mirror of
https://github.com/capstone-engine/llvm-capstone.git
synced 2025-04-03 05:32:24 +00:00
2faa569c0a
This replaces the use of %defaultOpts = '-basicaa -polly-prepare' with the minimal set of passes necessary for a test to succeed. Of the test cases that previously used %defaultOpts 76 test cases require none of these passes, 42 need -basicaa and only 2 need -polly-prepare. Our change makes this requirement explicit. In Polly many test cases have been using a macro '%defaultOpts' which run a couple of preparing passes before the actual Polly test case. This macro was introduced very early in the development of Polly and originally contained a large set of canonicalization passes. However, as the need for additional canonicalization passes makes test cases harder to understand and also more fragile in terms of changes in such passes, we aim since a longer time to only include the minimal set of passes necessary. This patch removes the last leftovers from of %defaultOpts and brings our tests cases more in line to what is usus in LLVM itself. llvm-svn: 216815
209 lines
7.4 KiB
LLVM
209 lines
7.4 KiB
LLVM
; RUN: opt %loadPolly -basicaa -polly-cloog -polly-codegen -enable-polly-openmp -analyze < %s | FileCheck %s
|
|
; RUN: opt %loadPolly -basicaa -polly-import-jscop -polly-import-jscop-dir=%S -polly-cloog -polly-codegen -enable-polly-openmp -analyze < %s | FileCheck -check-prefix=IMPORT %s
|
|
; RUN: opt %loadPolly -basicaa -polly-import-jscop -polly-import-jscop-dir=%S -polly-cloog -polly-codegen -enable-polly-openmp -analyze < %s | FileCheck -check-prefix=IMPORT %s
|
|
; RUN: opt %loadPolly -basicaa -polly-import-jscop -polly-import-jscop-postfix=tiled -polly-import-jscop-dir=%S -polly-cloog -polly-codegen -enable-polly-openmp -analyze -disable-polly-legality < %s | FileCheck -check-prefix=TILED %s
|
|
|
|
;#define M 1024
|
|
;#define N 1024
|
|
;#define K 1024
|
|
;
|
|
;float A[M][K], B[K][N], C[M][N], X[K];
|
|
;
|
|
;float parallel_loop() {
|
|
; int i, j, k;
|
|
;
|
|
; for (i = 0; i < M; i++)
|
|
; for (j = 0; j< N; j++)
|
|
; for (k = 0; k < K; k++)
|
|
; C[i][j] += A[i][k] * B[k][j];
|
|
;
|
|
; for (i = 0; i < M; i++)
|
|
; for (j = 0; j < N; j++)
|
|
; for (k = 0; k < K; k++)
|
|
; X[k] += X[k];
|
|
;
|
|
; return C[42][42] + X[42];
|
|
;}
|
|
|
|
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
|
|
target triple = "x86_64-unknown-linux-gnu"
|
|
|
|
@A = common global [1024 x [1024 x float]] zeroinitializer, align 16
|
|
@B = common global [1024 x [1024 x float]] zeroinitializer, align 16
|
|
@C = common global [1024 x [1024 x float]] zeroinitializer, align 16
|
|
