maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/polly/test/ScheduleOptimizer/prevectorization-without-ti...

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

75 lines
3.2 KiB
LLVM
Raw Normal View History

[BasicAA] Replace -basicaa with -basic-aa in polly Follow up to https://reviews.llvm.org/D82607.
2020-07-01 06:40:24 +08:00
; RUN: opt -S %loadPolly -basic-aa -polly-opt-isl -polly-tiling=false \
Make optimizations based on pattern matching be enabled by default Currently, pattern based optimizations of Polly can identify matrix multiplication and optimize it according to BLIS matmul optimization pattern (see ScheduleTreeOptimizer for details). This patch makes optimizations based on pattern matching be enabled by default. Reviewed-by: Tobias Grosser <tobias@grosser.es> Differential Revision: https://reviews.llvm.org/D30293 llvm-svn: 295958
2017-02-23 19:44:12 +08:00
; RUN: -polly-pattern-matching-based-opts=false -polly-vectorizer=polly \
; RUN: -polly-ast -analyze < %s | FileCheck %s
Prevectorize the schedule of the band (or the point loop in case of tiling) Contributed-by: Roman Gareev <gareevroman@gmail.com> llvm-svn: 243214
2015-07-25 20:28:56 +08:00
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
@C = common global [1536 x [1536 x float]] zeroinitializer, align 16
@A = common global [1536 x [1536 x float]] zeroinitializer, align 16
@B = common global [1536 x [1536 x float]] zeroinitializer, align 16
; Function Attrs: nounwind uwtable
define void @foo() #0 {
entry:
br label %entry.split
entry.split: ; preds = %entry
br label %for.cond1.preheader
for.cond1.preheader: ; preds = %entry.split, %for.inc28
%indvar4 = phi i64 [ 0, %entry.split ], [ %indvar.next5, %for.inc28 ]
br label %for.body3
for.body3: ; preds = %for.cond1.preheader, %for.inc25
%indvar6 = phi i64 [ 0, %for.cond1.preheader ], [ %indvar.next7, %for.inc25 ]
%arrayidx24 = getelementptr [1536 x [1536 x float]], [1536 x [1536 x float]]* @C, i64 0, i64 %indvar4, i64 %indvar6
store float 0.000000e+00, float* %arrayidx24, align 4
br label %for.body8
for.body8: ; preds = %for.body3, %for.body8
%indvar = phi i64 [ 0, %for.body3 ], [ %indvar.next, %for.body8 ]
%arrayidx16 = getelementptr [1536 x [1536 x float]], [1536 x [1536 x float]]* @A, i64 0, i64 %indvar4, i64 %indvar
%arrayidx20 = getelementptr [1536 x [1536 x float]], [1536 x [1536 x float]]* @B, i64 0, i64 %indvar, i64 %indvar6
%0 = load float, float* %arrayidx24, align 4
%1 = load float, float* %arrayidx16, align 4
%2 = load float, float* %arrayidx20, align 4
%mul = fmul float %1, %2
%add = fadd float %0, %mul
store float %add, float* %arrayidx24, align 4
%indvar.next = add i64 %indvar, 1
%exitcond = icmp ne i64 %indvar.next, 1536
br i1 %exitcond, label %for.body8, label %for.inc25
for.inc25: ; preds = %for.body8
%indvar.next7 = add i64 %indvar6, 1
%exitcond8 = icmp ne i64 %indvar.next7, 1536
br i1 %exitcond8, label %for.body3, label %for.inc28
for.inc28: ; preds = %for.inc25
%indvar.next5 = add i64 %indvar4, 1
%exitcond9 = icmp ne i64 %indvar.next5, 1536
br i1 %exitcond9, label %for.cond1.preheader, label %for.end30
for.end30: ; preds = %for.inc28
ret void
}
Revert "[NFC] remove explicit default value for strboolattr attribute in tests" This reverts commit bda6e5bee04c75b1f1332b4fd1ac4e8ef6c3c247. See https://lab.llvm.org/buildbot/#/builders/109/builds/15424 for instance
2021-05-25 01:43:40 +08:00
attributes #0 = { nounwind uwtable "less-precise-fpmad"="false" "frame-pointer"="all" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
Prevectorize the schedule of the band (or the point loop in case of tiling) Contributed-by: Roman Gareev <gareevroman@gmail.com> llvm-svn: 243214
2015-07-25 20:28:56 +08:00
; CHECK: #pragma known-parallel
; CHECK: for (int c0 = 0; c0 <= 1535; c0 += 1)
Rewrite getPrevectorMap using schedule trees operations Schedule trees are a lot easier to work with, for both humans and machines. For humans the more structured schedule representation is easier to reason about. Together with the more abstract isl programming interface this can result in a lot cleaner code (see this changeset). For machines, the structured schedule and the fact that we now use explicit piecewise affine expressions instead of integer maps makes it easier to generate code from this schedule tree. As a result, we can already see a slight compile-time improvement -- for 3mm from 0m0.593s to 0m0.551s seconds (-7 %). More importantly, future optimizations such as full-partial tile separation will most likely result in more streamlined code to be generated. Contributed-by: Roman Gareev <gareevroman@gmail.com> llvm-svn: 243458
2015-07-29 02:03:36 +08:00
; CHECK: for (int c1 = 0; c1 <= 383; c1 += 1)
