maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/Transforms/LoopFusion/simple.ll

543 lines
24 KiB
LLVM
Raw Normal View History

[LoopFusion] Regenerate test checks (NFC)
2020-11-03 05:42:06 +08:00
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
[test] Fix LoopFusion tests under NewPM The legacy pass depended on -loop-simplify running. The NPM does not allow for a non-analysis pass to depend on another non-analysis pass.
2020-12-08 09:36:18 +08:00
; RUN: opt -S -loop-simplify -loop-fusion < %s | FileCheck %s
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
@B = common global [1024 x i32] zeroinitializer, align 16
define void @dep_free(i32* noalias %arg) {
[LoopFusion] Regenerate test checks (NFC)
2020-11-03 05:42:06 +08:00
; CHECK-LABEL: @dep_free(
; CHECK-NEXT: bb:
; CHECK-NEXT: br label [[BB7:%.*]]
; CHECK: bb7:
; CHECK-NEXT: [[DOT014:%.*]] = phi i32 [ 0, [[BB:%.*]] ], [ [[TMP15:%.*]], [[BB27:%.*]] ]
; CHECK-NEXT: [[INDVARS_IV23:%.*]] = phi i64 [ 0, [[BB]] ], [ [[INDVARS_IV_NEXT3:%.*]], [[BB27]] ]
; CHECK-NEXT: [[DOT02:%.*]] = phi i32 [ 0, [[BB]] ], [ [[TMP28:%.*]], [[BB27]] ]
; CHECK-NEXT: [[INDVARS_IV1:%.*]] = phi i64 [ 0, [[BB]] ], [ [[INDVARS_IV_NEXT:%.*]], [[BB27]] ]
; CHECK-NEXT: [[TMP:%.*]] = add nsw i32 [[DOT014]], -3
; CHECK-NEXT: [[TMP8:%.*]] = add nuw nsw i64 [[INDVARS_IV23]], 3
; CHECK-NEXT: [[TMP9:%.*]] = trunc i64 [[TMP8]] to i32
; CHECK-NEXT: [[TMP10:%.*]] = mul nsw i32 [[TMP]], [[TMP9]]
; CHECK-NEXT: [[TMP11:%.*]] = trunc i64 [[INDVARS_IV23]] to i32
; CHECK-NEXT: [[TMP12:%.*]] = srem i32 [[TMP10]], [[TMP11]]
; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds i32, i32* [[ARG:%.*]], i64 [[INDVARS_IV23]]
; CHECK-NEXT: store i32 [[TMP12]], i32* [[TMP13]], align 4
; CHECK-NEXT: br label [[BB14:%.*]]
; CHECK: bb14:
; CHECK-NEXT: [[TMP20:%.*]] = add nsw i32 [[DOT02]], -3
; CHECK-NEXT: [[TMP21:%.*]] = add nuw nsw i64 [[INDVARS_IV1]], 3
; CHECK-NEXT: [[TMP22:%.*]] = trunc i64 [[TMP21]] to i32
; CHECK-NEXT: [[TMP23:%.*]] = mul nsw i32 [[TMP20]], [[TMP22]]
; CHECK-NEXT: [[TMP24:%.*]] = trunc i64 [[INDVARS_IV1]] to i32
; CHECK-NEXT: [[TMP25:%.*]] = srem i32 [[TMP23]], [[TMP24]]
; CHECK-NEXT: [[TMP26:%.*]] = getelementptr inbounds [1024 x i32], [1024 x i32]* @B, i64 0, i64 [[INDVARS_IV1]]
; CHECK-NEXT: store i32 [[TMP25]], i32* [[TMP26]], align 4
; CHECK-NEXT: br label [[BB27]]
; CHECK: bb27:
; CHECK-NEXT: [[INDVARS_IV_NEXT3]] = add nuw nsw i64 [[INDVARS_IV23]], 1
; CHECK-NEXT: [[TMP15]] = add nuw nsw i32 [[DOT014]], 1
; CHECK-NEXT: [[EXITCOND4:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT3]], 100
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV1]], 1
; CHECK-NEXT: [[TMP28]] = add nuw nsw i32 [[DOT02]], 1
; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], 100
; CHECK-NEXT: br i1 [[EXITCOND]], label [[BB7]], label [[BB18:%.*]]
; CHECK: bb18:
; CHECK-NEXT: br label [[BB29:%.*]]
; CHECK: bb29:
; CHECK-NEXT: ret void
;
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
bb:
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
br label %bb7
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
bb7: ; preds = %bb, %bb14
%.014 = phi i32 [ 0, %bb ], [ %tmp15, %bb14 ]
%indvars.iv23 = phi i64 [ 0, %bb ], [ %indvars.iv.next3, %bb14 ]
%tmp = add nsw i32 %.014, -3
%tmp8 = add nuw nsw i64 %indvars.iv23, 3
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
%tmp9 = trunc i64 %tmp8 to i32
%tmp10 = mul nsw i32 %tmp, %tmp9
