-; RUN: llc < %s -march=x86 -mattr=+sse2 | FileCheck %s
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i686-unknown -mattr=+sse2 | FileCheck %s --check-prefix=CHECK --check-prefix=X32
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+sse2 | FileCheck %s --check-prefix=CHECK --check-prefix=X64
; test vector shifts converted to proper SSE2 vector shifts when the shift
; amounts are the same.
define void @shift1a(<2 x i64> %val, <2 x i64>* %dst) nounwind {
+; X32-LABEL: shift1a:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: psllq $32, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift1a:
+; X64: # BB#0: # %entry
+; X64-NEXT: psllq $32, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift1a:
-; CHECK: psllq
%shl = shl <2 x i64> %val, < i64 32, i64 32 >
store <2 x i64> %shl, <2 x i64>* %dst
ret void
}
define void @shift1b(<2 x i64> %val, <2 x i64>* %dst, i64 %amt) nounwind {
+; X32-LABEL: shift1b:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; X32-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,0,1,1]
+; X32-NEXT: movd {{.*#+}} xmm2 = mem[0],zero,zero,zero
+; X32-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,0,1,1]
+; X32-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; X32-NEXT: psllq %xmm2, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift1b:
+; X64: # BB#0: # %entry
+; X64-NEXT: movd %rsi, %xmm1
+; X64-NEXT: psllq %xmm1, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift1b:
-; CHECK: movd
-; CHECK: psllq
%0 = insertelement <2 x i64> undef, i64 %amt, i32 0
%1 = insertelement <2 x i64> %0, i64 %amt, i32 1
%shl = shl <2 x i64> %val, %1
define void @shift2a(<4 x i32> %val, <4 x i32>* %dst) nounwind {
+; X32-LABEL: shift2a:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: pslld $5, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift2a:
+; X64: # BB#0: # %entry
+; X64-NEXT: pslld $5, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift2a:
-; CHECK: pslld
%shl = shl <4 x i32> %val, < i32 5, i32 5, i32 5, i32 5 >
store <4 x i32> %shl, <4 x i32>* %dst
ret void
}
define void @shift2b(<4 x i32> %val, <4 x i32>* %dst, i32 %amt) nounwind {
+; X32-LABEL: shift2b:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; X32-NEXT: pslld %xmm1, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift2b:
+; X64: # BB#0: # %entry
+; X64-NEXT: movd %esi, %xmm1
+; X64-NEXT: pslld %xmm1, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift2b:
-; CHECK: movd
-; CHECK: pslld
%0 = insertelement <4 x i32> undef, i32 %amt, i32 0
%1 = insertelement <4 x i32> %0, i32 %amt, i32 1
%2 = insertelement <4 x i32> %1, i32 %amt, i32 2
}
define void @shift3a(<8 x i16> %val, <8 x i16>* %dst) nounwind {
+; X32-LABEL: shift3a:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: psllw $5, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift3a:
+; X64: # BB#0: # %entry
+; X64-NEXT: psllw $5, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift3a:
-; CHECK: psllw
%shl = shl <8 x i16> %val, < i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5 >
store <8 x i16> %shl, <8 x i16>* %dst
ret void
; Make sure the shift amount is properly zero extended.
define void @shift3b(<8 x i16> %val, <8 x i16>* %dst, i16 %amt) nounwind {
+; X32-LABEL: shift3b:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT: movd %ecx, %xmm1
+; X32-NEXT: psllw %xmm1, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift3b:
+; X64: # BB#0: # %entry
+; X64-NEXT: movzwl %si, %eax
+; X64-NEXT: movd %eax, %xmm1
+; X64-NEXT: psllw %xmm1, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift3b:
-; CHECK: movzwl
-; CHECK: movd
-; CHECK-NEXT: psllw
%0 = insertelement <8 x i16> undef, i16 %amt, i32 0
%1 = insertelement <8 x i16> %0, i16 %amt, i32 1
%2 = insertelement <8 x i16> %1, i16 %amt, i32 2
-; RUN: llc < %s -march=x86 -mattr=+sse2 | FileCheck %s
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i686-unknown -mattr=+sse2 | FileCheck %s --check-prefix=CHECK --check-prefix=X32
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+sse2 | FileCheck %s --check-prefix=CHECK --check-prefix=X64
; test vector shifts converted to proper SSE2 vector shifts when the shift
; amounts are the same.
