maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/polly/test/Simplify/overwritten_3phi.ll

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

62 lines
1.4 KiB
LLVM
Raw Normal View History

[tests] Set -polly-import-jscop-dir=%S always This simplifies the test cases. llvm-svn: 307645
2017-07-11 18:39:01 +08:00
; RUN: opt %loadPolly -polly-import-jscop -polly-import-jscop-postfix=transformed -polly-simplify -analyze < %s | FileCheck -match-full-lines %s
[Simplify] Remove identical scalar writes. After DeLICM, it is possible to have two writes of the same value to the same location in the same statement when it determined that those writes do not conflict (write the same value). Teach -polly-simplify to remove one of the writes. It interferes with the pattern matching of matrix-multiplication kernels and also seem to not be optimized away by LLVM. The algorthm is simple, has O(n^2) behaviour (n = max number of MemoryAccesses in a statement) and only matches the most obvious cases, but seem to be enough to pattern-match Boost ublas gemm. Not handled cases include: - StoreInst instructions (a.k.a. explicit writes), since the value might be loaded or overwritten between the two stores. - PHINode, especially LCSSA, when the PHI value matches with on other's. - Partial writes (in preparation) llvm-svn: 302805
2017-05-11 23:07:38 +08:00
;
; Remove identical writes
; (two stores in the same statement that write the same value to the same
; destination)
;
; for (int j = 0; j < n; j += 1) {
; body:
; val = 21.0 + 21.0;
; A[1] = val;
; A[1] = val;
; A[1] = val;
;
; user:
; A[0] = A[1];
; }
;
[Simplify] Remove identical write removal. NFC. Removal of overwritten writes currently encompasses all the cases of the identical write removal. There is an observable behavioral change in that the last, instead of the first, MemoryAccess is kept. This should not affect the generated code, however. Differential Revision: https://reviews.llvm.org/D33143 llvm-svn: 302987
2017-05-13 20:20:57 +08:00
define void @overwritten_3phi(i32 %n, double* noalias nonnull %A) {
[Simplify] Remove identical scalar writes. After DeLICM, it is possible to have two writes of the same value to the same location in the same statement when it determined that those writes do not conflict (write the same value). Teach -polly-simplify to remove one of the writes. It interferes with the pattern matching of matrix-multiplication kernels and also seem to not be optimized away by LLVM. The algorthm is simple, has O(n^2) behaviour (n = max number of MemoryAccesses in a statement) and only matches the most obvious cases, but seem to be enough to pattern-match Boost ublas gemm. Not handled cases include: - StoreInst instructions (a.k.a. explicit writes), since the value might be loaded or overwritten between the two stores. - PHINode, especially LCSSA, when the PHI value matches with on other's. - Partial writes (in preparation) llvm-svn: 302805
2017-05-11 23:07:38 +08:00
entry:
br label %for
for:
%j = phi i32 [0, %entry], [%j.inc, %inc]
%j.cmp = icmp slt i32 %j, %n
br i1 %j.cmp, label %body, label %exit
body:
%val = fadd double 21.0, 21.0
br label %user
user:
%phi1 = phi double [%val, %body]
%phi2 = phi double [%val, %body]
%phi3 = phi double [%val, %body]
%add1 = fadd double %phi1, %phi2
%add2 = fadd double %add1, %phi3
store double %add2, double* %A
br label %inc
inc:
%j.inc = add nuw nsw i32 %j, 1
br label %for
exit:
br label %return
return:
ret void
}
; CHECK: Statistics {
[Simplify] Remove identical write removal. NFC. Removal of overwritten writes currently encompasses all the cases of the identical write removal. There is an observable behavioral change in that the last, instead of the first, MemoryAccess is kept. This should not affect the generated code, however. Differential Revision: https://reviews.llvm.org/D33143 llvm-svn: 302987
2017-05-13 20:20:57 +08:00
; CHECK: Overwrites removed: 2
[Simplify] Remove identical scalar writes. After DeLICM, it is possible to have two writes of the same value to the same location in the same statement when it determined that those writes do not conflict (write the same value). Teach -polly-simplify to remove one of the writes. It interferes with the pattern matching of matrix-multiplication kernels and also seem to not be optimized away by LLVM. The algorthm is simple, has O(n^2) behaviour (n = max number of MemoryAccesses in a statement) and only matches the most obvious cases, but seem to be enough to pattern-match Boost ublas gemm. Not handled cases include: - StoreInst instructions (a.k.a. explicit writes), since the value might be loaded or overwritten between the two stores. - PHINode, especially LCSSA, when the PHI value matches with on other's. - Partial writes (in preparation) llvm-svn: 302805
2017-05-11 23:07:38 +08:00
; CHECK: }
; CHECK: Stmt_body
; CHECK-NEXT: MustWriteAccess := [Reduction Type: NONE] [Scalar: 1]
[Simplify] Remove identical write removal. NFC. Removal of overwritten writes currently encompasses all the cases of the identical write removal. There is an observable behavioral change in that the last, instead of the first, MemoryAccess is kept. This should not affect the generated code, however. Differential Revision: https://reviews.llvm.org/D33143 llvm-svn: 302987
2017-05-13 20:20:57 +08:00
; CHECK-NEXT: [n] -> { Stmt_body[i0] -> MemRef_phi3__phi[] };
[Simplify] Remove identical scalar writes. After DeLICM, it is possible to have two writes of the same value to the same location in the same statement when it determined that those writes do not conflict (write the same value). Teach -polly-simplify to remove one of the writes. It interferes with the pattern matching of matrix-multiplication kernels and also seem to not be optimized away by LLVM. The algorthm is simple, has O(n^2) behaviour (n = max number of MemoryAccesses in a statement) and only matches the most obvious cases, but seem to be enough to pattern-match Boost ublas gemm. Not handled cases include: - StoreInst instructions (a.k.a. explicit writes), since the value might be loaded or overwritten between the two stores. - PHINode, especially LCSSA, when the PHI value matches with on other's. - Partial writes (in preparation) llvm-svn: 302805
2017-05-11 23:07:38 +08:00
; CHECK-NEXT: new: [n] -> { Stmt_body[i0] -> MemRef_A[1] };
; CHECK-NEXT: Stmt_user