1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-23 11:13:28 +01:00
llvm-mirror/test/CodeGen/X86/ret-mmx.ll
Sanjay Patel 4a6def05d5 [x86] favor vector constant load to avoid GPR to XMM transfer, part 2
This replaces the build_vector lowering code that was just added in
D80013
and matches the pattern later from the x86-specific "vzext_movl".
That seems to result in the same or better improvements and gets rid
of the 'TODO' items from that patch.

AFAICT, we always shrink wider constant vectors to 128-bit on these
patterns, so we still get the implicit zero-extension to ymm/zmm
without wasting space on larger vector constants. There's a trade-off
there because that means we miss potential load-folding.

Similarly, we could load scalar constants here with implicit
zero-extension even to 128-bit. That saves constant space, but it
means we forego load-folding, and so it increases register pressure.
This seems like a good middle-ground between those 2 options.

Differential Revision: https://reviews.llvm.org/D80131
2020-05-25 08:01:48 -04:00

48 lines
1.2 KiB
LLVM

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-apple-darwin11 -mcpu=core2 -mattr=+mmx,+sse2 | FileCheck %s
; rdar://6602459
@g_v1di = external global <1 x i64>
define void @t1() nounwind {
; CHECK-LABEL: t1:
; CHECK: ## %bb.0: ## %entry
; CHECK-NEXT: pushq %rax
; CHECK-NEXT: callq _return_v1di
; CHECK-NEXT: movq _g_v1di@{{.*}}(%rip), %rcx
; CHECK-NEXT: movq %rax, (%rcx)
; CHECK-NEXT: popq %rax
; CHECK-NEXT: retq
entry:
%call = call <1 x i64> @return_v1di() ; <<1 x i64>> [#uses=0]
store <1 x i64> %call, <1 x i64>* @g_v1di
ret void
}
declare <1 x i64> @return_v1di()
define <1 x i64> @t2() nounwind {
; CHECK-LABEL: t2:
; CHECK: ## %bb.0:
; CHECK-NEXT: movl $1, %eax
; CHECK-NEXT: retq
ret <1 x i64> <i64 1>
}
define <2 x i32> @t3() nounwind {
; CHECK-LABEL: t3:
; CHECK: ## %bb.0:
; CHECK-NEXT: movaps {{.*#+}} xmm0 = [1,0,0,0]
; CHECK-NEXT: retq
ret <2 x i32> <i32 1, i32 0>
}
define double @t4() nounwind {
; CHECK-LABEL: t4:
; CHECK: ## %bb.0:
; CHECK-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
; CHECK-NEXT: retq
ret double bitcast (<2 x i32> <i32 1, i32 0> to double)
}