define void @shift1a(<2 x i64> %val, <2 x i64>* %dst) nounwind {
+; X32-LABEL: shift1a:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: psrlq $32, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift1a:
+; X64: # BB#0: # %entry
+; X64-NEXT: psrlq $32, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift1a:
-; CHECK: psrlq
%lshr = lshr <2 x i64> %val, < i64 32, i64 32 >
store <2 x i64> %lshr, <2 x i64>* %dst
ret void
}
define void @shift1b(<2 x i64> %val, <2 x i64>* %dst, i64 %amt) nounwind {
+; X32-LABEL: shift1b:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; X32-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,0,1,1]
+; X32-NEXT: movd {{.*#+}} xmm2 = mem[0],zero,zero,zero
+; X32-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,0,1,1]
+; X32-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; X32-NEXT: psrlq %xmm2, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift1b:
+; X64: # BB#0: # %entry
+; X64-NEXT: movd %rsi, %xmm1
+; X64-NEXT: psrlq %xmm1, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift1b:
-; CHECK: movd
-; CHECK: psrlq
%0 = insertelement <2 x i64> undef, i64 %amt, i32 0
%1 = insertelement <2 x i64> %0, i64 %amt, i32 1
%lshr = lshr <2 x i64> %val, %1
}
define void @shift2a(<4 x i32> %val, <4 x i32>* %dst) nounwind {
+; X32-LABEL: shift2a:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: psrld $17, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift2a:
+; X64: # BB#0: # %entry
+; X64-NEXT: psrld $17, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift2a:
-; CHECK: psrld
%lshr = lshr <4 x i32> %val, < i32 17, i32 17, i32 17, i32 17 >
store <4 x i32> %lshr, <4 x i32>* %dst
ret void
}
define void @shift2b(<4 x i32> %val, <4 x i32>* %dst, i32 %amt) nounwind {
+; X32-LABEL: shift2b:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; X32-NEXT: psrld %xmm1, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift2b:
+; X64: # BB#0: # %entry
+; X64-NEXT: movd %esi, %xmm1
+; X64-NEXT: psrld %xmm1, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift2b:
-; CHECK: movd
-; CHECK: psrld
%0 = insertelement <4 x i32> undef, i32 %amt, i32 0
%1 = insertelement <4 x i32> %0, i32 %amt, i32 1
%2 = insertelement <4 x i32> %1, i32 %amt, i32 2
define void @shift3a(<8 x i16> %val, <8 x i16>* %dst) nounwind {
+; X32-LABEL: shift3a:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: psrlw $5, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift3a:
+; X64: # BB#0: # %entry
+; X64-NEXT: psrlw $5, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift3a:
-; CHECK: psrlw
%lshr = lshr <8 x i16> %val, < i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5 >
store <8 x i16> %lshr, <8 x i16>* %dst
ret void
; properly zero extend the shift amount
define void @shift3b(<8 x i16> %val, <8 x i16>* %dst, i16 %amt) nounwind {
+; X32-LABEL: shift3b:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT: movd %ecx, %xmm1
+; X32-NEXT: psrlw %xmm1, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift3b:
+; X64: # BB#0: # %entry
+; X64-NEXT: movzwl %si, %eax
+; X64-NEXT: movd %eax, %xmm1
+; X64-NEXT: psrlw %xmm1, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift3b:
-; CHECK: movzwl
-; CHECK: movd
-; CHECK: psrlw
%0 = insertelement <8 x i16> undef, i16 %amt, i32 0
%1 = insertelement <8 x i16> %0, i16 %amt, i32 1
%2 = insertelement <8 x i16> %1, i16 %amt, i32 2
-; RUN: llc < %s -march=x86 -mattr=+sse2 | FileCheck %s
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i686-unknown -mattr=+sse2 | FileCheck %s --check-prefix=CHECK --check-prefix=X32
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+sse2 | FileCheck %s --check-prefix=CHECK --check-prefix=X64
; test vector shifts converted to proper SSE2 vector shifts when the shift
; amounts are the same.