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%tmp11 = trunc i64 %indvars.iv23 to i32
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
%tmp12 = srem i32 %tmp10, %tmp11
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%tmp13 = getelementptr inbounds i32, i32* %arg, i64 %indvars.iv23
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
store i32 %tmp12, i32* %tmp13, align 4
br label %bb14
bb14: ; preds = %bb7
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%indvars.iv.next3 = add nuw nsw i64 %indvars.iv23, 1
%tmp15 = add nuw nsw i32 %.014, 1
%exitcond4 = icmp ne i64 %indvars.iv.next3, 100
br i1 %exitcond4, label %bb7, label %bb17.preheader
bb17.preheader: ; preds = %bb14
br label %bb19
bb19: ; preds = %bb17.preheader, %bb27
%.02 = phi i32 [ 0, %bb17.preheader ], [ %tmp28, %bb27 ]
%indvars.iv1 = phi i64 [ 0, %bb17.preheader ], [ %indvars.iv.next, %bb27 ]
%tmp20 = add nsw i32 %.02, -3
%tmp21 = add nuw nsw i64 %indvars.iv1, 3
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
%tmp22 = trunc i64 %tmp21 to i32
%tmp23 = mul nsw i32 %tmp20, %tmp22
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%tmp24 = trunc i64 %indvars.iv1 to i32
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
%tmp25 = srem i32 %tmp23, %tmp24
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%tmp26 = getelementptr inbounds [1024 x i32], [1024 x i32]* @B, i64 0, i64 %indvars.iv1
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
store i32 %tmp25, i32* %tmp26, align 4
br label %bb27
bb27: ; preds = %bb19
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%indvars.iv.next = add nuw nsw i64 %indvars.iv1, 1
%tmp28 = add nuw nsw i32 %.02, 1
%exitcond = icmp ne i64 %indvars.iv.next, 100
br i1 %exitcond, label %bb19, label %bb18
bb18: ; preds = %bb27
br label %bb29
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
bb29: ; preds = %bb18
ret void
}
define void @dep_free_parametric(i32* noalias %arg, i64 %arg2) {
[LoopFusion] Regenerate test checks (NFC)
2020-11-03 05:42:06 +08:00
; CHECK-LABEL: @dep_free_parametric(
; CHECK-NEXT: bb:
; CHECK-NEXT: [[TMP3:%.*]] = icmp slt i64 0, [[ARG2:%.*]]
; CHECK-NEXT: [[TMP161:%.*]] = icmp slt i64 0, [[ARG2]]
; CHECK-NEXT: br i1 [[TMP3]], label [[BB5_PREHEADER:%.*]], label [[BB27:%.*]]
; CHECK: bb5.preheader:
; CHECK-NEXT: br label [[BB5:%.*]]
; CHECK: bb5:
; CHECK-NEXT: [[DOT014:%.*]] = phi i64 [ [[TMP13:%.*]], [[BB25:%.*]] ], [ 0, [[BB5_PREHEADER]] ]
; CHECK-NEXT: [[DOT02:%.*]] = phi i64 [ [[TMP26:%.*]], [[BB25]] ], [ 0, [[BB5_PREHEADER]] ]
; CHECK-NEXT: [[TMP6:%.*]] = add nsw i64 [[DOT014]], -3
; CHECK-NEXT: [[TMP7:%.*]] = add nuw nsw i64 [[DOT014]], 3
; CHECK-NEXT: [[TMP8:%.*]] = mul nsw i64 [[TMP6]], [[TMP7]]
; CHECK-NEXT: [[TMP9:%.*]] = srem i64 [[TMP8]], [[DOT014]]
; CHECK-NEXT: [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
; CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds i32, i32* [[ARG:%.*]], i64 [[DOT014]]
; CHECK-NEXT: store i32 [[TMP10]], i32* [[TMP11]], align 4
; CHECK-NEXT: br label [[BB12:%.*]]
; CHECK: bb12:
; CHECK-NEXT: [[TMP19:%.*]] = add nsw i64 [[DOT02]], -3
; CHECK-NEXT: [[TMP20:%.*]] = add nuw nsw i64 [[DOT02]], 3
; CHECK-NEXT: [[TMP21:%.*]] = mul nsw i64 [[TMP19]], [[TMP20]]
; CHECK-NEXT: [[TMP22:%.*]] = srem i64 [[TMP21]], [[DOT02]]
; CHECK-NEXT: [[TMP23:%.*]] = trunc i64 [[TMP22]] to i32
; CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1024 x i32], [1024 x i32]* @B, i64 0, i64 [[DOT02]]
