diff options
| author | Chad Rosier <mcrosier@codeaurora.org> | 2014-03-21 19:34:41 +0000 |
|---|---|---|
| committer | Chad Rosier <mcrosier@codeaurora.org> | 2014-03-21 19:34:41 +0000 |
| commit | 1eb67a4f84d49d88454e2b6142d30e81c839209f (patch) | |
| tree | 6817ec55593037171b7a04f1730dc317f6507fa0 /test | |
| parent | 218c7c821b955398baf144199ac69d9c9665e102 (diff) | |
| download | llvm-1eb67a4f84d49d88454e2b6142d30e81c839209f.tar.gz llvm-1eb67a4f84d49d88454e2b6142d30e81c839209f.tar.bz2 llvm-1eb67a4f84d49d88454e2b6142d30e81c839209f.tar.xz | |
[AArch64] Add SchedRW lists to NEON instructions.
Previously, only regular AArch64 instructions were annotated with SchedRW lists.
This patch does the same for NEON enabling these instructions to be scheduled by
the MIScheduler. Additionally, store operations are now modeled and a few
SchedRW lists were updated for bug fixes (e.g. multiple def operands).
Reviewers: apazos, mcrosier, atrick
Patch by Dave Estes <cestes@codeaurora.org>!
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204505 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'test')
| -rw-r--r-- | test/CodeGen/AArch64/misched-basic-A53.ll | 31 |
1 files changed, 30 insertions, 1 deletions
diff --git a/test/CodeGen/AArch64/misched-basic-A53.ll b/test/CodeGen/AArch64/misched-basic-A53.ll index 0d5534eca5..1555c4868e 100644 --- a/test/CodeGen/AArch64/misched-basic-A53.ll +++ b/test/CodeGen/AArch64/misched-basic-A53.ll @@ -4,13 +4,15 @@ ; The Cortex-A53 machine model will cause the MADD instruction to be scheduled ; much higher than the ADD instructions in order to hide latency. When not ; specifying a subtarget, the MADD will remain near the end of the block. +; +; CHECK: ********** MI Scheduling ********** ; CHECK: main ; CHECK: *** Final schedule for BB#2 *** ; CHECK: SU(13) ; CHECK: MADDwwww ; CHECK: SU(4) ; CHECK: ADDwwi_lsl0_s -; CHECK: ********** MI Scheduling ********** +; CHECK: ********** INTERVALS ********** @main.x = private unnamed_addr constant [8 x i32] [i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1], align 4 @main.y = private unnamed_addr constant [8 x i32] [i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2], align 4 @@ -76,6 +78,33 @@ for.end: ; preds = %for.cond ret i32 %add6 } + +; The Cortex-A53 machine model will cause the FDIVvvv_42 to be raised to +; hide latency. Whereas normally there would only be a single FADDvvv_4s +; after it, this test checks to make sure there are more than one. +; +; CHECK: ********** MI Scheduling ********** +; CHECK: neon4xfloat:BB#0 +; CHECK: *** Final schedule for BB#0 *** +; CHECK: FDIVvvv_4S +; CHECK: FADDvvv_4S +; CHECK: FADDvvv_4S +; CHECK: ********** INTERVALS ********** +define <4 x float> @neon4xfloat(<4 x float> %A, <4 x float> %B) { + %tmp1 = fadd <4 x float> %A, %B; + %tmp2 = fadd <4 x float> %A, %tmp1; + %tmp3 = fadd <4 x float> %A, %tmp2; + %tmp4 = fadd <4 x float> %A, %tmp3; + %tmp5 = fadd <4 x float> %A, %tmp4; + %tmp6 = fadd <4 x float> %A, %tmp5; + %tmp7 = fadd <4 x float> %A, %tmp6; + %tmp8 = fadd <4 x float> %A, %tmp7; + %tmp9 = fdiv <4 x float> %A, %B; + %tmp10 = fadd <4 x float> %tmp8, %tmp9; + + ret <4 x float> %tmp10 +} + ; Function Attrs: nounwind declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture readonly, i64, i32, i1) #1 |