) nounwind {
; X86-LABEL: mul_v2i64_neg_15_63:
; X86: # BB#0:
; X86-NEXT: movdqa {{.*#+}} xmm1 = [4294967281,4294967295,4294967233,4294967295]
; X86-NEXT: movdqa %xmm0, %xmm2
; X86-NEXT: pmuludq %xmm1, %xmm2
; X86-NEXT: movdqa %xmm0, %xmm3
; X86-NEXT: psrlq $32, %xmm3
; X86-NEXT: pmuludq %xmm1, %xmm3
; X86-NEXT: pmuludq {{\.LCPI.*}}, %xmm0
; X86-NEXT: paddq %xmm3, %xmm0
; X86-NEXT: psllq $32, %xmm0
; X86-NEXT: paddq %xmm2, %xmm0
; X86-NEXT: retl
;
; X64-LABEL: mul_v2i64_neg_15_63:
; X64: # BB#0:
; X64-NEXT: movdqa {{.*#+}} xmm1 = [18446744073709551601,18446744073709551553]
; X64-NEXT: movdqa %xmm0, %xmm2
; X64-NEXT: pmuludq %xmm1, %xmm2
; X64-NEXT: movdqa %xmm0, %xmm3
; X64-NEXT: psrlq $32, %xmm3
; X64-NEXT: pmuludq %xmm1, %xmm3
; X64-NEXT: pmuludq {{.*}}(%rip), %xmm0
; X64-NEXT: paddq %xmm3, %xmm0
; X64-NEXT: psllq $32, %xmm0
; X64-NEXT: paddq %xmm2, %xmm0
; X64-NEXT: retq
;
; X64-AVX-LABEL: mul_v2i64_neg_15_63:
; X64-AVX: # BB#0:
; X64-AVX-NEXT: vmovdqa {{.*#+}} xmm1 = [18446744073709551601,18446744073709551553]
; X64-AVX-NEXT: vpmuludq %xmm1, %xmm0, %xmm2
; X64-AVX-NEXT: vpsrlq $32, %xmm0, %xmm3
; X64-AVX-NEXT: vpmuludq %xmm1, %xmm3, %xmm1
; X64-AVX-NEXT: vpmuludq {{.*}}(%rip), %xmm0, %xmm0
; X64-AVX-NEXT: vpaddq %xmm1, %xmm0, %xmm0
; X64-AVX-NEXT: vpsllq $32, %xmm0, %xmm0
; X64-AVX-NEXT: vpaddq %xmm0, %xmm2, %xmm0
; X64-AVX-NEXT: retq
%1 = mul <2 x i64> %a0, <i64 -15, i64 -63>
ret <2 x i64> %1
}
define <2 x i64> @mul_v2i64_neg_17_65(<2 x i64> %a0) nounwind {
; X86-LABEL: mul_v2i64_neg_17_65:
; X86: # BB#0:
; X86-NEXT: movdqa {{.*#+}} xmm1 = [4294967279,4294967295,4294967231,4294967295]
; X86-NEXT: movdqa %xmm0, %xmm2
; X86-NEXT: pmuludq %xmm1, %xmm2
; X86-NEXT: movdqa %xmm0, %xmm3
; X86-NEXT: psrlq $32, %xmm3
; X86-NEXT: pmuludq %xmm1, %xmm3
; X86-NEXT: pmuludq {{\.LCPI.*}}, %xmm0
; X86-NEXT: paddq %xmm3, %xmm0
; X86-NEXT: psllq $32, %xmm0
; X86-NEXT: paddq %xmm2, %xmm0
; X86-NEXT: retl
;
; X64-LABEL: mul_v2i64_neg_17_65:
; X64: # BB#0:
; X64-NEXT: movdqa {{.*#+}} xmm1 = [18446744073709551599,18446744073709551551]
; X64-NEXT: movdqa %xmm0, %xmm2
; X64-NEXT: pmuludq %xmm1, %xmm2
; X64-NEXT: movdqa %xmm0, %xmm3
; X64-NEXT: psrlq $32, %xmm3
; X64-NEXT: pmuludq %xmm1, %xmm3
