mirror of
https://github.com/c64scene-ar/llvm-6502.git
synced 2024-12-15 04:30:12 +00:00
6a7770b7ae
This changes the SelectionDAG scheduling preference to source order. Soon, the SelectionDAG scheduler can be bypassed saving a nice chunk of compile time. Performance differences that result from this change are often a consequence of register coalescing. The register coalescer is far from perfect. Bugs can be filed for deficiencies. On x86 SandyBridge/Haswell, the source order schedule is often preserved, particularly for small blocks. Register pressure is generally improved over the SD scheduler's ILP mode. However, we are still able to handle large blocks that require latency hiding, unlike the SD scheduler's BURR mode. MI scheduler also attempts to discover the critical path in single-block loops and adjust heuristics accordingly. The MI scheduler relies on the new machine model. This is currently unimplemented for AVX, so we may not be generating the best code yet. Unit tests are updated so they don't depend on SD scheduling heuristics. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192750 91177308-0d34-0410-b5e6-96231b3b80d8
87 lines
2.8 KiB
LLVM
87 lines
2.8 KiB
LLVM
; RUN: llc < %s -mtriple=x86_64-linux -mcpu=penryn | FileCheck %s
|
|
; rdar://10050222, rdar://10134392
|
|
|
|
define <4 x float> @t1(<4 x float> %a, <1 x i64>* nocapture %p) nounwind {
|
|
entry:
|
|
; CHECK-LABEL: t1:
|
|
; CHECK: movlps (%rdi), %xmm0
|
|
; CHECK: ret
|
|
%p.val = load <1 x i64>* %p, align 1
|
|
%0 = bitcast <1 x i64> %p.val to <2 x float>
|
|
%shuffle.i = shufflevector <2 x float> %0, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
|
|
%shuffle1.i = shufflevector <4 x float> %a, <4 x float> %shuffle.i, <4 x i32> <i32 4, i32 5, i32 2, i32 3>
|
|
ret <4 x float> %shuffle1.i
|
|
}
|
|
|
|
define <4 x float> @t1a(<4 x float> %a, <1 x i64>* nocapture %p) nounwind {
|
|
entry:
|
|
; CHECK-LABEL: t1a:
|
|
; CHECK: movlps (%rdi), %xmm0
|
|
; CHECK: ret
|
|
%0 = bitcast <1 x i64>* %p to double*
|
|
%1 = load double* %0
|
|
%2 = insertelement <2 x double> undef, double %1, i32 0
|
|
%3 = bitcast <2 x double> %2 to <4 x float>
|
|
%4 = shufflevector <4 x float> %a, <4 x float> %3, <4 x i32> <i32 4, i32 5, i32 2, i32 3>
|
|
ret <4 x float> %4
|
|
}
|
|
|
|
define void @t2(<1 x i64>* nocapture %p, <4 x float> %a) nounwind {
|
|
entry:
|
|
; CHECK-LABEL: t2:
|
|
; CHECK: movlps %xmm0, (%rdi)
|
|
; CHECK: ret
|
|
%cast.i = bitcast <4 x float> %a to <2 x i64>
|
|
%extract.i = extractelement <2 x i64> %cast.i, i32 0
|
|
%0 = getelementptr inbounds <1 x i64>* %p, i64 0, i64 0
|
|
store i64 %extract.i, i64* %0, align 8
|
|
ret void
|
|
}
|
|
|
|
define void @t2a(<1 x i64>* nocapture %p, <4 x float> %a) nounwind {
|
|
entry:
|
|
; CHECK-LABEL: t2a:
|
|
; CHECK: movlps %xmm0, (%rdi)
|
|
; CHECK: ret
|
|
%0 = bitcast <1 x i64>* %p to double*
|
|
%1 = bitcast <4 x float> %a to <2 x double>
|
|
%2 = extractelement <2 x double> %1, i32 0
|
|
store double %2, double* %0
|
|
ret void
|
|
}
|
|
|
|
; rdar://10436044
|
|
define <2 x double> @t3() nounwind readonly {
|
|
bb:
|
|
; CHECK-LABEL: t3:
|
|
; CHECK: movq (%rax), %xmm1
|
|
; CHECK: punpcklqdq %xmm2, %xmm0
|
|
; CHECK: movsd %xmm1, %xmm0
|
|
%tmp0 = load i128* null, align 1
|
|
%tmp1 = load <2 x i32>* undef, align 8
|
|
%tmp2 = bitcast i128 %tmp0 to <16 x i8>
|
|
%tmp3 = bitcast <2 x i32> %tmp1 to i64
|
|
%tmp4 = insertelement <2 x i64> undef, i64 %tmp3, i32 0
|
|
%tmp5 = bitcast <16 x i8> %tmp2 to <2 x double>
|
|
%tmp6 = bitcast <2 x i64> %tmp4 to <2 x double>
|
|
%tmp7 = shufflevector <2 x double> %tmp5, <2 x double> %tmp6, <2 x i32> <i32 2, i32 1>
|
|
ret <2 x double> %tmp7
|
|
}
|
|
|
|
; rdar://10450317
|
|
define <2 x i64> @t4() nounwind readonly {
|
|
bb:
|
|
; CHECK-LABEL: t4:
|
|
; CHECK: movq (%rax), %xmm0
|
|
; CHECK: punpcklqdq %{{xmm.}}, %[[XMM:xmm[0-9]]]
|
|
; CHECK: movsd %[[XMM]], %xmm0
|
|
%tmp0 = load i128* null, align 1
|
|
%tmp1 = load <2 x i32>* undef, align 8
|
|
%tmp2 = bitcast i128 %tmp0 to <16 x i8>
|
|
%tmp3 = bitcast <2 x i32> %tmp1 to i64
|
|
%tmp4 = insertelement <2 x i64> undef, i64 %tmp3, i32 0
|
|
%tmp5 = bitcast <16 x i8> %tmp2 to <2 x i64>
|
|
%tmp6 = shufflevector <2 x i64> %tmp4, <2 x i64> %tmp5, <2 x i32> <i32 2, i32 1>
|
|
ret <2 x i64> %tmp6
|
|
}
|