[llvm] r342838 - [x86] add tests for mul decomposition with negative constant; NFC
Sanjay Patel via llvm-commits
llvm-commits at lists.llvm.org
Sun Sep 23 09:07:46 PDT 2018
Author: spatel
Date: Sun Sep 23 09:07:46 2018
New Revision: 342838
URL: http://llvm.org/viewvc/llvm-project?rev=342838&view=rev
Log:
[x86] add tests for mul decomposition with negative constant; NFC
Modified:
llvm/trunk/test/CodeGen/X86/vector-mul.ll
Modified: llvm/trunk/test/CodeGen/X86/vector-mul.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vector-mul.ll?rev=342838&r1=342837&r2=342838&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/vector-mul.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vector-mul.ll Sun Sep 23 09:07:46 2018
@@ -335,6 +335,152 @@ define <16 x i8> @mul_v16i8_17(<16 x i8>
}
;
+; -(PowOf2 + 1) (uniform)
+;
+
+define <2 x i64> @mul_v2i64_neg1025(<2 x i64> %a0) nounwind {
+; X86-LABEL: mul_v2i64_neg1025:
+; X86: # %bb.0:
+; X86-NEXT: movdqa {{.*#+}} xmm1 = [4294966271,4294967295,4294966271,4294967295]
+; 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_neg1025:
+; X64: # %bb.0:
+; X64-NEXT: movdqa {{.*#+}} xmm1 = [18446744073709550591,18446744073709550591]
+; 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_neg1025:
+; X64-AVX: # %bb.0:
+; X64-AVX-NEXT: vmovdqa {{.*#+}} xmm1 = [18446744073709550591,18446744073709550591]
+; 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 -1025, i64 -1025>
+ ret <2 x i64> %1
+}
+
+define <4 x i32> @mul_v4i32_neg33(<4 x i32> %a0) nounwind {
+; X86-LABEL: mul_v4i32_neg33:
+; X86: # %bb.0:
+; X86-NEXT: pmulld {{\.LCPI.*}}, %xmm0
+; X86-NEXT: retl
+;
+; X64-LABEL: mul_v4i32_neg33:
+; X64: # %bb.0:
+; X64-NEXT: pmulld {{.*}}(%rip), %xmm0
+; X64-NEXT: retq
+;
+; X64-XOP-LABEL: mul_v4i32_neg33:
+; X64-XOP: # %bb.0:
+; X64-XOP-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0
+; X64-XOP-NEXT: retq
+;
+; X64-AVX2-LABEL: mul_v4i32_neg33:
+; X64-AVX2: # %bb.0:
+; X64-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm1 = [4294967263,4294967263,4294967263,4294967263]
+; X64-AVX2-NEXT: vpmulld %xmm1, %xmm0, %xmm0
+; X64-AVX2-NEXT: retq
+ %1 = mul <4 x i32> %a0, <i32 -33, i32 -33, i32 -33, i32 -33>
+ ret <4 x i32> %1
+}
+
+define <8 x i16> @mul_v8i16_neg9(<8 x i16> %a0) nounwind {
+; X86-LABEL: mul_v8i16_neg9:
+; X86: # %bb.0:
+; X86-NEXT: pmullw {{\.LCPI.*}}, %xmm0
+; X86-NEXT: retl
+;
+; X64-LABEL: mul_v8i16_neg9:
+; X64: # %bb.0:
+; X64-NEXT: pmullw {{.*}}(%rip), %xmm0
+; X64-NEXT: retq
+;
+; X64-AVX-LABEL: mul_v8i16_neg9:
+; X64-AVX: # %bb.0:
+; X64-AVX-NEXT: vpmullw {{.*}}(%rip), %xmm0, %xmm0
+; X64-AVX-NEXT: retq
+ %1 = mul <8 x i16> %a0, <i16 -9, i16 -9, i16 -9, i16 -9, i16 -9, i16 -9, i16 -9, i16 -9>
+ ret <8 x i16> %1
+}
+
+define <16 x i8> @mul_v16i8_neg5(<16 x i8> %a0) nounwind {
+; X86-LABEL: mul_v16i8_neg5:
+; X86: # %bb.0:
+; X86-NEXT: pmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; X86-NEXT: punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X86-NEXT: movdqa {{.*#+}} xmm2 = [251,251,251,251,251,251,251,251,251,251,251,251,251,251,251,251]
+; X86-NEXT: pmullw %xmm2, %xmm0
+; X86-NEXT: movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
+; X86-NEXT: pand %xmm3, %xmm0
+; X86-NEXT: pmullw %xmm2, %xmm1