; X64-NEXT: pmuludq {{.*}}(%rip), %xmm0
; X64-NEXT: paddq %xmm3, %xmm0
; X64-NEXT: psllq $32, %xmm0
; X64-NEXT: paddq %xmm2, %xmm0
; X64-NEXT: retq
;
; X64-AVX-LABEL: mul_v2i64_neg_17_65:
; X64-AVX: # BB#0:
; X64-AVX-NEXT: vmovdqa {{.*#+}} xmm1 = [18446744073709551599,18446744073709551551]
; X64-AVX-NEXT: vpmuludq %xmm1, %xmm0, %xmm2
; X64-AVX-NEXT: vpsrlq $32, %xmm0, %xmm3
; X64-AVX-NEXT: vpmuludq %xmm1, %xmm3, %xmm1
; X64-AVX-NEXT: vpmuludq {{.*}}(%rip), %xmm0, %xmm0
; X64-AVX-NEXT: vpaddq %xmm1, %xmm0, %xmm0
; X64-AVX-NEXT: vpsllq $32, %xmm0, %xmm0
; X64-AVX-NEXT: vpaddq %xmm0, %xmm2, %xmm0
; X64-AVX-NEXT: retq
%1 = mul <2 x i64> %a0, <i64 -17, i64 -65>
ret <2 x i64> %1
}
define <2 x i64> @mul_v2i64_0_1(<2 x i64> %a0) nounwind {
; X86-LABEL: mul_v2i64_0_1:
; X86: # BB#0:
; X86-NEXT: movdqa {{.*#+}} xmm1 = [0,0,1,0]
; X86-NEXT: movdqa %xmm0, %xmm2
; X86-NEXT: pmuludq %xmm1, %xmm2
; X86-NEXT: psrlq $32, %xmm0
; X86-NEXT: pmuludq %xmm1, %xmm0
; X86-NEXT: psllq $32, %xmm0
; X86-NEXT: paddq %xmm2, %xmm0
; X86-NEXT: retl
;
; X64-LABEL: mul_v2i64_0_1:
; X64: # BB#0:
; X64-NEXT: movl $1, %eax
; X64-NEXT: movq %rax, %xmm1
; X64-NEXT: pslldq {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0,1,2,3,4,5,6,7]
; X64-NEXT: movdqa %xmm0, %xmm2
; X64-NEXT: pmuludq %xmm1, %xmm2
; X64-NEXT: psrlq $32, %xmm0
; X64-NEXT: pmuludq %xmm1, %xmm0
; X64-NEXT: psllq $32, %xmm0
; X64-NEXT: paddq %xmm2, %xmm0
; X64-NEXT: retq
;
; X64-AVX-LABEL: mul_v2i64_0_1:
; X64-AVX: # BB#0:
; X64-AVX-NEXT: movl $1, %eax
; X64-AVX-NEXT: vmovq %rax, %xmm1
; X64-AVX-NEXT: vpslldq {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0,1,2,3,4,5,6,7]
; X64-AVX-NEXT: vpmuludq %xmm1, %xmm0, %xmm2
; X64-AVX-NEXT: vpsrlq $32, %xmm0, %xmm0
; X64-AVX-NEXT: vpmuludq %xmm1, %xmm0, %xmm0
; X64-AVX-NEXT: vpsllq $32, %xmm0, %xmm0
; X64-AVX-NEXT: vpaddq %xmm0, %xmm2, %xmm0
; X64-AVX-NEXT: retq
%1 = mul <2 x i64> %a0, <i64 0, i64 1>
ret <2 x i64> %1
}
define <2 x i64> @mul_v2i64_neg_0_1(<2 x i64> %a0) nounwind {
; X86-LABEL: mul_v2i64_neg_0_1:
; X86: # BB#0:
; X86-NEXT: movdqa {{.*#+}} xmm1 = [0,0,4294967295,4294967295]
; X86-NEXT: movdqa %xmm0, %xmm2
; X86-NEXT: pmuludq %xmm1, %xmm2
; X86-NEXT: movdqa %xmm0, %xmm3
