1 files changed, 216 insertions, 0 deletions
diff --git a/llvm/test/CodeGen/X86/vector-idiv-v2i32.ll b/llvm/test/CodeGen/X86/vector-idiv-v2i32.ll
index 6f6776ea89e..a76c4d10afa 100644
--- a/llvm/test/CodeGen/X86/vector-idiv-v2i32.ll
+++ b/llvm/test/CodeGen/X86/vector-idiv-v2i32.ll
@@ -1,6 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X64
 ; RUN: llc < %s -mtriple=i386-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X86
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 -x86-experimental-vector-widening-legalization | FileCheck %s --check-prefix=X64_WIDEN
+; RUN: llc < %s -mtriple=i386-unknown-unknown -mattr=+sse2 -x86-experimental-vector-widening-legalization | FileCheck %s --check-prefix=X86_WIDEN
 
 define void @test_udiv7_v2i32(<2 x i32>* %x, <2 x i32>* %y) nounwind {
 ; X64-LABEL: test_udiv7_v2i32:
@@ -41,6 +43,45 @@ define void @test_udiv7_v2i32(<2 x i32>* %x, <2 x i32>* %y) nounwind {
 ; X86-NEXT:    psrld $2, %xmm0
 ; X86-NEXT:    movq %xmm0, (%eax)
 ; X86-NEXT:    retl
+;
+; X64_WIDEN-LABEL: test_udiv7_v2i32:
+; X64_WIDEN:       # %bb.0:
+; X64_WIDEN-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; X64_WIDEN-NEXT:    movdqa {{.*#+}} xmm1 = [613566757,613566757,613566757,613566757]
+; X64_WIDEN-NEXT:    movdqa %xmm0, %xmm2
+; X64_WIDEN-NEXT:    pmuludq %xmm1, %xmm2
+; X64_WIDEN-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
+; X64_WIDEN-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
+; X64_WIDEN-NEXT:    pmuludq %xmm1, %xmm3
+; X64_WIDEN-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[1,3,2,3]
+; X64_WIDEN-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; X64_WIDEN-NEXT:    psubd %xmm2, %xmm0
+; X64_WIDEN-NEXT:    psrld $1, %xmm0
+; X64_WIDEN-NEXT:    paddd %xmm2, %xmm0
+; X64_WIDEN-NEXT:    psrld $2, %xmm0
+; X64_WIDEN-NEXT:    movq %xmm0, (%rsi)
+; X64_WIDEN-NEXT:    retq
+;
+; X86_WIDEN-LABEL: test_udiv7_v2i32:
+; X86_WIDEN:       # %bb.0:
+; X86_WIDEN-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86_WIDEN-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86_WIDEN-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; X86_WIDEN-NEXT:    movdqa {{.*#+}} xmm1 = [613566757,613566757,613566757,613566757]
+; X86_WIDEN-NEXT:    movdqa %xmm0, %xmm2
+; X86_WIDEN-NEXT:    pmuludq %xmm1, %xmm2
+; X86_WIDEN-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
+; X86_WIDEN-NEXT:    movdqa %xmm0, %xmm3
+; X86_WIDEN-NEXT:    shufps {{.*#+}} xmm3 = xmm3[1,1],xmm0[2,3]
+; X86_WIDEN-NEXT:    pmuludq %xmm1, %xmm3
+; X86_WIDEN-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[1,3,2,3]
+; X86_WIDEN-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; X86_WIDEN-NEXT:    psubd %xmm2, %xmm0
+; X86_WIDEN-NEXT:    psrld $1, %xmm0
+; X86_WIDEN-NEXT:    paddd %xmm2, %xmm0
+; X86_WIDEN-NEXT:    psrld $2, %xmm0
+; X86_WIDEN-NEXT:    movq %xmm0, (%eax)
+; X86_WIDEN-NEXT:    retl
   %a = load <2 x i32>, <2 x i32>* %x
   %b = udiv <2 x i32> %a, <i32 7, i32 7>
   store <2 x i32> %b, <2 x i32>* %y