Annotation of SIMD loops Use 'mark' nodes annotate a SIMD loop during ScheduleTransformation and skip parallelism checks. The buildbot shows the following compile/execution time changes: Compile time: Improvements Δ Previous Current σ …/gesummv -6.06% 0.2640 0.2480 0.0055 …/gemver -4.46% 0.4480 0.4280 0.0044 …/covariance -4.31% 0.8360 0.8000 0.0065 …/adi -3.23% 0.9920 0.9600 0.0065 …/doitgen -2.53% 0.9480 0.9240 0.0090 …/3mm -2.33% 1.0320 1.0080 0.0087 Execution time: Regressions Δ Previous Current σ …/viterbi 1.70% 5.1840 5.2720 0.0074 …/smallpt 1.06% 12.4920 12.6240 0.0040 Reviewed-by: Tobias Grosser <tobias@grosser.es> Differential Revision: http://reviews.llvm.org/D14491 llvm-svn: 261620
2016-02-23 17:00:13 +08:00
; CHECK: // SIMD
Rewrite getPrevectorMap using schedule trees operations Schedule trees are a lot easier to work with, for both humans and machines. For humans the more structured schedule representation is easier to reason about. Together with the more abstract isl programming interface this can result in a lot cleaner code (see this changeset). For machines, the structured schedule and the fact that we now use explicit piecewise affine expressions instead of integer maps makes it easier to generate code from this schedule tree. As a result, we can already see a slight compile-time improvement -- for 3mm from 0m0.593s to 0m0.551s seconds (-7 %). More importantly, future optimizations such as full-partial tile separation will most likely result in more streamlined code to be generated. Contributed-by: Roman Gareev <gareevroman@gmail.com> llvm-svn: 243458
2015-07-29 02:03:36 +08:00
; CHECK: for (int c2 = 0; c2 <= 3; c2 += 1)
; CHECK: Stmt_for_body3(c0, 4 * c1 + c2);
Prevectorize the schedule of the band (or the point loop in case of tiling) Contributed-by: Roman Gareev <gareevroman@gmail.com> llvm-svn: 243214
2015-07-25 20:28:56 +08:00
; CHECK: #pragma known-parallel
; CHECK: for (int c0 = 0; c0 <= 1535; c0 += 1)
Rewrite getPrevectorMap using schedule trees operations Schedule trees are a lot easier to work with, for both humans and machines. For humans the more structured schedule representation is easier to reason about. Together with the more abstract isl programming interface this can result in a lot cleaner code (see this changeset). For machines, the structured schedule and the fact that we now use explicit piecewise affine expressions instead of integer maps makes it easier to generate code from this schedule tree. As a result, we can already see a slight compile-time improvement -- for 3mm from 0m0.593s to 0m0.551s seconds (-7 %). More importantly, future optimizations such as full-partial tile separation will most likely result in more streamlined code to be generated. Contributed-by: Roman Gareev <gareevroman@gmail.com> llvm-svn: 243458
2015-07-29 02:03:36 +08:00
; CHECK: for (int c1 = 0; c1 <= 383; c1 += 1)
Prevectorize the schedule of the band (or the point loop in case of tiling) Contributed-by: Roman Gareev <gareevroman@gmail.com> llvm-svn: 243214
2015-07-25 20:28:56 +08:00
; CHECK: for (int c2 = 0; c2 <= 1535; c2 += 1)
Annotation of SIMD loops Use 'mark' nodes annotate a SIMD loop during ScheduleTransformation and skip parallelism checks. The buildbot shows the following compile/execution time changes: Compile time: Improvements Δ Previous Current σ …/gesummv -6.06% 0.2640 0.2480 0.0055 …/gemver -4.46% 0.4480 0.4280 0.0044 …/covariance -4.31% 0.8360 0.8000 0.0065 …/adi -3.23% 0.9920 0.9600 0.0065 …/doitgen -2.53% 0.9480 0.9240 0.0090 …/3mm -2.33% 1.0320 1.0080 0.0087 Execution time: Regressions Δ Previous Current σ …/viterbi 1.70% 5.1840 5.2720 0.0074 …/smallpt 1.06% 12.4920 12.6240 0.0040 Reviewed-by: Tobias Grosser <tobias@grosser.es> Differential Revision: http://reviews.llvm.org/D14491 llvm-svn: 261620
2016-02-23 17:00:13 +08:00
; CHECK: // SIMD
Rewrite getPrevectorMap using schedule trees operations Schedule trees are a lot easier to work with, for both humans and machines. For humans the more structured schedule representation is easier to reason about. Together with the more abstract isl programming interface this can result in a lot cleaner code (see this changeset). For machines, the structured schedule and the fact that we now use explicit piecewise affine expressions instead of integer maps makes it easier to generate code from this schedule tree. As a result, we can already see a slight compile-time improvement -- for 3mm from 0m0.593s to 0m0.551s seconds (-7 %). More importantly, future optimizations such as full-partial tile separation will most likely result in more streamlined code to be generated. Contributed-by: Roman Gareev <gareevroman@gmail.com> llvm-svn: 243458
2015-07-29 02:03:36 +08:00
; CHECK: for (int c3 = 0; c3 <= 3; c3 += 1)
; CHECK: Stmt_for_body8(c0, 4 * c1 + c3, c2);
Prevectorize the schedule of the band (or the point loop in case of tiling) Contributed-by: Roman Gareev <gareevroman@gmail.com> llvm-svn: 243214
2015-07-25 20:28:56 +08:00
!llvm.ident = !{!0}
!0 = !{!"clang version 3.5.0 "}