; Note that x86 does have ashr
define void @shift1a(<2 x i64> %val, <2 x i64>* %dst) nounwind {
+; X32-LABEL: shift1a:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,3,2,3]
+; X32-NEXT: psrad $31, %xmm0
+; X32-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,3,2,3]
+; X32-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; X32-NEXT: movdqa %xmm1, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift1a:
+; X64: # BB#0: # %entry
+; X64-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,3,2,3]
+; X64-NEXT: psrad $31, %xmm0
+; X64-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,3,2,3]
+; X64-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; X64-NEXT: movdqa %xmm1, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift1a:
-; CHECK: psrad $31
%ashr = ashr <2 x i64> %val, < i64 32, i64 32 >
store <2 x i64> %ashr, <2 x i64>* %dst
ret void
}
define void @shift2a(<4 x i32> %val, <4 x i32>* %dst) nounwind {
+; X32-LABEL: shift2a:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: psrad $5, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift2a:
+; X64: # BB#0: # %entry
+; X64-NEXT: psrad $5, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift2a:
-; CHECK: psrad $5
%ashr = ashr <4 x i32> %val, < i32 5, i32 5, i32 5, i32 5 >
store <4 x i32> %ashr, <4 x i32>* %dst
ret void
}
define void @shift2b(<4 x i32> %val, <4 x i32>* %dst, i32 %amt) nounwind {
+; X32-LABEL: shift2b:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; X32-NEXT: psrad %xmm1, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift2b:
+; X64: # BB#0: # %entry
+; X64-NEXT: movd %esi, %xmm1
+; X64-NEXT: psrad %xmm1, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift2b:
-; CHECK: movd
-; CHECK: psrad
%0 = insertelement <4 x i32> undef, i32 %amt, i32 0
%1 = insertelement <4 x i32> %0, i32 %amt, i32 1
%2 = insertelement <4 x i32> %1, i32 %amt, i32 2
}
define void @shift3a(<8 x i16> %val, <8 x i16>* %dst) nounwind {
+; X32-LABEL: shift3a:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: psraw $5, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift3a:
+; X64: # BB#0: # %entry
+; X64-NEXT: psraw $5, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift3a:
-; CHECK: psraw $5
%ashr = ashr <8 x i16> %val, < i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5 >
store <8 x i16> %ashr, <8 x i16>* %dst
ret void
}
define void @shift3b(<8 x i16> %val, <8 x i16>* %dst, i16 %amt) nounwind {
+; X32-LABEL: shift3b:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT: movd %ecx, %xmm1
+; X32-NEXT: psraw %xmm1, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift3b:
+; X64: # BB#0: # %entry
+; X64-NEXT: movzwl %si, %eax
+; X64-NEXT: movd %eax, %xmm1
+; X64-NEXT: psraw %xmm1, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift3b:
-; CHECK: movzwl
-; CHECK: movd
-; CHECK: psraw
%0 = insertelement <8 x i16> undef, i16 %amt, i32 0
%1 = insertelement <8 x i16> %0, i16 %amt, i32 1
%2 = insertelement <8 x i16> %1, i16 %amt, i32 2
-; RUN: llc < %s -march=x86 -mcpu=core2 | FileCheck %s
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i686-unknown -mattr=+sse2 | FileCheck %s --check-prefix=CHECK --check-prefix=X32
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+sse2 | FileCheck %s --check-prefix=CHECK --check-prefix=X64
; test vector shifts converted to proper SSE2 vector shifts when the shift
; amounts are the same when using a shuffle splat.
