834cc88c5d
lowerBuildVectorAsBroadcast will not broadcast splat constants in all cases, resulting in a lot of situations where a full width vector load that has failed to fold but is loading splat constant values could use a broadcast load instruction just as cheaply, and save constant pool space. NOTE: SSE3 targets can use MOVDDUP but not all SSE era CPUs can perform this as cheaply as a vector load, we will need to add scheduler model checks if we want to pursue this. This is an updated commit of98061013e0after being reverted ata279a09ab9
19 lines
1.0 KiB
LLVM
19 lines
1.0 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc < %s -mtriple=x86_64-unknown -mcpu=btver2 | FileCheck %s
|
|
|
|
define <8 x double> @test(<4 x double> %a, <4 x double> %b) {
|
|
; CHECK-LABEL: test:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vbroadcastsd {{.*#+}} ymm1 = [8.2071743224100002E-1,8.2071743224100002E-1,8.2071743224100002E-1,8.2071743224100002E-1]
|
|
; CHECK-NEXT: vblendps {{.*#+}} ymm2 = ymm0[0,1,2,3],ymm1[4,5,6,7]
|
|
; CHECK-NEXT: vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm0[4,5,6,7]
|
|
; CHECK-NEXT: vunpckhpd {{.*#+}} ymm1 = ymm1[1],ymm2[1],ymm1[3],ymm2[3]
|
|
; CHECK-NEXT: vmovddup {{.*#+}} xmm2 = [8.2071743224100002E-1,8.2071743224100002E-1]
|
|
; CHECK-NEXT: # xmm2 = mem[0,0]
|
|
; CHECK-NEXT: vblendps {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
|
|
; CHECK-NEXT: retq
|
|
%1 = shufflevector <4 x double> %a, <4 x double> <double undef, double 0x3FEA435134576E1C, double 0x3FEA435134576E1C, double 0x3FEA435134576E1C>, <8 x i32> <i32 6, i32 5, i32 2, i32 3, i32 5, i32 1, i32 3, i32 7>
|
|
ret <8 x double> %1
|
|
}
|
|
|