; CHECK-NEXT: store i32 [[TMP23]], i32* [[TMP24]], align 4
; CHECK-NEXT: br label [[BB25]]
; CHECK: bb25:
; CHECK-NEXT: [[TMP13]] = add nuw nsw i64 [[DOT014]], 1
; CHECK-NEXT: [[TMP:%.*]] = icmp slt i64 [[TMP13]], [[ARG2]]
; CHECK-NEXT: [[TMP26]] = add nuw nsw i64 [[DOT02]], 1
; CHECK-NEXT: [[TMP16:%.*]] = icmp slt i64 [[TMP26]], [[ARG2]]
; CHECK-NEXT: br i1 [[TMP16]], label [[BB5]], label [[BB27_LOOPEXIT:%.*]]
; CHECK: bb27.loopexit:
; CHECK-NEXT: br label [[BB27]]
; CHECK: bb27:
; CHECK-NEXT: ret void
;
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
bb:
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%tmp3 = icmp slt i64 0, %arg2
br i1 %tmp3, label %bb5, label %bb15.preheader
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
bb5: ; preds = %bb5, %bb12
%.014 = phi i64 [ 0, %bb ], [ %tmp13, %bb12 ]
%tmp6 = add nsw i64 %.014, -3
%tmp7 = add nuw nsw i64 %.014, 3
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
%tmp8 = mul nsw i64 %tmp6, %tmp7
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%tmp9 = srem i64 %tmp8, %.014
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
%tmp10 = trunc i64 %tmp9 to i32
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%tmp11 = getelementptr inbounds i32, i32* %arg, i64 %.014
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
store i32 %tmp10, i32* %tmp11, align 4
br label %bb12
bb12: ; preds = %bb5
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%tmp13 = add nuw nsw i64 %.014, 1
%tmp = icmp slt i64 %tmp13, %arg2
br i1 %tmp, label %bb5, label %bb15.preheader
bb15.preheader: ; preds = %bb12, %bb
%tmp161 = icmp slt i64 0, %arg2
br i1 %tmp161, label %bb18, label %bb27
bb18: ; preds = %bb15.preheader, %bb25
%.02 = phi i64 [ 0, %bb15.preheader ], [ %tmp26, %bb25 ]
%tmp19 = add nsw i64 %.02, -3
%tmp20 = add nuw nsw i64 %.02, 3
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
%tmp21 = mul nsw i64 %tmp19, %tmp20
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%tmp22 = srem i64 %tmp21, %.02
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
%tmp23 = trunc i64 %tmp22 to i32
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%tmp24 = getelementptr inbounds [1024 x i32], [1024 x i32]* @B, i64 0, i64 %.02
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
store i32 %tmp23, i32* %tmp24, align 4
br label %bb25
bb25: ; preds = %bb18
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%tmp26 = add nuw nsw i64 %.02, 1
%tmp16 = icmp slt i64 %tmp26, %arg2
br i1 %tmp16, label %bb18, label %bb27
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
bb27: ; preds = %bb17
ret void
}
define void @raw_only(i32* noalias %arg) {
[LoopFusion] Regenerate test checks (NFC)
2020-11-03 05:42:06 +08:00
; CHECK-LABEL: @raw_only(
; CHECK-NEXT: bb:
; CHECK-NEXT: br label [[BB7:%.*]]
; CHECK: bb7:
; CHECK-NEXT: [[INDVARS_IV22:%.*]] = phi i64 [ 0, [[BB:%.*]] ], [ [[INDVARS_IV_NEXT3:%.*]], [[BB18:%.*]] ]
; CHECK-NEXT: [[INDVARS_IV1:%.*]] = phi i64 [ 0, [[BB]] ], [ [[INDVARS_IV_NEXT:%.*]], [[BB18]] ]
; CHECK-NEXT: [[TMP:%.*]] = getelementptr inbounds i32, i32* [[ARG:%.*]], i64 [[INDVARS_IV22]]
; CHECK-NEXT: [[TMP8:%.*]] = trunc i64 [[INDVARS_IV22]] to i32
; CHECK-NEXT: store i32 [[TMP8]], i32* [[TMP]], align 4
; CHECK-NEXT: br label [[BB9:%.*]]
; CHECK: bb9:
; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds i32, i32* [[ARG]], i64 [[INDVARS_IV1]]
; CHECK-NEXT: [[TMP15:%.*]] = load i32, i32* [[TMP14]], align 4
; CHECK-NEXT: [[TMP16:%.*]] = shl nsw i32 [[TMP15]], 1
; CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1024 x i32], [1024 x i32]* @B, i64 0, i64 [[INDVARS_IV1]]
; CHECK-NEXT: store i32 [[TMP16]], i32* [[TMP17]], align 4
; CHECK-NEXT: br label [[BB18]]
; CHECK: bb18:
; CHECK-NEXT: [[INDVARS_IV_NEXT3]] = add nuw nsw i64 [[INDVARS_IV22]], 1
; CHECK-NEXT: [[EXITCOND4:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT3]], 100
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV1]], 1
; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], 100
; CHECK-NEXT: br i1 [[EXITCOND]], label [[BB7]], label [[BB19:%.*]]
; CHECK: bb19:
; CHECK-NEXT: ret void
;
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
bb:
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
br label %bb7
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
bb11.preheader: ; preds = %bb9
br label %bb13
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
bb7: ; preds = %bb, %bb9
%indvars.iv22 = phi i64 [ 0, %bb ], [ %indvars.iv.next3, %bb9 ]
%tmp = getelementptr inbounds i32, i32* %arg, i64 %indvars.iv22
%tmp8 = trunc i64 %indvars.iv22 to i32
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
store i32 %tmp8, i32* %tmp, align 4
br label %bb9
bb9: ; preds = %bb7
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%indvars.iv.next3 = add nuw nsw i64 %indvars.iv22, 1
%exitcond4 = icmp ne i64 %indvars.iv.next3, 100
br i1 %exitcond4, label %bb7, label %bb11.preheader
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
bb13: ; preds = %bb11.preheader, %bb18
%indvars.iv1 = phi i64 [ 0, %bb11.preheader ], [ %indvars.iv.next, %bb18 ]
%tmp14 = getelementptr inbounds i32, i32* %arg, i64 %indvars.iv1
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
%tmp15 = load i32, i32* %tmp14, align 4
%tmp16 = shl nsw i32 %tmp15, 1
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%tmp17 = getelementptr inbounds [1024 x i32], [1024 x i32]* @B, i64 0, i64 %indvars.iv1
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
store i32 %tmp16, i32* %tmp17, align 4
br label %bb18
bb18: ; preds = %bb13
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%indvars.iv.next = add nuw nsw i64 %indvars.iv1, 1
%exitcond = icmp ne i64 %indvars.iv.next, 100 br i1 %exitcond, label %bb13, label %bb19
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
bb19: ; preds = %bb18
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
ret void
}
define void @raw_only_parametric(i32* noalias %arg, i32 %arg4) {
[LoopFusion] Regenerate test checks (NFC)
2020-11-03 05:42:06 +08:00
; CHECK-LABEL: @raw_only_parametric(
; CHECK-NEXT: bb:
; CHECK-NEXT: [[TMP:%.*]] = sext i32 [[ARG4:%.*]] to i64
; CHECK-NEXT: [[TMP64:%.*]] = icmp sgt i32 [[ARG4]], 0
; CHECK-NEXT: br i1 [[TMP64]], label [[BB8_PREHEADER:%.*]], label [[BB23:%.*]]
; CHECK: bb8.preheader:
; CHECK-NEXT: br label [[BB8:%.*]]
; CHECK: bb8:
; CHECK-NEXT: [[INDVARS_IV25:%.*]] = phi i64 [ [[INDVARS_IV_NEXT3:%.*]], [[BB8]] ], [ 0, [[BB8_PREHEADER]] ]
; CHECK-NEXT: [[INDVARS_IV3:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[BB8]] ], [ 0, [[BB8_PREHEADER]] ]
; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds i32, i32* [[ARG:%.*]], i64 [[INDVARS_IV25]]
; CHECK-NEXT: [[TMP10:%.*]] = trunc i64 [[INDVARS_IV25]] to i32
; CHECK-NEXT: store i32 [[TMP10]], i32* [[TMP9]], align 4
; CHECK-NEXT: [[INDVARS_IV_NEXT3]] = add nuw nsw i64 [[INDVARS_IV25]], 1
; CHECK-NEXT: [[TMP6:%.*]] = icmp slt i64 [[INDVARS_IV_NEXT3]], [[TMP]]
; CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds i32, i32* [[ARG]], i64 [[INDVARS_IV3]]
; CHECK-NEXT: [[TMP19:%.*]] = load i32, i32* [[TMP18]], align 4
; CHECK-NEXT: [[TMP20:%.*]] = shl nsw i32 [[TMP19]], 1
; CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1024 x i32], [1024 x i32]* @B, i64 0, i64 [[INDVARS_IV3]]
; CHECK-NEXT: store i32 [[TMP20]], i32* [[TMP21]], align 4
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV3]], 1
; CHECK-NEXT: [[TMP15:%.*]] = icmp slt i64 [[INDVARS_IV_NEXT]], [[TMP]]
; CHECK-NEXT: br i1 [[TMP15]], label [[BB8]], label [[BB23_LOOPEXIT:%.*]]
; CHECK: bb23.loopexit:
; CHECK-NEXT: br label [[BB23]]
; CHECK: bb23:
; CHECK-NEXT: ret void
;
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
bb:
%tmp = sext i32 %arg4 to i64
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%tmp64 = icmp sgt i32 %arg4, 0
br i1 %tmp64, label %bb8, label %bb23
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
bb8: ; preds = %bb, %bb8
%indvars.iv25 = phi i64 [ %indvars.iv.next3, %bb8 ], [ 0, %bb ]
%tmp9 = getelementptr inbounds i32, i32* %arg, i64 %indvars.iv25
%tmp10 = trunc i64 %indvars.iv25 to i32
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
store i32 %tmp10, i32* %tmp9, align 4
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%indvars.iv.next3 = add nuw nsw i64 %indvars.iv25, 1
%tmp6 = icmp slt i64 %indvars.iv.next3, %tmp
br i1 %tmp6, label %bb8, label %bb17
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
bb17: ; preds = %bb8, %bb17
%indvars.iv3 = phi i64 [ %indvars.iv.next, %bb17 ], [ 0, %bb8 ]
%tmp18 = getelementptr inbounds i32, i32* %arg, i64 %indvars.iv3
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
%tmp19 = load i32, i32* %tmp18, align 4
%tmp20 = shl nsw i32 %tmp19, 1
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%tmp21 = getelementptr inbounds [1024 x i32], [1024 x i32]* @B, i64 0, i64 %indvars.iv3
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
store i32 %tmp20, i32* %tmp21, align 4
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%indvars.iv.next = add nuw nsw i64 %indvars.iv3, 1
%tmp15 = icmp slt i64 %indvars.iv.next, %tmp
br i1 %tmp15, label %bb17, label %bb23
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
bb23: ; preds = %bb17, %bb
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
ret void
}
define void @forward_dep(i32* noalias %arg) {
[LoopFusion] Regenerate test checks (NFC)
2020-11-03 05:42:06 +08:00
; CHECK-LABEL: @forward_dep(
; CHECK-NEXT: bb:
; CHECK-NEXT: br label [[BB7:%.*]]
; CHECK: bb7:
[SCEV] Fix nsw flags for GEP expressions The SCEV code for constructing GEP expressions currently assumes that the addition of the base and all the offsets is nsw if the GEP is inbounds. While the addition of the offsets is indeed nsw, the addition to the base address is not, as the base address is interpreted as an unsigned value. Fix the GEP expression code to not assume nsw for the base+offset calculation. However, do assume nuw if we know that the offset is non-negative. With this, we use the same behavior as the construction of GEP addrecs does. (Modulo the fact that we disregard SCEV unification, as the pre-existing FIXME points out). Differential Revision: https://reviews.llvm.org/D90648
2020-11-03 04:40:08 +08:00
; CHECK-NEXT: [[DOT013:%.*]] = phi i32 [ 0, [[BB:%.*]] ], [ [[TMP15:%.*]], [[BB14:%.*]] ]
; CHECK-NEXT: [[INDVARS_IV22:%.*]] = phi i64 [ 0, [[BB]] ], [ [[INDVARS_IV_NEXT3:%.*]], [[BB14]] ]