define void @shift1a(<2 x i64> %val, <2 x i64>* %dst, <2 x i64> %sh) nounwind {
+; X32-LABEL: shift1a:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: movq {{.*#+}} xmm1 = xmm1[0],zero
+; X32-NEXT: psllq %xmm1, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift1a:
+; X64: # BB#0: # %entry
+; X64-NEXT: psllq %xmm1, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift1a:
-; CHECK: psllq
%shamt = shufflevector <2 x i64> %sh, <2 x i64> undef, <2 x i32> <i32 0, i32 0>
%shl = shl <2 x i64> %val, %shamt
store <2 x i64> %shl, <2 x i64>* %dst
; shift1b can't use a packed shift but can shift lanes separately and shuffle back together
define void @shift1b(<2 x i64> %val, <2 x i64>* %dst, <2 x i64> %sh) nounwind {
+; X32-LABEL: shift1b:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: pshufd {{.*#+}} xmm2 = xmm1[2,3,0,1]
+; X32-NEXT: movdqa %xmm0, %xmm3
+; X32-NEXT: psllq %xmm2, %xmm3
+; X32-NEXT: movq {{.*#+}} xmm1 = xmm1[0],zero
+; X32-NEXT: psllq %xmm1, %xmm0
+; X32-NEXT: movsd {{.*#+}} xmm3 = xmm0[0],xmm3[1]
+; X32-NEXT: movapd %xmm3, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift1b:
+; X64: # BB#0: # %entry
+; X64-NEXT: pshufd {{.*#+}} xmm2 = xmm1[2,3,0,1]
+; X64-NEXT: movdqa %xmm0, %xmm3
+; X64-NEXT: psllq %xmm2, %xmm3
+; X64-NEXT: psllq %xmm1, %xmm0
+; X64-NEXT: movsd {{.*#+}} xmm3 = xmm0[0],xmm3[1]
+; X64-NEXT: movapd %xmm3, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift1b:
-; CHECK: pshufd {{.*#+}} xmm2 = xmm1[2,3,0,1]
-; CHECK-NEXT: movdqa %xmm0, %xmm3
-; CHECK-NEXT: psllq %xmm2, %xmm3
-; CHECK-NEXT: movq {{.*#+}} xmm1 = xmm1[0],zero
-; CHECK-NEXT: psllq %xmm1, %xmm0
-; CHECK-NEXT: movsd {{.*#+}} xmm3 = xmm0[0],xmm3[1]
%shamt = shufflevector <2 x i64> %sh, <2 x i64> undef, <2 x i32> <i32 0, i32 1>
%shl = shl <2 x i64> %val, %shamt
store <2 x i64> %shl, <2 x i64>* %dst
}
define void @shift2a(<4 x i32> %val, <4 x i32>* %dst, <2 x i32> %amt) nounwind {
+; X32-LABEL: shift2a:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
+; X32-NEXT: xorps %xmm2, %xmm2
+; X32-NEXT: movss {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
+; X32-NEXT: pslld %xmm2, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift2a:
+; X64: # BB#0: # %entry
+; X64-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
+; X64-NEXT: xorps %xmm2, %xmm2
+; X64-NEXT: movss {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
+; X64-NEXT: pslld %xmm2, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift2a:
-; CHECK: pslld
%shamt = shufflevector <2 x i32> %amt, <2 x i32> undef, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
%shl = shl <4 x i32> %val, %shamt
store <4 x i32> %shl, <4 x i32>* %dst
}
define void @shift2b(<4 x i32> %val, <4 x i32>* %dst, <2 x i32> %amt) nounwind {
+; X32-LABEL: shift2b:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
+; X32-NEXT: xorps %xmm2, %xmm2
+; X32-NEXT: movss {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
+; X32-NEXT: pslld %xmm2, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift2b:
