llvm-6502/test/CodeGen/X86/combine-sse2-intrinsics.ll
Andrea Di Biagio 2360e51fd0 [X86] Add target specific combine rules to fold SSE2/AVX2 packed arithmetic shift intrinsics.
This patch teaches the backend how to combine packed SSE2/AVX2 arithmetic shift
intrinsics.

The rules are:
 - Always fold a packed arithmetic shift by zero to its first operand;
 - Convert a packed arithmetic shift intrinsic dag node into a ISD::SRA only if
   the shift count is known to be smaller than the vector element size.

This patch also teaches to function 'getTargetVShiftByConstNode' how fold
target specific vector shifts by zero.

Added two new tests to verify that the DAGCombiner is able to fold
sequences of SSE2/AVX2 packed arithmetic shift calls.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@208342 91177308-0d34-0410-b5e6-96231b3b80d8
2014-05-08 17:44:04 +00:00

54 lines
1.9 KiB
LLVM

; RUN: llc < %s -march=x86 -mcpu=core2 | FileCheck %s
; RUN: llc < %s -march=x86-64 -mcpu=corei7 | FileCheck %s
; Verify that the backend correctly combines SSE2 builtin intrinsics.
define <4 x i32> @test_psra_1(<4 x i32> %A) {
%1 = tail call <4 x i32> @llvm.x86.sse2.psrai.d(<4 x i32> %A, i32 3)
%2 = tail call <4 x i32> @llvm.x86.sse2.psra.d(<4 x i32> %1, <4 x i32> <i32 3, i32 0, i32 7, i32 0>)
%3 = tail call <4 x i32> @llvm.x86.sse2.psrai.d(<4 x i32> %2, i32 2)
ret <4 x i32> %3
}
; CHECK-LABEL: test_psra_1
; CHECK: psrad $8, %xmm0
; CHECK-NEXT: ret
define <8 x i16> @test_psra_2(<8 x i16> %A) {
%1 = tail call <8 x i16> @llvm.x86.sse2.psrai.w(<8 x i16> %A, i32 3)
%2 = tail call <8 x i16> @llvm.x86.sse2.psra.w(<8 x i16> %1, <8 x i16> <i16 3, i16 0, i16 0, i16 0, i16 7, i16 0, i16 0, i16 0>)
%3 = tail call <8 x i16> @llvm.x86.sse2.psrai.w(<8 x i16> %2, i32 2)
ret <8 x i16> %3
}
; CHECK-LABEL: test_psra_2
; CHECK: psraw $8, %xmm0
; CHECK-NEXT: ret
define <4 x i32> @test_psra_3(<4 x i32> %A) {
%1 = tail call <4 x i32> @llvm.x86.sse2.psrai.d(<4 x i32> %A, i32 0)
%2 = tail call <4 x i32> @llvm.x86.sse2.psra.d(<4 x i32> %1, <4 x i32> <i32 0, i32 0, i32 7, i32 0>)
%3 = tail call <4 x i32> @llvm.x86.sse2.psrai.d(<4 x i32> %2, i32 0)
ret <4 x i32> %3
}
; CHECK-LABEL: test_psra_3
; CHECK-NOT: psrad
; CHECK: ret
define <8 x i16> @test_psra_4(<8 x i16> %A) {
%1 = tail call <8 x i16> @llvm.x86.sse2.psrai.w(<8 x i16> %A, i32 0)
%2 = tail call <8 x i16> @llvm.x86.sse2.psra.w(<8 x i16> %1, <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 7, i16 0, i16 0, i16 0>)
%3 = tail call <8 x i16> @llvm.x86.sse2.psrai.w(<8 x i16> %2, i32 0)
ret <8 x i16> %3
}
; CHECK-LABEL: test_psra_4
; CHECK-NOT: psraw
; CHECK: ret
declare <8 x i16> @llvm.x86.sse2.psra.w(<8 x i16>, <8 x i16>)
declare <8 x i16> @llvm.x86.sse2.psrai.w(<8 x i16>, i32)
declare <4 x i32> @llvm.x86.sse2.psra.d(<4 x i32>, <4 x i32>)
declare <4 x i32> @llvm.x86.sse2.psrai.d(<4 x i32>, i32)