[LoopFusion] Regenerate test checks (NFC)
2020-11-03 05:42:06 +08:00
; CHECK-NEXT: [[TMP:%.*]] = add nsw i32 [[DOT013]], -3
; CHECK-NEXT: [[TMP8:%.*]] = add nuw nsw i64 [[INDVARS_IV22]], 3
; CHECK-NEXT: [[TMP9:%.*]] = trunc i64 [[TMP8]] to i32
; CHECK-NEXT: [[TMP10:%.*]] = mul nsw i32 [[TMP]], [[TMP9]]
; CHECK-NEXT: [[TMP11:%.*]] = trunc i64 [[INDVARS_IV22]] to i32
; CHECK-NEXT: [[TMP12:%.*]] = srem i32 [[TMP10]], [[TMP11]]
; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds i32, i32* [[ARG:%.*]], i64 [[INDVARS_IV22]]
; CHECK-NEXT: store i32 [[TMP12]], i32* [[TMP13]], align 4
[SCEV] Fix nsw flags for GEP expressions The SCEV code for constructing GEP expressions currently assumes that the addition of the base and all the offsets is nsw if the GEP is inbounds. While the addition of the offsets is indeed nsw, the addition to the base address is not, as the base address is interpreted as an unsigned value. Fix the GEP expression code to not assume nsw for the base+offset calculation. However, do assume nuw if we know that the offset is non-negative. With this, we use the same behavior as the construction of GEP addrecs does. (Modulo the fact that we disregard SCEV unification, as the pre-existing FIXME points out). Differential Revision: https://reviews.llvm.org/D90648
2020-11-03 04:40:08 +08:00
; CHECK-NEXT: br label [[BB14]]
[LoopFusion] Regenerate test checks (NFC)
2020-11-03 05:42:06 +08:00
; CHECK: bb14:
[SCEV] Fix nsw flags for GEP expressions The SCEV code for constructing GEP expressions currently assumes that the addition of the base and all the offsets is nsw if the GEP is inbounds. While the addition of the offsets is indeed nsw, the addition to the base address is not, as the base address is interpreted as an unsigned value. Fix the GEP expression code to not assume nsw for the base+offset calculation. However, do assume nuw if we know that the offset is non-negative. With this, we use the same behavior as the construction of GEP addrecs does. (Modulo the fact that we disregard SCEV unification, as the pre-existing FIXME points out). Differential Revision: https://reviews.llvm.org/D90648
2020-11-03 04:40:08 +08:00
; CHECK-NEXT: [[INDVARS_IV_NEXT3]] = add nuw nsw i64 [[INDVARS_IV22]], 1
; CHECK-NEXT: [[TMP15]] = add nuw nsw i32 [[DOT013]], 1
; CHECK-NEXT: [[EXITCOND4:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT3]], 100
; CHECK-NEXT: br i1 [[EXITCOND4]], label [[BB7]], label [[BB19_PREHEADER:%.*]]
; CHECK: bb19.preheader:
; CHECK-NEXT: br label [[BB19:%.*]]
; CHECK: bb19:
; CHECK-NEXT: [[INDVARS_IV1:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[BB25:%.*]] ], [ 0, [[BB19_PREHEADER]] ]
[LoopFusion] Regenerate test checks (NFC)
2020-11-03 05:42:06 +08:00
; CHECK-NEXT: [[TMP20:%.*]] = add nsw i64 [[INDVARS_IV1]], -3
; CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds i32, i32* [[ARG]], i64 [[TMP20]]
; CHECK-NEXT: [[TMP22:%.*]] = load i32, i32* [[TMP21]], align 4
; CHECK-NEXT: [[TMP23:%.*]] = mul nsw i32 [[TMP22]], 3
; CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds i32, i32* [[ARG]], i64 [[INDVARS_IV1]]
; CHECK-NEXT: store i32 [[TMP23]], i32* [[TMP24]], align 4
; CHECK-NEXT: br label [[BB25]]
; CHECK: bb25:
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV1]], 1
; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], 100