@@ -96,6 +137,55 @@ define void @test_urem7_v2i32(<2 x i32>* %x, <2 x i32>* %y) nounwind {
 ; X86-NEXT:    paddd %xmm0, %xmm1
 ; X86-NEXT:    movq %xmm1, (%eax)
 ; X86-NEXT:    retl
+;
+; X64_WIDEN-LABEL: test_urem7_v2i32:
+; X64_WIDEN:       # %bb.0:
+; X64_WIDEN-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; X64_WIDEN-NEXT:    movdqa {{.*#+}} xmm1 = [613566757,613566757,613566757,613566757]
+; X64_WIDEN-NEXT:    movdqa %xmm0, %xmm2
+; X64_WIDEN-NEXT:    pmuludq %xmm1, %xmm2
+; X64_WIDEN-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
+; X64_WIDEN-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
+; X64_WIDEN-NEXT:    pmuludq %xmm1, %xmm3
+; X64_WIDEN-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[1,3,2,3]
+; X64_WIDEN-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; X64_WIDEN-NEXT:    movdqa %xmm0, %xmm1
+; X64_WIDEN-NEXT:    psubd %xmm2, %xmm1
+; X64_WIDEN-NEXT:    psrld $1, %xmm1
+; X64_WIDEN-NEXT:    paddd %xmm2, %xmm1
+; X64_WIDEN-NEXT:    psrld $2, %xmm1
+; X64_WIDEN-NEXT:    movdqa %xmm1, %xmm2
+; X64_WIDEN-NEXT:    pslld $3, %xmm2
+; X64_WIDEN-NEXT:    psubd %xmm2, %xmm1
+; X64_WIDEN-NEXT:    paddd %xmm0, %xmm1
+; X64_WIDEN-NEXT:    movq %xmm1, (%rsi)
+; X64_WIDEN-NEXT:    retq
+;
+; X86_WIDEN-LABEL: test_urem7_v2i32:
+; X86_WIDEN:       # %bb.0:
+; X86_WIDEN-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86_WIDEN-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86_WIDEN-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; X86_WIDEN-NEXT:    movdqa {{.*#+}} xmm1 = [613566757,613566757,613566757,613566757]
+; X86_WIDEN-NEXT:    movdqa %xmm0, %xmm2
+; X86_WIDEN-NEXT:    pmuludq %xmm1, %xmm2
+; X86_WIDEN-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
+; X86_WIDEN-NEXT:    movdqa %xmm0, %xmm3
+; X86_WIDEN-NEXT:    shufps {{.*#+}} xmm3 = xmm3[1,1],xmm0[2,3]
+; X86_WIDEN-NEXT:    pmuludq %xmm1, %xmm3
+; X86_WIDEN-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[1,3,2,3]
+; X86_WIDEN-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; X86_WIDEN-NEXT:    movdqa %xmm0, %xmm1
+; X86_WIDEN-NEXT:    psubd %xmm2, %xmm1
+; X86_WIDEN-NEXT:    psrld $1, %xmm1
+; X86_WIDEN-NEXT:    paddd %xmm2, %xmm1
+; X86_WIDEN-NEXT:    psrld $2, %xmm1
+; X86_WIDEN-NEXT:    movdqa %xmm1, %xmm2
+; X86_WIDEN-NEXT:    pslld $3, %xmm2
+; X86_WIDEN-NEXT:    psubd %xmm2, %xmm1
+; X86_WIDEN-NEXT:    paddd %xmm0, %xmm1
+; X86_WIDEN-NEXT:    movq %xmm1, (%eax)
+; X86_WIDEN-NEXT:    retl
   %a = load <2 x i32>, <2 x i32>* %x
   %b = urem <2 x i32> %a, <i32 7, i32 7>
   store <2 x i32> %b, <2 x i32>* %y
@@ -153,6 +243,57 @@ define void @test_sdiv7_v2i32(<2 x i32>* %x, <2 x i32>* %y) nounwind {
 ; X86-NEXT:    paddd %xmm0, %xmm2
 ; X86-NEXT:    movq %xmm2, (%eax)
 ; X86-NEXT:    retl
+;
+; X64_WIDEN-LABEL: test_sdiv7_v2i32:
+; X64_WIDEN:       # %bb.0:
+; X64_WIDEN-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; X64_WIDEN-NEXT:    movdqa {{.*#+}} xmm1 = [2454267027,2454267027,2454267027,2454267027]
+; X64_WIDEN-NEXT:    movdqa %xmm0, %xmm2
+; X64_WIDEN-NEXT:    pmuludq %xmm1, %xmm2
