2016-07-31 04:28:02 +08:00
|
|
|
; 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
|
2009-01-28 16:13:56 +08:00
|
|
|
|
|
|
|
; 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 {
|
2016-07-31 04:28:02 +08:00
|
|
|
; X32-LABEL: shift1a:
|
2017-12-05 01:18:51 +08:00
|
|
|
; X32: # %bb.0: # %entry
|
2016-07-31 04:28:02 +08:00
|
|
|
; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
|
|
|
|
; X32-NEXT: psllq $32, %xmm0
|
|
|
|
; X32-NEXT: movdqa %xmm0, (%eax)
|
|
|
|
; X32-NEXT: retl
|
|
|
|
;
|
|
|
|
; X64-LABEL: shift1a:
|
2017-12-05 01:18:51 +08:00
|
|
|
; X64: # %bb.0: # %entry
|
2016-07-31 04:28:02 +08:00
|
|
|
; X64-NEXT: psllq $32, %xmm0
|
|
|
|
; X64-NEXT: movdqa %xmm0, (%rdi)
|
|
|
|
; X64-NEXT: retq
|
2009-01-28 16:13:56 +08:00
|
|
|
entry:
|
|
|
|
%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 {
|
2016-07-31 04:28:02 +08:00
|
|
|
; X32-LABEL: shift1b:
|
2017-12-05 01:18:51 +08:00
|
|
|
; X32: # %bb.0: # %entry
|
2016-07-31 04:28:02 +08:00
|
|
|
; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
|
[X86][SSE] Change BUILD_VECTOR interleaving ordering to improve coalescing/combine opportunities
We currently generate BUILD_VECTOR as a tree of UNPCKL shuffles of the same type:
e.g. for v4f32:
Step 1: unpcklps 0, 2 ==> X: <?, ?, 2, 0>
: unpcklps 1, 3 ==> Y: <?, ?, 3, 1>
Step 2: unpcklps X, Y ==> <3, 2, 1, 0>
The issue is because we are not placing sequential vector elements together early enough, we fail to recognise many combinable patterns - consecutive scalar loads, extractions etc.
Instead, this patch unpacks progressively larger sequential vector elements together:
e.g. for v4f32:
Step 1: unpcklps 0, 2 ==> X: <?, ?, 1, 0>
: unpcklps 1, 3 ==> Y: <?, ?, 3, 2>
Step 2: unpcklpd X, Y ==> <3, 2, 1, 0>
This does mean that we are creating UNPCKL shuffle of different value types, but the relevant combines that benefit from this are quite capable of handling the additional BITCASTs that are now included in the shuffle tree.
Differential Revision: https://reviews.llvm.org/D33864
llvm-svn: 304688
2017-06-05 04:12:04 +08:00
|
|
|
; X32-NEXT: movq {{.*#+}} xmm1 = mem[0],zero
|
|
|
|
; X32-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
|
|
|
|
; X32-NEXT: psllq %xmm1, %xmm0
|
2016-07-31 04:28:02 +08:00
|
|
|
; X32-NEXT: movdqa %xmm0, (%eax)
|
|
|
|
; X32-NEXT: retl
|
|
|
|
;
|
|
|
|
; X64-LABEL: shift1b:
|
2017-12-05 01:18:51 +08:00
|
|
|
; X64: # %bb.0: # %entry
|
2017-04-26 15:08:44 +08:00
|
|
|
; X64-NEXT: movq %rsi, %xmm1
|
2016-07-31 04:28:02 +08:00
|
|
|
; X64-NEXT: psllq %xmm1, %xmm0
|
|
|
|
; X64-NEXT: movdqa %xmm0, (%rdi)
|
|
|
|
; X64-NEXT: retq
|
2009-01-28 16:13:56 +08:00
|
|
|
entry:
|
|
|
|
%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
|
|
|
|
store <2 x i64> %shl, <2 x i64>* %dst
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
define void @shift2a(<4 x i32> %val, <4 x i32>* %dst) nounwind {
|
2016-07-31 04:28:02 +08:00
|
|
|
; X32-LABEL: shift2a:
|
2017-12-05 01:18:51 +08:00
|
|
|
; X32: # %bb.0: # %entry
|
2016-07-31 04:28:02 +08:00
|
|
|
; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
|
|
|
|
; X32-NEXT: pslld $5, %xmm0
|
|
|
|
; X32-NEXT: movdqa %xmm0, (%eax)
|
|
|