+; X64: # BB#0: # %entry
+; X64-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
+; X64-NEXT: xorps %xmm2, %xmm2
+; X64-NEXT: movss {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
+; X64-NEXT: pslld %xmm2, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift2b:
-; CHECK: pslld
%shamt = shufflevector <2 x i32> %amt, <2 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 1, i32 1>
%shl = shl <4 x i32> %val, %shamt
store <4 x i32> %shl, <4 x i32>* %dst
}
define void @shift2c(<4 x i32> %val, <4 x i32>* %dst, <2 x i32> %amt) nounwind {
+; X32-LABEL: shift2c:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
+; X32-NEXT: xorps %xmm2, %xmm2
+; X32-NEXT: movss {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
+; X32-NEXT: pslld %xmm2, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift2c:
+; X64: # BB#0: # %entry
+; X64-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
+; X64-NEXT: xorps %xmm2, %xmm2
+; X64-NEXT: movss {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
+; X64-NEXT: pslld %xmm2, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift2c:
-; CHECK: pslld
%shamt = shufflevector <2 x i32> %amt, <2 x i32> undef, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
%shl = shl <4 x i32> %val, %shamt
store <4 x i32> %shl, <4 x i32>* %dst
}
define void @shift3a(<8 x i16> %val, <8 x i16>* %dst, <8 x i16> %amt) nounwind {
+; X32-LABEL: shift3a:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: pextrw $6, %xmm1, %ecx
+; X32-NEXT: movd %ecx, %xmm1
+; X32-NEXT: psllw %xmm1, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift3a:
+; X64: # BB#0: # %entry
+; X64-NEXT: pextrw $6, %xmm1, %eax
+; X64-NEXT: movd %eax, %xmm1
+; X64-NEXT: psllw %xmm1, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift3a:
-; CHECK: pextrw $6
-; CHECK: psllw
%shamt = shufflevector <8 x i16> %amt, <8 x i16> undef, <8 x i32> <i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6>
%shl = shl <8 x i16> %val, %shamt
store <8 x i16> %shl, <8 x i16>* %dst
}
define void @shift3b(<8 x i16> %val, <8 x i16>* %dst, i16 %amt) nounwind {
+; X32-LABEL: shift3b:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT: movd %ecx, %xmm1
+; X32-NEXT: psllw %xmm1, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift3b:
+; X64: # BB#0: # %entry
+; X64-NEXT: movzwl %si, %eax
+; X64-NEXT: movd %eax, %xmm1
+; X64-NEXT: psllw %xmm1, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift3b:
-; CHECK: movzwl
-; CHECK: psllw
%0 = insertelement <8 x i16> undef, i16 %amt, i32 0
%1 = insertelement <8 x i16> %0, i16 %amt, i32 1
%2 = insertelement <8 x i16> %1, i16 %amt, i32 2
-; RUN: llc < %s -march=x86 -mattr=+sse2 | FileCheck %s
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i686-unknown -mattr=+sse2 | FileCheck %s --check-prefix=CHECK --check-prefix=X32
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+sse2 | FileCheck %s --check-prefix=CHECK --check-prefix=X64
; When loading the shift amount from memory, avoid generating the splat.
