RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-23 11:13:28 +01:00
Code Issues Projects Releases Wiki Activity
95f0cd3953
llvm-mirror/test/CodeGen/X86/misched-matrix.ll
Craig Topper eb74bf6b9b [X86] Remove some InstRWs for plain store instructions on Sandy Bridge. 
We were forcing the latency of these instructions to 5 cycles, but every other scheduler model had them as 1 cycle. I'm sure I didn't get everything, but this gets a big portion.

llvm-svn: 329339
2018-04-05 20:04:06 +00:00

192 lines
7.7 KiB
LLVM
Raw Blame History

; RUN: llc < %s -mtriple=x86_64-- -mcpu=core2 -pre-RA-sched=source -enable-misched \
; RUN: -misched-topdown -verify-machineinstrs \
; RUN: | FileCheck %s -check-prefix=TOPDOWN
; RUN: llc < %s -mtriple=x86_64-- -mcpu=core2 -pre-RA-sched=source -enable-misched \
; RUN: -misched=ilpmin -verify-machineinstrs \
; RUN: | FileCheck %s -check-prefix=ILPMIN
; RUN: llc < %s -mtriple=x86_64-- -mcpu=core2 -pre-RA-sched=source -enable-misched \
; RUN: -misched=ilpmax -verify-machineinstrs \
; RUN: | FileCheck %s -check-prefix=ILPMAX
;
; Verify that the MI scheduler minimizes register pressure for a
; uniform set of bottom-up subtrees (unrolled matrix multiply).
;
; For current top-down heuristics, ensure that some folded imulls have
; been reordered with the stores. This tests the scheduler's cheap
; alias analysis ability (that doesn't require any AliasAnalysis pass).
;
; TOPDOWN-LABEL: %for.body
; TOPDOWN: movl %{{.*}}, (
; TOPDOWN: imull {{[0-9]*}}(
; TOPDOWN: movl %{{.*}}, 4(
; TOPDOWN: imull {{[0-9]*}}(
; TOPDOWN: movl %{{.*}}, 8(
; TOPDOWN: movl %{{.*}}, 12(
; TOPDOWN-LABEL: %for.end
;
; For -misched=ilpmin, verify that each expression subtree is
; scheduled independently, and that the imull/adds are interleaved.
;
; ILPMIN-LABEL: %for.body
; ILPMIN: movl %{{.*}}, (
; ILPMIN: imull
; ILPMIN: imull
; ILPMIN: addl
; ILPMIN: imull
; ILPMIN: addl
; ILPMIN: imull
; ILPMIN: addl
; ILPMIN: movl %{{.*}}, 4(
; ILPMIN: imull
; ILPMIN: imull
; ILPMIN: addl
; ILPMIN: imull
; ILPMIN: addl
; ILPMIN: imull
; ILPMIN: addl
; ILPMIN: movl %{{.*}}, 8(
; ILPMIN: imull
; ILPMIN: imull
; ILPMIN: addl
; ILPMIN: imull
; ILPMIN: addl
; ILPMIN: imull
; ILPMIN: addl
; ILPMIN: movl %{{.*}}, 12(
; ILPMIN-LABEL: %for.end
;
; For -misched=ilpmax, verify that each expression subtree is
; scheduled independently, and that the imull/adds are clustered.