; X86-NEXT: psrlq $32, %xmm3
; X86-NEXT: pmuludq %xmm1, %xmm3
; X86-NEXT: pmuludq {{\.LCPI.*}}, %xmm0
; X86-NEXT: paddq %xmm3, %xmm0
; X86-NEXT: psllq $32, %xmm0
; X86-NEXT: paddq %xmm2, %xmm0
; X86-NEXT: retl
;
; X64-LABEL: mul_v2i64_neg_0_1:
; X64: # BB#0:
; X64-NEXT: movdqa %xmm0, %xmm1
; X64-NEXT: psrlq $32, %xmm1
; X64-NEXT: movq $-1, %rax
; X64-NEXT: movq %rax, %xmm2
; X64-NEXT: pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0,1,2,3,4,5,6,7]
; X64-NEXT: pmuludq %xmm2, %xmm1
; X64-NEXT: movl $4294967295, %eax # imm = 0xFFFFFFFF
; X64-NEXT: movq %rax, %xmm3
; X64-NEXT: pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0,1,2,3,4,5,6,7]
; X64-NEXT: pmuludq %xmm0, %xmm3
; X64-NEXT: paddq %xmm1, %xmm3
; X64-NEXT: psllq $32, %xmm3
; X64-NEXT: pmuludq %xmm2, %xmm0
; X64-NEXT: paddq %xmm3, %xmm0
; X64-NEXT: retq
;
; X64-AVX-LABEL: mul_v2i64_neg_0_1:
; X64-AVX: # BB#0:
; X64-AVX-NEXT: vpsrlq $32, %xmm0, %xmm1
; X64-AVX-NEXT: movq $-1, %rax
; X64-AVX-NEXT: vmovq %rax, %xmm2
; X64-AVX-NEXT: vpslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0,1,2,3,4,5,6,7]
; X64-AVX-NEXT: vpmuludq %xmm2, %xmm1, %xmm1
; X64-AVX-NEXT: movl $4294967295, %eax # imm = 0xFFFFFFFF
; X64-AVX-NEXT: vmovq %rax, %xmm3
; X64-AVX-NEXT: vpslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0,1,2,3,4,5,6,7]
; X64-AVX-NEXT: vpmuludq %xmm3, %xmm0, %xmm3
; X64-AVX-NEXT: vpaddq %xmm1, %xmm3, %xmm1
; X64-AVX-NEXT: vpsllq $32, %xmm1, %xmm1
; X64-AVX-NEXT: vpmuludq %xmm2, %xmm0, %xmm0
; X64-AVX-NEXT: vpaddq %xmm1, %xmm0, %xmm0
; X64-AVX-NEXT: retq
%1 = mul <2 x i64> %a0, <i64 0, i64 -1>
ret <2 x i64> %1
}
define <2 x i64> @mul_v2i64_15_neg_63(<2 x i64> %a0) nounwind {
; X86-LABEL: mul_v2i64_15_neg_63:
; X86: # BB#0:
; X86-NEXT: movdqa {{.*#+}} xmm1 = [15,0,4294967233,4294967295]
; X86-NEXT: movdqa %xmm0, %xmm2
; X86-NEXT: pmuludq %xmm1, %xmm2
; X86-NEXT: movdqa %xmm0, %xmm3
; X86-NEXT: psrlq $32, %xmm3
; X86-NEXT: pmuludq %xmm1, %xmm3
; X86-NEXT: pmuludq {{\.LCPI.*}}, %xmm0
; X86-NEXT: paddq %xmm3, %xmm0
; X86-NEXT: psllq $32, %xmm0
; X86-NEXT: paddq %xmm2, %xmm0
; X86-NEXT: retl
;
; X64-LABEL: mul_v2i64_15_neg_63:
; X64: # BB#0:
; X64-NEXT: movdqa %xmm0, %xmm1
; X64-NEXT: psrlq $32, %xmm1