+; X86-NEXT: pand %xmm3, %xmm1
+; X86-NEXT: packuswb %xmm0, %xmm1
+; X86-NEXT: movdqa %xmm1, %xmm0
+; X86-NEXT: retl
+;
+; X64-LABEL: mul_v16i8_neg5:
+; X64: # %bb.0:
+; X64-NEXT: pmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; X64-NEXT: punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X64-NEXT: movdqa {{.*#+}} xmm2 = [251,251,251,251,251,251,251,251,251,251,251,251,251,251,251,251]
+; X64-NEXT: pmullw %xmm2, %xmm0
+; X64-NEXT: movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
+; X64-NEXT: pand %xmm3, %xmm0
+; X64-NEXT: pmullw %xmm2, %xmm1
+; X64-NEXT: pand %xmm3, %xmm1
+; X64-NEXT: packuswb %xmm0, %xmm1
+; X64-NEXT: movdqa %xmm1, %xmm0
+; X64-NEXT: retq
+;
+; X64-XOP-LABEL: mul_v16i8_neg5:
+; X64-XOP: # %bb.0:
+; X64-XOP-NEXT: vpunpckhbw {{.*#+}} xmm1 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X64-XOP-NEXT: vmovdqa {{.*#+}} xmm2 = [251,251,251,251,251,251,251,251,251,251,251,251,251,251,251,251]
+; X64-XOP-NEXT: vpmullw %xmm2, %xmm1, %xmm1
+; X64-XOP-NEXT: vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; X64-XOP-NEXT: vpmullw %xmm2, %xmm0, %xmm0
+; X64-XOP-NEXT: vpperm {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],xmm1[0,2,4,6,8,10,12,14]
+; X64-XOP-NEXT: retq
+;
+; X64-AVX2-LABEL: mul_v16i8_neg5:
+; X64-AVX2: # %bb.0:
+; X64-AVX2-NEXT: vpmovsxbw %xmm0, %ymm0
+; X64-AVX2-NEXT: vpmullw {{.*}}(%rip), %ymm0, %ymm0
+; X64-AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
+; X64-AVX2-NEXT: vmovdqa {{.*#+}} xmm2 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>
+; X64-AVX2-NEXT: vpshufb %xmm2, %xmm1, %xmm1
+; X64-AVX2-NEXT: vpshufb %xmm2, %xmm0, %xmm0
+; X64-AVX2-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; X64-AVX2-NEXT: vzeroupper
+; X64-AVX2-NEXT: retq
+ %1 = mul <16 x i8> %a0, <i8 -5, i8 -5, i8 -5, i8 -5, i8 -5, i8 -5, i8 -5, i8 -5, i8 -5, i8 -5, i8 -5, i8 -5, i8 -5, i8 -5, i8 -5, i8 -5>
+ ret <16 x i8> %1
+}
+
+;
; PowOf2 + 1 (non-uniform)
;
@@ -578,6 +724,152 @@ define <16 x i8> @mul_v16i8_31(<16 x i8>
ret <16 x i8> %1
}
+;
+; -(PowOf2 - 1) (uniform)
+;
+
+define <2 x i64> @mul_v2i64_neg7(<2 x i64> %a0) nounwind {
+; X86-LABEL: mul_v2i64_neg7:
+; X86: # %bb.0:
+; X86-NEXT: movdqa {{.*#+}} xmm1 = [4294967289,4294967295,4294967289,4294967295]
+; 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_neg7:
+; X64: # %bb.0:
+; X64-NEXT: movdqa {{.*#+}} xmm1 = [18446744073709551609,18446744073709551609]
+; 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_neg7:
+; X64-AVX: # %bb.0:
+; X64-AVX-NEXT: vmovdqa {{.*#+}} xmm1 = [18446744073709551609,18446744073709551609]
+; 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 -7, i64 -7>
+ ret <2 x i64> %1
+}
+
+define <4 x i32> @mul_v4i32_neg63(<4 x i32> %a0) nounwind {
+; X86-LABEL: mul_v4i32_neg63:
+; X86: # %bb.0:
+; X86-NEXT: pmulld {{\.LCPI.*}}, %xmm0
+; X86-NEXT: retl
+;
+; X64-LABEL: mul_v4i32_neg63:
+; X64: # %bb.0:
+; X64-NEXT: pmulld {{.*}}(%rip), %xmm0
+; X64-NEXT: retq
+;
+; X64-XOP-LABEL: mul_v4i32_neg63:
+; X64-XOP: # %bb.0:
+; X64-XOP-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0
+; X64-XOP-NEXT: retq
+;
+; X64-AVX2-LABEL: mul_v4i32_neg63:
+; X64-AVX2: # %bb.0:
+; X64-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm1 = [4294967233,4294967233,4294967233,4294967233]