define void @shift5a(<4 x i32> %val, <4 x i32>* %dst, i32* %pamt) nounwind {
+; X32-LABEL: shift5a:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT: movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; X32-NEXT: pslld %xmm1, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift5a:
+; X64: # BB#0: # %entry
+; X64-NEXT: movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; X64-NEXT: pslld %xmm1, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift5a:
-; CHECK: movd
-; CHECK: pslld
- %amt = load i32, i32* %pamt
+ %amt = load i32, i32* %pamt
%tmp0 = insertelement <4 x i32> undef, i32 %amt, i32 0
- %shamt = shufflevector <4 x i32> %tmp0, <4 x i32> undef, <4 x i32> zeroinitializer
+ %shamt = shufflevector <4 x i32> %tmp0, <4 x i32> undef, <4 x i32> zeroinitializer
%shl = shl <4 x i32> %val, %shamt
store <4 x i32> %shl, <4 x i32>* %dst
ret void
define void @shift5b(<4 x i32> %val, <4 x i32>* %dst, i32* %pamt) nounwind {
+; X32-LABEL: shift5b:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT: movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; X32-NEXT: psrad %xmm1, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift5b:
+; X64: # BB#0: # %entry
+; X64-NEXT: movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; X64-NEXT: psrad %xmm1, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift5b:
-; CHECK: movd
-; CHECK: psrad
- %amt = load i32, i32* %pamt
+ %amt = load i32, i32* %pamt
%tmp0 = insertelement <4 x i32> undef, i32 %amt, i32 0
- %shamt = shufflevector <4 x i32> %tmp0, <4 x i32> undef, <4 x i32> zeroinitializer
+ %shamt = shufflevector <4 x i32> %tmp0, <4 x i32> undef, <4 x i32> zeroinitializer
%shr = ashr <4 x i32> %val, %shamt
store <4 x i32> %shr, <4 x i32>* %dst
ret void
define void @shift5c(<4 x i32> %val, <4 x i32>* %dst, i32 %amt) nounwind {
+; X32-LABEL: shift5c:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; X32-NEXT: pslld %xmm1, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift5c:
+; X64: # BB#0: # %entry
+; X64-NEXT: movd %esi, %xmm1
+; X64-NEXT: pslld %xmm1, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift5c:
-; CHECK: movd
-; CHECK: pslld
%tmp0 = insertelement <4 x i32> undef, i32 %amt, i32 0
%shamt = shufflevector <4 x i32> %tmp0, <4 x i32> undef, <4 x i32> zeroinitializer
%shl = shl <4 x i32> %val, %shamt
define void @shift5d(<4 x i32> %val, <4 x i32>* %dst, i32 %amt) nounwind {
+; X32-LABEL: shift5d:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; X32-NEXT: psrad %xmm1, %xmm0
+; X32-NEXT: movdqa %xmm0, (%eax)
+; X32-NEXT: retl
+;
+; X64-LABEL: shift5d:
+; X64: # BB#0: # %entry
+; X64-NEXT: movd %esi, %xmm1
+; X64-NEXT: psrad %xmm1, %xmm0
+; X64-NEXT: movdqa %xmm0, (%rdi)
+; X64-NEXT: retq
entry:
-; CHECK-LABEL: shift5d:
-; CHECK: movd
-; CHECK: psrad
%tmp0 = insertelement <4 x i32> undef, i32 %amt, i32 0
%shamt = shufflevector <4 x i32> %tmp0, <4 x i32> undef, <4 x i32> zeroinitializer
%shr = ashr <4 x i32> %val, %shamt
-; RUN: llc < %s -march=x86-64 -mattr=+sse2 | FileCheck %s
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i686-unknown -mattr=+sse2 | FileCheck %s --check-prefix=CHECK --check-prefix=X32
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+sse2 | FileCheck %s --check-prefix=CHECK --check-prefix=X64
; This test makes sure that the compiler does not crash with an
; assertion failure when trying to fold a vector shift left
; 'count' is the vector shift count.