@X = common global [1024 x float] zeroinitializer, align 16
|
|
|
|
define float @parallel_loop() nounwind {
|
|
bb:
|
|
br label %bb18
|
|
|
|
bb18: ; preds = %bb32, %bb
|
|
%indvar9 = phi i64 [ %indvar.next10, %bb32 ], [ 0, %bb ]
|
|
%exitcond15 = icmp ne i64 %indvar9, 1024
|
|
br i1 %exitcond15, label %bb19, label %bb33
|
|
|
|
bb19: ; preds = %bb18
|
|
br label %bb20
|
|
|
|
bb20: ; preds = %bb30, %bb19
|
|
%indvar6 = phi i64 [ %indvar.next7, %bb30 ], [ 0, %bb19 ]
|
|
%scevgep14 = getelementptr [1024 x [1024 x float]]* @C, i64 0, i64 %indvar9, i64 %indvar6
|
|
%exitcond12 = icmp ne i64 %indvar6, 1024
|
|
br i1 %exitcond12, label %bb21, label %bb31
|
|
|
|
bb21: ; preds = %bb20
|
|
br label %bb22
|
|
|
|
bb22: ; preds = %bb28, %bb21
|
|
%indvar3 = phi i64 [ %indvar.next4, %bb28 ], [ 0, %bb21 ]
|
|
%scevgep11 = getelementptr [1024 x [1024 x float]]* @A, i64 0, i64 %indvar9, i64 %indvar3
|
|
%scevgep8 = getelementptr [1024 x [1024 x float]]* @B, i64 0, i64 %indvar3, i64 %indvar6
|
|
%exitcond5 = icmp ne i64 %indvar3, 1024
|
|
br i1 %exitcond5, label %bb23, label %bb29
|
|
|
|
bb23: ; preds = %bb22
|
|
%tmp = load float* %scevgep11, align 4
|
|
%tmp24 = load float* %scevgep8, align 4
|
|
%tmp25 = fmul float %tmp, %tmp24
|
|
%tmp26 = load float* %scevgep14, align 4
|
|
%tmp27 = fadd float %tmp26, %tmp25
|
|
store float %tmp27, float* %scevgep14, align 4
|
|
br label %bb28
|
|
|
|
bb28: ; preds = %bb23
|
|
%indvar.next4 = add i64 %indvar3, 1
|
|
br label %bb22
|
|
|
|
bb29: ; preds = %bb22
|
|
br label %bb30
|
|
|
|
bb30: ; preds = %bb29
|
|
%indvar.next7 = add i64 %indvar6, 1
|
|
br label %bb20
|
|
|
|
bb31: ; preds = %bb20
|
|
br label %bb32
|
|
|
|
bb32: ; preds = %bb31
|
|
%indvar.next10 = add i64 %indvar9, 1
|
|
br label %bb18
|
|
|
|
bb33: ; preds = %bb18
|
|
br label %bb34
|
|
|
|
bb34: ; preds = %bb48, %bb33
|
|
%i.1 = phi i32 [ 0, %bb33 ], [ %tmp49, %bb48 ]
|
|
%exitcond2 = icmp ne i32 %i.1, 1024
|
|
br i1 %exitcond2, label %bb35, label %bb50
|
|
|
|
bb35: ; preds = %bb34
|
|
br label %bb36
|
|
|
|
bb36: ; preds = %bb45, %bb35
|
|
%j.1 = phi i32 [ 0, %bb35 ], [ %tmp46, %bb45 ]
|
|
%exitcond1 = icmp ne i32 %j.1, 1024
|
|
br i1 %exitcond1, label %bb37, label %bb47
|
|
|
|
bb37: ; preds = %bb36
|
|
br label %bb38
|
|
|
|
bb38: ; preds = %bb43, %bb37
|
|
%indvar = phi i64 [ %indvar.next, %bb43 ], [ 0, %bb37 ]
|
|
%scevgep = getelementptr [1024 x float]* @X, i64 0, i64 %indvar
|
|
%exitcond = icmp ne i64 %indvar, 1024
|
|
br i1 %exitcond, label %bb39, label %bb44
|
|
|
|
bb39: ; preds = %bb38
|
|
%tmp40 = load float* %scevgep, align 4
|
|
%tmp41 = load float* %scevgep, align 4
|
|
%tmp42 = fadd float %tmp41, %tmp40
|
|