; X64-NEXT: movdqa {{.*#+}} xmm2 = [15,18446744073709551553]
; X64-NEXT: pmuludq %xmm2, %xmm1
; X64-NEXT: movl $4294967295, %eax # imm = 0xFFFFFFFF
; X64-NEXT: movq %rax, %xmm3
; X64-NEXT: pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0,1,2,3,4,5,6,7]
; X64-NEXT: pmuludq %xmm0, %xmm3
; X64-NEXT: paddq %xmm1, %xmm3
; X64-NEXT: psllq $32, %xmm3
; X64-NEXT: pmuludq %xmm2, %xmm0
; X64-NEXT: paddq %xmm3, %xmm0
; X64-NEXT: retq
;
; X64-AVX-LABEL: mul_v2i64_15_neg_63:
; X64-AVX: # BB#0:
; X64-AVX-NEXT: vpsrlq $32, %xmm0, %xmm1
; X64-AVX-NEXT: vmovdqa {{.*#+}} xmm2 = [15,18446744073709551553]
; X64-AVX-NEXT: vpmuludq %xmm2, %xmm1, %xmm1
; X64-AVX-NEXT: movl $4294967295, %eax # imm = 0xFFFFFFFF
; X64-AVX-NEXT: vmovq %rax, %xmm3
; X64-AVX-NEXT: vpslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0,1,2,3,4,5,6,7]
; X64-AVX-NEXT: vpmuludq %xmm3, %xmm0, %xmm3
; X64-AVX-NEXT: vpaddq %xmm1, %xmm3, %xmm1
; X64-AVX-NEXT: vpsllq $32, %xmm1, %xmm1
; X64-AVX-NEXT: vpmuludq %xmm2, %xmm0, %xmm0
; X64-AVX-NEXT: vpaddq %xmm1, %xmm0, %xmm0
; X64-AVX-NEXT: retq
%1 = mul <2 x i64> %a0, <i64 15, i64 -63>
ret <2 x i64> %1
}
define <4 x i32> @mul_v4i32_0_15_31_7(<4 x i32> %a0) nounwind { define <4 x i32> @mul_v4i32_0_15_31_7(<4 x i32> %a0) nounwind {
; X86-LABEL: mul_v4i32_0_15_31_7: ; X86-LABEL: mul_v4i32_0_15_31_7:
; X86: # BB#0: ; X86: # BB#0:
; X86-NEXT: pmulld {{\.LCPI.*}}, %xmm0 ; X86-NEXT: pmulld {{\.LCPI.*}}, %xmm0
; X86-NEXT: retl ; X86-NEXT: retl
; ;
; X64-LABEL: mul_v4i32_0_15_31_7: ; X64-LABEL: mul_v4i32_0_15_31_7:
; X64: # BB#0: ; X64: # BB#0:
▲ Show 20 Lines • Show All 81 Lines • ▼ Show 20 Lines
; X64-AVX2-NEXT: vpshufb %xmm2, %xmm1, %xmm1 ; X64-AVX2-NEXT: vpshufb %xmm2, %xmm1, %xmm1
; X64-AVX2-NEXT: vpshufb %xmm2, %xmm0, %xmm0 ; X64-AVX2-NEXT: vpshufb %xmm2, %xmm0, %xmm0
; X64-AVX2-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] ; X64-AVX2-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
; X64-AVX2-NEXT: vzeroupper ; X64-AVX2-NEXT: vzeroupper
; X64-AVX2-NEXT: retq ; X64-AVX2-NEXT: retq
%1 = mul <16 x i8> %a0, <i8 0, i8 1, i8 3, i8 7, i8 15, i8 31, i8 63, i8 127, i8 0, i8 1, i8 3, i8 7, i8 15, i8 31, i8 63, i8 127> %1 = mul <16 x i8> %a0, <i8 0, i8 1, i8 3, i8 7, i8 15, i8 31, i8 63, i8 127, i8 0, i8 1, i8 3, i8 7, i8 15, i8 31, i8 63, i8 127>