[SCEV] Fix nsw flags for GEP expressions The SCEV code for constructing GEP expressions currently assumes that the addition of the base and all the offsets is nsw if the GEP is inbounds. While the addition of the offsets is indeed nsw, the addition to the base address is not, as the base address is interpreted as an unsigned value. Fix the GEP expression code to not assume nsw for the base+offset calculation. However, do assume nuw if we know that the offset is non-negative. With this, we use the same behavior as the construction of GEP addrecs does. (Modulo the fact that we disregard SCEV unification, as the pre-existing FIXME points out). Differential Revision: https://reviews.llvm.org/D90648
2020-11-03 04:40:08 +08:00
; CHECK-NEXT: br i1 [[EXITCOND]], label [[BB19]], label [[BB26:%.*]]
[LoopFusion] Regenerate test checks (NFC)
2020-11-03 05:42:06 +08:00
; CHECK: bb26:
; CHECK-NEXT: ret void
;
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
bb:
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
br label %bb7
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
bb7: ; preds = %bb, %bb14
%.013 = phi i32 [ 0, %bb ], [ %tmp15, %bb14 ]
%indvars.iv22 = phi i64 [ 0, %bb ], [ %indvars.iv.next3, %bb14 ]
%tmp = add nsw i32 %.013, -3
%tmp8 = add nuw nsw i64 %indvars.iv22, 3
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
%tmp9 = trunc i64 %tmp8 to i32
%tmp10 = mul nsw i32 %tmp, %tmp9
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%tmp11 = trunc i64 %indvars.iv22 to i32
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
%tmp12 = srem i32 %tmp10, %tmp11
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%tmp13 = getelementptr inbounds i32, i32* %arg, i64 %indvars.iv22
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
store i32 %tmp12, i32* %tmp13, align 4
br label %bb14
bb14: ; preds = %bb7
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%indvars.iv.next3 = add nuw nsw i64 %indvars.iv22, 1
%tmp15 = add nuw nsw i32 %.013, 1
%exitcond4 = icmp ne i64 %indvars.iv.next3, 100
br i1 %exitcond4, label %bb7, label %bb19
bb19: ; preds = %bb14, %bb25
%indvars.iv1 = phi i64 [ 0, %bb14 ], [ %indvars.iv.next, %bb25 ]
%tmp20 = add nsw i64 %indvars.iv1, -3
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
%tmp21 = getelementptr inbounds i32, i32* %arg, i64 %tmp20
%tmp22 = load i32, i32* %tmp21, align 4
%tmp23 = mul nsw i32 %tmp22, 3
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%tmp24 = getelementptr inbounds i32, i32* %arg, i64 %indvars.iv1
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
store i32 %tmp23, i32* %tmp24, align 4
br label %bb25
bb25: ; preds = %bb19
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
%indvars.iv.next = add nuw nsw i64 %indvars.iv1, 1
%exitcond = icmp ne i64 %indvars.iv.next, 100
br i1 %exitcond, label %bb19, label %bb26
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
[LoopFusion] Restrict loop fusion to rotated loops. Summary: This patch restricts loop fusion to only consider rotated loops as valid candidates. This simplifies the analysis and transformation and aligns with other loop optimizations. Reviewers: jdoerfert, Meinersbur, dmgreen, etiotto, Whitney, fhahn, hfinkel Reviewed By: Meinersbur Subscribers: ormris, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D71025
2019-12-17 00:57:52 +08:00
bb26: ; preds = %bb25