;
; ILPMAX-LABEL: %for.body
; ILPMAX: movl %{{.*}}, (
; ILPMAX: imull
; ILPMAX: imull
; ILPMAX: imull
; ILPMAX: imull
; ILPMAX: addl
; ILPMAX: addl
; ILPMAX: addl
; ILPMAX: movl %{{.*}}, 4(
; ILPMAX: imull
; ILPMAX: imull
; ILPMAX: imull
; ILPMAX: imull
; ILPMAX: addl
; ILPMAX: addl
; ILPMAX: addl
; ILPMAX: movl %{{.*}}, 8(
; ILPMAX: imull
; ILPMAX: imull
; ILPMAX: imull
; ILPMAX: imull
; ILPMAX: addl
; ILPMAX: addl
; ILPMAX: addl
; ILPMAX: movl %{{.*}}, 12(
; ILPMAX-LABEL: %for.end
define void @mmult([4 x i32]* noalias nocapture %m1, [4 x i32]* noalias nocapture %m2,
[4 x i32]* noalias nocapture %m3) nounwind uwtable ssp {
entry:
br label %for.body
for.body: ; preds = %for.body, %entry
%indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
%arrayidx8 = getelementptr inbounds [4 x i32], [4 x i32]* %m1, i64 %indvars.iv, i64 0
%tmp = load i32, i32* %arrayidx8, align 4
%arrayidx12 = getelementptr inbounds [4 x i32], [4 x i32]* %m2, i64 0, i64 0
%tmp1 = load i32, i32* %arrayidx12, align 4
%arrayidx8.1 = getelementptr inbounds [4 x i32], [4 x i32]* %m1, i64 %indvars.iv, i64 1
%tmp2 = load i32, i32* %arrayidx8.1, align 4
%arrayidx12.1 = getelementptr inbounds [4 x i32], [4 x i32]* %m2, i64 1, i64 0
%tmp3 = load i32, i32* %arrayidx12.1, align 4
%arrayidx8.2 = getelementptr inbounds [4 x i32], [4 x i32]* %m1, i64 %indvars.iv, i64 2
%tmp4 = load i32, i32* %arrayidx8.2, align 4
%arrayidx12.2 = getelementptr inbounds [4 x i32], [4 x i32]* %m2, i64 2, i64 0
%tmp5 = load i32, i32* %arrayidx12.2, align 4
%arrayidx8.3 = getelementptr inbounds [4 x i32], [4 x i32]* %m1, i64 %indvars.iv, i64 3
%tmp6 = load i32, i32* %arrayidx8.3, align 4
%arrayidx12.3 = getelementptr inbounds [4 x i32], [4 x i32]* %m2, i64 3, i64 0
%tmp8 = load i32, i32* %arrayidx8, align 4
%arrayidx12.137 = getelementptr inbounds [4 x i32], [4 x i32]* %m2, i64 0, i64 1
%tmp9 = load i32, i32* %arrayidx12.137, align 4
%tmp10 = load i32, i32* %arrayidx8.1, align 4
%arrayidx12.1.1 = getelementptr inbounds [4 x i32], [4 x i32]* %m2, i64 1, i64 1
%tmp11 = load i32, i32* %arrayidx12.1.1, align 4
%tmp12 = load i32, i32* %arrayidx8.2, align 4
%arrayidx12.2.1 = getelementptr inbounds [4 x i32], [4 x i32]* %m2, i64 2, i64 1
%tmp13 = load i32, i32* %arrayidx12.2.1, align 4
%tmp14 = load i32, i32* %arrayidx8.3, align 4
%arrayidx12.3.1 = getelementptr inbounds [4 x i32], [4 x i32]* %m2, i64 3, i64 1
%tmp15 = load i32, i32* %arrayidx12.3.1, align 4
%tmp16 = load i32, i32* %arrayidx8, align 4
%arrayidx12.239 = getelementptr inbounds [4 x i32], [4 x i32]* %m2, i64 0, i64 2
%tmp17 = load i32, i32* %arrayidx12.239, align 4
%tmp18 = load i32, i32* %arrayidx8.1, align 4
%arrayidx12.1.2 = getelementptr inbounds [4 x i32], [4 x i32]* %m2, i64 1, i64 2
%tmp19 = load i32, i32* %arrayidx12.1.2, align 4
%tmp20 = load i32, i32* %arrayidx8.2, align 4
%arrayidx12.2.2 = getelementptr inbounds [4 x i32], [4 x i32]* %m2, i64 2, i64 2