|
; X32-NEXT: retl
|
|
|
|
;
|
|
|
|
; X64-LABEL: shift2a:
|
2017-12-05 01:18:51 +08:00
|
|
|
; X64: # %bb.0: # %entry
|
2016-07-31 04:28:02 +08:00
|
|
|
; X64-NEXT: pslld $5, %xmm0
|
|
|
|
; X64-NEXT: movdqa %xmm0, (%rdi)
|
|
|
|
; X64-NEXT: retq
|
2009-01-28 16:13:56 +08:00
|
|
|
entry:
|
|
|
|
%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 {
|
2016-07-31 04:28:02 +08:00
|
|
|
; X32-LABEL: shift2b:
|
2017-12-05 01:18:51 +08:00
|
|
|
; X32: # %bb.0: # %entry
|
2016-07-31 04:28:02 +08:00
|
|
|
; 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:
|
2017-12-05 01:18:51 +08:00
|
|
|
; X64: # %bb.0: # %entry
|
2016-07-31 04:28:02 +08:00
|
|
|
; X64-NEXT: movd %esi, %xmm1
|
|
|
|
; X64-NEXT: pslld %xmm1, %xmm0
|
|
|
|
; X64-NEXT: movdqa %xmm0, (%rdi)
|
|
|
|
; X64-NEXT: retq
|
2009-01-28 16:13:56 +08:00
|
|
|
entry:
|
|
|
|
%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
|
|
|
|
%3 = insertelement <4 x i32> %2, i32 %amt, i32 3
|
|
|
|
%shl = shl <4 x i32> %val, %3
|
|
|
|
store <4 x i32> %shl, <4 x i32>* %dst
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
define void @shift3a(<8 x i16> %val, <8 x i16>* %dst) nounwind {
|
2016-07-31 04:28:02 +08:00
|
|
|
; X32-LABEL: shift3a:
|
2017-12-05 01:18:51 +08:00
|
|
|
; X32: # %bb.0: # %entry
|
2016-07-31 04:28:02 +08:00
|
|
|
; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
|
|
|
|
; X32-NEXT: psllw $5, %xmm0
|
|
|
|
; X32-NEXT: movdqa %xmm0, (%eax)
|
|
|
|
; X32-NEXT: retl
|
|
|
|
;
|
|
|
|
; X64-LABEL: shift3a:
|
2017-12-05 01:18:51 +08:00
|
|
|
; X64: # %bb.0: # %entry
|
2016-07-31 04:28:02 +08:00
|
|
|
; X64-NEXT: psllw $5, %xmm0
|
|
|
|
; X64-NEXT: movdqa %xmm0, (%rdi)
|
|
|
|
; X64-NEXT: retq
|
2009-01-28 16:13:56 +08:00
|
|
|
entry:
|
|
|
|
%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
|
|
|
|
}
|
|
|
|
|
2009-09-04 03:57:35 +08:00
|
|
|
; Make sure the shift amount is properly zero extended.
|
2009-01-28 16:13:56 +08:00
|
|
|
define void @shift3b(<8 x i16> %val, <8 x i16>* %dst, i16 %amt) nounwind {
|
2016-07-31 04:28:02 +08:00
|
|
|
; X32-LABEL: shift3b:
|
2017-12-05 01:18:51 +08:00
|
|
|
; X32: # %bb.0: # %entry
|
2016-07-31 04:28:02 +08:00
|
|
|
; 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:
|
2017-12-05 01:18:51 +08:00
|
|
|
; X64: # %bb.0: # %entry
|
2016-07-31 04:28:02 +08:00
|
|
|
; X64-NEXT: movzwl %si, %eax
|
|
|
|
; X64-NEXT: movd %eax, %xmm1
|
|
|
|
; X64-NEXT: psllw %xmm1, %xmm0
|
|
|
|
; X64-NEXT: movdqa %xmm0, (%rdi)
|
|
|
|
; X64-NEXT: retq
|
2009-01-28 16:13:56 +08:00
|
|
|
entry:
|
|
|
|
%0 = insertelement <8 x i16> undef, i16 %amt, i32 0
|
|
|
|
%1 = insertelement <8 x i16> %0, i16 %amt, i32 1
|
2013-07-30 08:24:09 +08:00
|
|
|
%2 = insertelement <8 x i16> %1, i16 %amt, i32 2
|
|
|
|
%3 = insertelement <8 x i16> %2, i16 %amt, i32 3
|
|
|
|
%4 = insertelement <8 x i16> %3, i16 %amt, i32 4
|
|
|
|
%5 = insertelement <8 x i16> %4, i16 %amt, i32 5
|
|
|
|
%6 = insertelement <8 x i16> %5, i16 %amt, i32 6
|
|
|
|
%7 = insertelement <8 x i16> %6, i16 %amt, i32 7
|
2009-01-28 16:13:56 +08:00
|
|
|
%shl = shl <8 x i16> %val, %7
|
|
|
|
store <8 x i16> %shl, <8 x i16>* %dst
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|