define <16 x i8> @do_not_crash(i8*, i32*, i64*, i32, i64, i8) {
+; X32-LABEL: do_not_crash:
+; X32: # BB#0: # %entry
+; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT: movb %al, (%ecx)
+; X32-NEXT: movd %eax, %xmm0
+; X32-NEXT: psllq $56, %xmm0
+; X32-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,255]
+; X32-NEXT: movdqa %xmm2, %xmm1
+; X32-NEXT: pandn %xmm0, %xmm1
+; X32-NEXT: por %xmm2, %xmm1
+; X32-NEXT: pcmpeqd %xmm2, %xmm2
+; X32-NEXT: psllw $5, %xmm1
+; X32-NEXT: pxor %xmm3, %xmm3
+; X32-NEXT: pxor %xmm0, %xmm0
+; X32-NEXT: pcmpgtb %xmm1, %xmm0
+; X32-NEXT: pxor %xmm0, %xmm2
+; X32-NEXT: pand {{\.LCPI.*}}, %xmm0
+; X32-NEXT: por %xmm2, %xmm0
+; X32-NEXT: paddb %xmm1, %xmm1
+; X32-NEXT: pxor %xmm2, %xmm2
+; X32-NEXT: pcmpgtb %xmm1, %xmm2
+; X32-NEXT: movdqa %xmm2, %xmm4
+; X32-NEXT: pandn %xmm0, %xmm4
+; X32-NEXT: psllw $2, %xmm0
+; X32-NEXT: pand {{\.LCPI.*}}, %xmm0
+; X32-NEXT: pand %xmm2, %xmm0
+; X32-NEXT: por %xmm4, %xmm0
+; X32-NEXT: paddb %xmm1, %xmm1
+; X32-NEXT: pcmpgtb %xmm1, %xmm3
+; X32-NEXT: movdqa %xmm3, %xmm1
+; X32-NEXT: pandn %xmm0, %xmm1
+; X32-NEXT: paddb %xmm0, %xmm0
+; X32-NEXT: pand %xmm3, %xmm0
+; X32-NEXT: por %xmm1, %xmm0
+; X32-NEXT: retl
+;
+; X64-LABEL: do_not_crash:
+; X64: # BB#0: # %entry
+; X64-NEXT: movb %r9b, (%rdi)
+; X64-NEXT: movd %r9d, %xmm0
+; X64-NEXT: psllq $56, %xmm0
+; X64-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,255]
+; X64-NEXT: movdqa %xmm2, %xmm1
+; X64-NEXT: pandn %xmm0, %xmm1
+; X64-NEXT: por %xmm2, %xmm1
+; X64-NEXT: pcmpeqd %xmm2, %xmm2
+; X64-NEXT: psllw $5, %xmm1
+; X64-NEXT: pxor %xmm3, %xmm3
+; X64-NEXT: pxor %xmm0, %xmm0
+; X64-NEXT: pcmpgtb %xmm1, %xmm0
+; X64-NEXT: pxor %xmm0, %xmm2
+; X64-NEXT: pand {{.*}}(%rip), %xmm0
+; X64-NEXT: por %xmm2, %xmm0
+; X64-NEXT: paddb %xmm1, %xmm1
+; X64-NEXT: pxor %xmm2, %xmm2
+; X64-NEXT: pcmpgtb %xmm1, %xmm2
+; X64-NEXT: movdqa %xmm2, %xmm4
+; X64-NEXT: pandn %xmm0, %xmm4
+; X64-NEXT: psllw $2, %xmm0
+; X64-NEXT: pand {{.*}}(%rip), %xmm0
+; X64-NEXT: pand %xmm2, %xmm0
+; X64-NEXT: por %xmm4, %xmm0
+; X64-NEXT: paddb %xmm1, %xmm1
+; X64-NEXT: pcmpgtb %xmm1, %xmm3
+; X64-NEXT: movdqa %xmm3, %xmm1
+; X64-NEXT: pandn %xmm0, %xmm1
+; X64-NEXT: paddb %xmm0, %xmm0
+; X64-NEXT: pand %xmm3, %xmm0
+; X64-NEXT: por %xmm1, %xmm0
+; X64-NEXT: retq
entry:
store i8 %5, i8* %0
%L5 = load i8, i8* %0
%B51 = shl <16 x i8> <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>, %I8
ret <16 x i8> %B51
}
-
-; CHECK-LABEL: do_not_crash
-; CHECK: ret
-