+; X64_WIDEN-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
+; X64_WIDEN-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
+; X64_WIDEN-NEXT:    pmuludq %xmm1, %xmm3
+; X64_WIDEN-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,3,2,3]
+; X64_WIDEN-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
+; X64_WIDEN-NEXT:    pxor %xmm3, %xmm3
+; X64_WIDEN-NEXT:    pcmpgtd %xmm0, %xmm3
+; X64_WIDEN-NEXT:    pand %xmm1, %xmm3
+; X64_WIDEN-NEXT:    paddd %xmm0, %xmm3
+; X64_WIDEN-NEXT:    psubd %xmm3, %xmm2
+; X64_WIDEN-NEXT:    paddd %xmm0, %xmm2
+; X64_WIDEN-NEXT:    movdqa %xmm2, %xmm0
+; X64_WIDEN-NEXT:    psrld $31, %xmm0
+; X64_WIDEN-NEXT:    psrad $2, %xmm2
+; X64_WIDEN-NEXT:    paddd %xmm0, %xmm2
+; X64_WIDEN-NEXT:    movq %xmm2, (%rsi)
+; X64_WIDEN-NEXT:    retq
+;
+; X86_WIDEN-LABEL: test_sdiv7_v2i32:
+; X86_WIDEN:       # %bb.0:
+; X86_WIDEN-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86_WIDEN-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86_WIDEN-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; X86_WIDEN-NEXT:    movdqa {{.*#+}} xmm1 = [2454267027,2454267027,2454267027,2454267027]
+; X86_WIDEN-NEXT:    movdqa %xmm0, %xmm2
+; X86_WIDEN-NEXT:    pmuludq %xmm1, %xmm2
+; X86_WIDEN-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
+; X86_WIDEN-NEXT:    movdqa %xmm0, %xmm3
+; X86_WIDEN-NEXT:    shufps {{.*#+}} xmm3 = xmm3[1,1],xmm0[2,3]
+; X86_WIDEN-NEXT:    pmuludq %xmm1, %xmm3
+; X86_WIDEN-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,3,2,3]
+; X86_WIDEN-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
+; X86_WIDEN-NEXT:    pxor %xmm3, %xmm3
+; X86_WIDEN-NEXT:    pcmpgtd %xmm0, %xmm3
+; X86_WIDEN-NEXT:    pand %xmm1, %xmm3
+; X86_WIDEN-NEXT:    paddd %xmm0, %xmm3
+; X86_WIDEN-NEXT:    psubd %xmm3, %xmm2
+; X86_WIDEN-NEXT:    paddd %xmm0, %xmm2
+; X86_WIDEN-NEXT:    movdqa %xmm2, %xmm0
+; X86_WIDEN-NEXT:    psrld $31, %xmm0
+; X86_WIDEN-NEXT:    psrad $2, %xmm2
+; X86_WIDEN-NEXT:    paddd %xmm0, %xmm2
+; X86_WIDEN-NEXT:    movq %xmm2, (%eax)
+; X86_WIDEN-NEXT:    retl
   %a = load <2 x i32>, <2 x i32>* %x
   %b = sdiv <2 x i32> %a, <i32 7, i32 7>
   store <2 x i32> %b, <2 x i32>* %y
@@ -218,6 +359,65 @@ define void @test_srem7_v2i32(<2 x i32>* %x, <2 x i32>* %y) nounwind {
 ; X86-NEXT:    paddd %xmm0, %xmm2
 ; X86-NEXT:    movq %xmm2, (%eax)
 ; X86-NEXT:    retl
+;
+; X64_WIDEN-LABEL: test_srem7_v2i32:
+; X64_WIDEN:       # %bb.0:
+; X64_WIDEN-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; X64_WIDEN-NEXT:    movdqa {{.*#+}} xmm1 = [2454267027,2454267027,2454267027,2454267027]
+; X64_WIDEN-NEXT:    movdqa %xmm0, %xmm2
+; X64_WIDEN-NEXT:    pmuludq %xmm1, %xmm2
+; X64_WIDEN-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
+; X64_WIDEN-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
+; X64_WIDEN-NEXT:    pmuludq %xmm1, %xmm3
+; X64_WIDEN-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,3,2,3]
+; X64_WIDEN-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
+; X64_WIDEN-NEXT:    pxor %xmm3, %xmm3