+; X64-AVX2-NEXT: vpmulld %xmm1, %xmm0, %xmm0
+; X64-AVX2-NEXT: retq
+ %1 = mul <4 x i32> %a0, <i32 -63, i32 -63, i32 -63, i32 -63>
+ ret <4 x i32> %1
+}
+
+define <8 x i16> @mul_v8i16_neg31(<8 x i16> %a0) nounwind {
+; X86-LABEL: mul_v8i16_neg31:
+; X86: # %bb.0:
+; X86-NEXT: pmullw {{\.LCPI.*}}, %xmm0
+; X86-NEXT: retl
+;
+; X64-LABEL: mul_v8i16_neg31:
+; X64: # %bb.0:
+; X64-NEXT: pmullw {{.*}}(%rip), %xmm0
+; X64-NEXT: retq
+;
+; X64-AVX-LABEL: mul_v8i16_neg31:
+; X64-AVX: # %bb.0:
+; X64-AVX-NEXT: vpmullw {{.*}}(%rip), %xmm0, %xmm0
+; X64-AVX-NEXT: retq
+ %1 = mul <8 x i16> %a0, <i16 -31, i16 -31, i16 -31, i16 -31, i16 -31, i16 -31, i16 -31, i16 -31>
+ ret <8 x i16> %1
+}
+
+define <16 x i8> @mul_v16i8_neg15(<16 x i8> %a0) nounwind {
+; X86-LABEL: mul_v16i8_neg15:
+; X86: # %bb.0:
+; X86-NEXT: pmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; X86-NEXT: punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X86-NEXT: movdqa {{.*#+}} xmm2 = [241,241,241,241,241,241,241,241,241,241,241,241,241,241,241,241]
+; X86-NEXT: pmullw %xmm2, %xmm0
+; X86-NEXT: movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
+; X86-NEXT: pand %xmm3, %xmm0
+; X86-NEXT: pmullw %xmm2, %xmm1
+; X86-NEXT: pand %xmm3, %xmm1
+; X86-NEXT: packuswb %xmm0, %xmm1
+; X86-NEXT: movdqa %xmm1, %xmm0
+; X86-NEXT: retl
+;
+; X64-LABEL: mul_v16i8_neg15:
+; X64: # %bb.0:
+; X64-NEXT: pmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; X64-NEXT: punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X64-NEXT: movdqa {{.*#+}} xmm2 = [241,241,241,241,241,241,241,241,241,241,241,241,241,241,241,241]
+; X64-NEXT: pmullw %xmm2, %xmm0
+; X64-NEXT: movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
+; X64-NEXT: pand %xmm3, %xmm0
+; X64-NEXT: pmullw %xmm2, %xmm1
+; X64-NEXT: pand %xmm3, %xmm1
+; X64-NEXT: packuswb %xmm0, %xmm1
+; X64-NEXT: movdqa %xmm1, %xmm0
+; X64-NEXT: retq
+;
+; X64-XOP-LABEL: mul_v16i8_neg15:
+; X64-XOP: # %bb.0:
+; X64-XOP-NEXT: vpunpckhbw {{.*#+}} xmm1 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X64-XOP-NEXT: vmovdqa {{.*#+}} xmm2 = [241,241,241,241,241,241,241,241,241,241,241,241,241,241,241,241]
+; X64-XOP-NEXT: vpmullw %xmm2, %xmm1, %xmm1
+; X64-XOP-NEXT: vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; X64-XOP-NEXT: vpmullw %xmm2, %xmm0, %xmm0
+; X64-XOP-NEXT: vpperm {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],xmm1[0,2,4,6,8,10,12,14]
+; X64-XOP-NEXT: retq
+;
+; X64-AVX2-LABEL: mul_v16i8_neg15:
+; X64-AVX2: # %bb.0:
+; X64-AVX2-NEXT: vpmovsxbw %xmm0, %ymm0
+; X64-AVX2-NEXT: vpmullw {{.*}}(%rip), %ymm0, %ymm0
+; X64-AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
+; X64-AVX2-NEXT: vmovdqa {{.*#+}} xmm2 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>
+; X64-AVX2-NEXT: vpshufb %xmm2, %xmm1, %xmm1
+; X64-AVX2-NEXT: vpshufb %xmm2, %xmm0, %xmm0
+; X64-AVX2-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; X64-AVX2-NEXT: vzeroupper
+; X64-AVX2-NEXT: retq
+ %1 = mul <16 x i8> %a0, <i8 -15, i8 -15, i8 -15, i8 -15, i8 -15, i8 -15, i8 -15, i8 -15, i8 -15, i8 -15, i8 -15, i8 -15, i8 -15, i8 -15, i8 -15, i8 -15>
+ ret <16 x i8> %1
+}
+
;
; PowOf2 - 1 (non-uniform)
;
More information about the llvm-commits
mailing list