store float %tmp42, float* %scevgep, align 4
|
|
br label %bb43
|
|
|
|
bb43: ; preds = %bb39
|
|
%indvar.next = add i64 %indvar, 1
|
|
br label %bb38
|
|
|
|
bb44: ; preds = %bb38
|
|
br label %bb45
|
|
|
|
bb45: ; preds = %bb44
|
|
%tmp46 = add nsw i32 %j.1, 1
|
|
br label %bb36
|
|
|
|
bb47: ; preds = %bb36
|
|
br label %bb48
|
|
|
|
bb48: ; preds = %bb47
|
|
%tmp49 = add nsw i32 %i.1, 1
|
|
br label %bb34
|
|
|
|
bb50: ; preds = %bb34
|
|
%tmp51 = load float* getelementptr inbounds ([1024 x [1024 x float]]* @C, i64 0, i64 42, i64 42), align 8
|
|
%tmp52 = load float* getelementptr inbounds ([1024 x float]* @X, i64 0, i64 42), align 8
|
|
%tmp53 = fadd float %tmp51, %tmp52
|
|
ret float %tmp53
|
|
}
|
|
|
|
; CHECK: for (c2=0;c2<=1023;c2++) {
|
|
; CHECK: for (c4=0;c4<=1023;c4++) {
|
|
; CHECK: for (c6=0;c6<=1023;c6++) {
|
|
; CHECK: Stmt_bb23(c2,c4,c6);
|
|
; CHECK: }
|
|
; CHECK: }
|
|
; CHECK: }
|
|
; CHECK: for (c2=0;c2<=1023;c2++) {
|
|
; CHECK: for (c4=0;c4<=1023;c4++) {
|
|
; CHECK: for (c6=0;c6<=1023;c6++) {
|
|
; CHECK: Stmt_bb39(c2,c4,c6);
|
|
; CHECK: }
|
|
; CHECK: }
|
|
; CHECK: }
|
|
; CHECK: Parallel loop with iterator 'c2' generated
|
|
; CHECK: Parallel loop with iterator 'c6' generated
|
|
; CHECK-NOT: Parallel loop
|
|
|
|
|
|
; IMPORT: for (c2=0;c2<=1023;c2++) {
|
|
; IMPORT: for (c4=0;c4<=1023;c4++) {
|
|
; IMPORT: for (c6=0;c6<=1023;c6++) {
|
|
; IMPORT: Stmt_bb23(c2,c4,c6);
|
|
; IMPORT: Stmt_bb39(c2,c4,c6);
|
|
; IMPORT: }
|
|
; IMPORT: }
|
|
; IMPORT: }
|
|
; IMPORT-NOT: Parallel loop
|
|
|
|
; TILED: for (c2=0;c2<=1023;c2+=4) {
|
|
; TILED: for (c4=0;c4<=1023;c4+=4) {
|
|
; TILED: for (c6=0;c6<=1023;c6+=4) {
|
|
; TILED: for (c8=c2;c8<=c2+3;c8++) {
|
|
; TILED: for (c9=c4;c9<=c4+3;c9++) {
|
|
; TILED: for (c10=c6;c10<=c6+3;c10++) {
|
|
; TILED: Stmt_bb23(c8,c9,c10);
|
|
; TILED: }
|
|
; TILED: }
|
|
; TILED: }
|
|
; TILED: }
|
|
; TILED: }
|
|
; TILED: }
|
|
; TILED: for (c2=0;c2<=1023;c2+=4) {
|
|
; TILED: for (c4=0;c4<=1023;c4+=4) {
|
|
; TILED: for (c6=0;c6<=1023;c6+=4) {
|
|
; TILED: for (c8=c2;c8<=c2+3;c8++) {
|
|
; TILED: for (c9=c4;c9<=c4+3;c9++) {
|
|
; TILED: for (c10=c6;c10<=c6+3;c10++) {
|
|
; TILED: Stmt_bb39(c8,c9,c10);
|
|
; TILED: }
|
|
; TILED: }
|
|
; TILED: }
|
|
; TILED: }
|
|
; TILED: }
|
|
; TILED: }
|
|
; I am not sure if we actually may have parallel loops here. The dependency
|
|
; analysis does not detect any. This may however be because we do not
|
|
; correctly update the imported schedule. Add a check that hopefully fails
|
|
; after this is corrected. Or someone proves there are no parallel loops and
|
|
; we can remove this comment.
|
|
; TILDED-NOT: Parallel loop
|