%tmp21 = load i32, i32* %arrayidx12.2.2, align 4
%tmp22 = load i32, i32* %arrayidx8.3, align 4
%arrayidx12.3.2 = getelementptr inbounds [4 x i32], [4 x i32]* %m2, i64 3, i64 2
%tmp23 = load i32, i32* %arrayidx12.3.2, align 4
%tmp24 = load i32, i32* %arrayidx8, align 4
%arrayidx12.341 = getelementptr inbounds [4 x i32], [4 x i32]* %m2, i64 0, i64 3
%tmp25 = load i32, i32* %arrayidx12.341, align 4
%tmp26 = load i32, i32* %arrayidx8.1, align 4
%arrayidx12.1.3 = getelementptr inbounds [4 x i32], [4 x i32]* %m2, i64 1, i64 3
%tmp27 = load i32, i32* %arrayidx12.1.3, align 4
%tmp28 = load i32, i32* %arrayidx8.2, align 4
%arrayidx12.2.3 = getelementptr inbounds [4 x i32], [4 x i32]* %m2, i64 2, i64 3
%tmp29 = load i32, i32* %arrayidx12.2.3, align 4
%tmp30 = load i32, i32* %arrayidx8.3, align 4
%arrayidx12.3.3 = getelementptr inbounds [4 x i32], [4 x i32]* %m2, i64 3, i64 3
%tmp31 = load i32, i32* %arrayidx12.3.3, align 4
%tmp7 = load i32, i32* %arrayidx12.3, align 4
%mul = mul nsw i32 %tmp1, %tmp
%mul.1 = mul nsw i32 %tmp3, %tmp2
%mul.2 = mul nsw i32 %tmp5, %tmp4
%mul.3 = mul nsw i32 %tmp7, %tmp6
%mul.138 = mul nsw i32 %tmp9, %tmp8
%mul.1.1 = mul nsw i32 %tmp11, %tmp10
%mul.2.1 = mul nsw i32 %tmp13, %tmp12
%mul.3.1 = mul nsw i32 %tmp15, %tmp14
%mul.240 = mul nsw i32 %tmp17, %tmp16
%mul.1.2 = mul nsw i32 %tmp19, %tmp18
%mul.2.2 = mul nsw i32 %tmp21, %tmp20
%mul.3.2 = mul nsw i32 %tmp23, %tmp22
%mul.342 = mul nsw i32 %tmp25, %tmp24
%mul.1.3 = mul nsw i32 %tmp27, %tmp26
%mul.2.3 = mul nsw i32 %tmp29, %tmp28
%mul.3.3 = mul nsw i32 %tmp31, %tmp30
%add.1 = add nsw i32 %mul.1, %mul
%add.2 = add nsw i32 %mul.2, %add.1
%add.3 = add nsw i32 %mul.3, %add.2
%add.1.1 = add nsw i32 %mul.1.1, %mul.138
%add.2.1 = add nsw i32 %mul.2.1, %add.1.1
%add.3.1 = add nsw i32 %mul.3.1, %add.2.1
%add.1.2 = add nsw i32 %mul.1.2, %mul.240
%add.2.2 = add nsw i32 %mul.2.2, %add.1.2
%add.3.2 = add nsw i32 %mul.3.2, %add.2.2
%add.1.3 = add nsw i32 %mul.1.3, %mul.342
%add.2.3 = add nsw i32 %mul.2.3, %add.1.3
%add.3.3 = add nsw i32 %mul.3.3, %add.2.3
%arrayidx16 = getelementptr inbounds [4 x i32], [4 x i32]* %m3, i64 %indvars.iv, i64 0
store i32 %add.3, i32* %arrayidx16, align 4
%arrayidx16.1 = getelementptr inbounds [4 x i32], [4 x i32]* %m3, i64 %indvars.iv, i64 1
store i32 %add.3.1, i32* %arrayidx16.1, align 4
%arrayidx16.2 = getelementptr inbounds [4 x i32], [4 x i32]* %m3, i64 %indvars.iv, i64 2
store i32 %add.3.2, i32* %arrayidx16.2, align 4
%arrayidx16.3 = getelementptr inbounds [4 x i32], [4 x i32]* %m3, i64 %indvars.iv, i64 3
store i32 %add.3.3, i32* %arrayidx16.3, align 4
%indvars.iv.next = add i64 %indvars.iv, 1
%lftr.wideiv = trunc i64 %indvars.iv.next to i32
%exitcond = icmp eq i32 %lftr.wideiv, 4
br i1 %exitcond, label %for.end, label %for.body
for.end: ; preds = %for.body
ret void
}
Reference in New Issue View Git Blame Copy Permalink