+; X64_WIDEN-NEXT:    pcmpgtd %xmm0, %xmm3
+; X64_WIDEN-NEXT:    pand %xmm1, %xmm3
+; X64_WIDEN-NEXT:    paddd %xmm0, %xmm3
+; X64_WIDEN-NEXT:    psubd %xmm3, %xmm2
+; X64_WIDEN-NEXT:    paddd %xmm0, %xmm2
+; X64_WIDEN-NEXT:    movdqa %xmm2, %xmm1
+; X64_WIDEN-NEXT:    psrld $31, %xmm1
+; X64_WIDEN-NEXT:    psrad $2, %xmm2
+; X64_WIDEN-NEXT:    paddd %xmm1, %xmm2
+; X64_WIDEN-NEXT:    movdqa %xmm2, %xmm1
+; X64_WIDEN-NEXT:    pslld $3, %xmm1
+; X64_WIDEN-NEXT:    psubd %xmm1, %xmm2
+; X64_WIDEN-NEXT:    paddd %xmm0, %xmm2
+; X64_WIDEN-NEXT:    movq %xmm2, (%rsi)
+; X64_WIDEN-NEXT:    retq
+;
+; X86_WIDEN-LABEL: test_srem7_v2i32:
+; X86_WIDEN:       # %bb.0:
+; X86_WIDEN-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86_WIDEN-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86_WIDEN-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; X86_WIDEN-NEXT:    movdqa {{.*#+}} xmm1 = [2454267027,2454267027,2454267027,2454267027]
+; X86_WIDEN-NEXT:    movdqa %xmm0, %xmm2
+; X86_WIDEN-NEXT:    pmuludq %xmm1, %xmm2
+; X86_WIDEN-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
+; X86_WIDEN-NEXT:    movdqa %xmm0, %xmm3
+; X86_WIDEN-NEXT:    shufps {{.*#+}} xmm3 = xmm3[1,1],xmm0[2,3]
+; X86_WIDEN-NEXT:    pmuludq %xmm1, %xmm3
+; X86_WIDEN-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,3,2,3]
+; X86_WIDEN-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
+; X86_WIDEN-NEXT:    pxor %xmm3, %xmm3
+; X86_WIDEN-NEXT:    pcmpgtd %xmm0, %xmm3
+; X86_WIDEN-NEXT:    pand %xmm1, %xmm3
+; X86_WIDEN-NEXT:    paddd %xmm0, %xmm3
+; X86_WIDEN-NEXT:    psubd %xmm3, %xmm2
+; X86_WIDEN-NEXT:    paddd %xmm0, %xmm2
+; X86_WIDEN-NEXT:    movdqa %xmm2, %xmm1
+; X86_WIDEN-NEXT:    psrld $31, %xmm1
+; X86_WIDEN-NEXT:    psrad $2, %xmm2
+; X86_WIDEN-NEXT:    paddd %xmm1, %xmm2
+; X86_WIDEN-NEXT:    movdqa %xmm2, %xmm1
+; X86_WIDEN-NEXT:    pslld $3, %xmm1
+; X86_WIDEN-NEXT:    psubd %xmm1, %xmm2
+; X86_WIDEN-NEXT:    paddd %xmm0, %xmm2
+; X86_WIDEN-NEXT:    movq %xmm2, (%eax)
+; X86_WIDEN-NEXT:    retl
   %a = load <2 x i32>, <2 x i32>* %x
   %b = srem <2 x i32> %a, <i32 7, i32 7>
   store <2 x i32> %b, <2 x i32>* %y
@@ -240,6 +440,22 @@ define void @test_udiv_pow2_v2i32(<2 x i32>* %x, <2 x i32>* %y) nounwind {
 ; X86-NEXT:    psrld $3, %xmm0
 ; X86-NEXT:    movq %xmm0, (%eax)
 ; X86-NEXT:    retl
+;
+; X64_WIDEN-LABEL: test_udiv_pow2_v2i32:
+; X64_WIDEN:       # %bb.0:
+; X64_WIDEN-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; X64_WIDEN-NEXT:    psrld $3, %xmm0
+; X64_WIDEN-NEXT:    movq %xmm0, (%rsi)
+; X64_WIDEN-NEXT:    retq
+;
+; X86_WIDEN-LABEL: test_udiv_pow2_v2i32:
+; X86_WIDEN:       # %bb.0:
+; X86_WIDEN-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86_WIDEN-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86_WIDEN-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; X86_WIDEN-NEXT:    psrld $3, %xmm0
+; X86_WIDEN-NEXT:    movq %xmm0, (%eax)
+; X86_WIDEN-NEXT:    retl
   %a = load <2 x i32>, <2 x i32>* %x
   %b = udiv <2 x i32> %a, <i32 8, i32 8>
   store <2 x i32> %b, <2 x i32>* %y