ret <16 x i8> %1 ret <16 x i8> %1
} }
define <2 x i64> @mul_v2i64_68_132(<2 x i64> %x) nounwind {
; X86-LABEL: mul_v2i64_68_132:
; X86: # BB#0:
; X86-NEXT: movdqa {{.*#+}} xmm1 = [68,0,132,0]
; X86-NEXT: movdqa %xmm0, %xmm2
; X86-NEXT: pmuludq %xmm1, %xmm2
; X86-NEXT: psrlq $32, %xmm0
; X86-NEXT: pmuludq %xmm1, %xmm0
; X86-NEXT: psllq $32, %xmm0
; X86-NEXT: paddq %xmm2, %xmm0
; X86-NEXT: retl
;
; X64-LABEL: mul_v2i64_68_132:
; X64: # BB#0:
; X64-NEXT: movdqa {{.*#+}} xmm1 = [68,132]
; X64-NEXT: movdqa %xmm0, %xmm2
; X64-NEXT: pmuludq %xmm1, %xmm2
; X64-NEXT: psrlq $32, %xmm0
; X64-NEXT: pmuludq %xmm1, %xmm0
; X64-NEXT: psllq $32, %xmm0
; X64-NEXT: paddq %xmm2, %xmm0
; X64-NEXT: retq
;
; X64-AVX-LABEL: mul_v2i64_68_132:
; X64-AVX: # BB#0:
; X64-AVX-NEXT: vmovdqa {{.*#+}} xmm1 = [68,132]
; X64-AVX-NEXT: vpmuludq %xmm1, %xmm0, %xmm2
; X64-AVX-NEXT: vpsrlq $32, %xmm0, %xmm0
; X64-AVX-NEXT: vpmuludq %xmm1, %xmm0, %xmm0
; X64-AVX-NEXT: vpsllq $32, %xmm0, %xmm0
; X64-AVX-NEXT: vpaddq %xmm0, %xmm2, %xmm0
; X64-AVX-NEXT: retq
%mul = mul <2 x i64> %x, <i64 68, i64 132>
ret <2 x i64> %mul
}
define <2 x i64> @mul_v2i64_60_120(<2 x i64> %x) nounwind {
; X86-LABEL: mul_v2i64_60_120:
; X86: # BB#0:
; X86-NEXT: movdqa {{.*#+}} xmm1 = [60,0,124,0]
; X86-NEXT: movdqa %xmm0, %xmm2
; X86-NEXT: pmuludq %xmm1, %xmm2
; X86-NEXT: psrlq $32, %xmm0
; X86-NEXT: pmuludq %xmm1, %xmm0
; X86-NEXT: psllq $32, %xmm0
; X86-NEXT: paddq %xmm2, %xmm0
; X86-NEXT: retl
;
; X64-LABEL: mul_v2i64_60_120:
; X64: # BB#0:
; X64-NEXT: movdqa {{.*#+}} xmm1 = [60,124]
; X64-NEXT: movdqa %xmm0, %xmm2
; X64-NEXT: pmuludq %xmm1, %xmm2
; X64-NEXT: psrlq $32, %xmm0
; X64-NEXT: pmuludq %xmm1, %xmm0
; X64-NEXT: psllq $32, %xmm0
; X64-NEXT: paddq %xmm2, %xmm0
; X64-NEXT: retq
;
; X64-AVX-LABEL: mul_v2i64_60_120:
; X64-AVX: # BB#0:
; X64-AVX-NEXT: vmovdqa {{.*#+}} xmm1 = [60,124]
; X64-AVX-NEXT: vpmuludq %xmm1, %xmm0, %xmm2
; X64-AVX-NEXT: vpsrlq $32, %xmm0, %xmm0
; X64-AVX-NEXT: vpmuludq %xmm1, %xmm0, %xmm0
; X64-AVX-NEXT: vpsllq $32, %xmm0, %xmm0
; X64-AVX-NEXT: vpaddq %xmm0, %xmm2, %xmm0
; X64-AVX-NEXT: retq
%mul = mul <2 x i64> %x, <i64 60, i64 124>
ret <2 x i64> %mul
}