Add basic loop fusion pass. This patch adds a basic loop fusion pass. It will fuse loops that conform to the following 4 conditions: 1. Adjacent (no code between them) 2. Control flow equivalent (if one loop executes, the other loop executes) 3. Identical bounds (both loops iterate the same number of iterations) 4. No negative distance dependencies between the loop bodies. The pass does not make any changes to the IR to create opportunities for fusion. Instead, it checks if the necessary conditions are met and if so it fuses two loops together. The pass has not been added to the pass pipeline yet, and thus is not enabled by default. It can be run stand alone using the -loop-fusion option. Differential Revision: https://reviews.llvm.org/D55851 llvm-svn: 358607
2019-04-18 02:53:27 +08:00
ret void
}
[LoopFusion] Move instructions from FC0.Latch to FC1.Latch. Summary:This PR move instructions from FC0.Latch bottom up to the beginning of FC1.Latch as long as they are proven safe. To illustrate why this is beneficial, let's consider the following example: Before Fusion: header1: br header2 header2: br header2, latch1 latch1: br header1, preheader3 preheader3: br header3 header3: br header4 header4: br header4, latch3 latch3: br header3, exit3 After Fusion (before this PR): header1: br header2 header2: br header2, latch1 latch1: br header3 header3: br header4 header4: br header4, latch3 latch3: br header1, exit3 Note that preheader3 is removed during fusion before this PR. Notice that we cannot fuse loop2 with loop4 as there exists block latch1 in between. This PR move instructions from latch1 to beginning of latch3, and remove block latch1. LoopFusion is now able to fuse loop nest recursively. After Fusion (after this PR): header1: br header2 header2: br header3 header3: br header4 header4: br header2, latch3 latch3: br header1, exit3 Reviewer: kbarton, jdoerfert, Meinersbur, dmgreen, fhahn, hfinkel, bmahjour, etiotto Reviewed By: kbarton, Meinersbur Subscribers: hiraditya, llvm-commits Tag: LLVM Differential Revision: https://reviews.llvm.org/D71165
2019-12-18 05:44:06 +08:00
; Test that instructions in loop 1 latch are moved to the beginning of loop 2
; latch iff it is proven safe. %inc.first and %cmp.first are moved, but
; `store i32 0, i32* %Ai.first` is not.
define void @flow_dep(i32* noalias %A, i32* noalias %B) {
[LoopFusion] Regenerate test checks (NFC)
2020-11-03 05:42:06 +08:00
; CHECK-LABEL: @flow_dep(
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[FOR_FIRST:%.*]]
; CHECK: for.first:
; CHECK-NEXT: [[I_FIRST:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INC_FIRST:%.*]], [[FOR_SECOND_LATCH:%.*]] ]
; CHECK-NEXT: [[I_SECOND:%.*]] = phi i64 [ [[INC_SECOND:%.*]], [[FOR_SECOND_LATCH]] ], [ 0, [[ENTRY]] ]
; CHECK-NEXT: [[AI_FIRST:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[I_FIRST]]
; CHECK-NEXT: store i32 0, i32* [[AI_FIRST]], align 4
; CHECK-NEXT: [[AI_SECOND:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[I_SECOND]]
; CHECK-NEXT: [[TMP0:%.*]] = load i32, i32* [[AI_SECOND]], align 4
; CHECK-NEXT: [[BI:%.*]] = getelementptr inbounds i32, i32* [[B:%.*]], i64 [[I_SECOND]]
; CHECK-NEXT: store i32 [[TMP0]], i32* [[BI]], align 4
; CHECK-NEXT: br label [[FOR_SECOND_LATCH]]
; CHECK: for.second.latch:
; CHECK-NEXT: [[INC_FIRST]] = add nsw i64 [[I_FIRST]], 1
; CHECK-NEXT: [[CMP_FIRST:%.*]] = icmp slt i64 [[INC_FIRST]], 100
; CHECK-NEXT: [[INC_SECOND]] = add nsw i64 [[I_SECOND]], 1
; CHECK-NEXT: [[CMP_SECOND:%.*]] = icmp slt i64 [[INC_SECOND]], 100
; CHECK-NEXT: br i1 [[CMP_SECOND]], label [[FOR_FIRST]], label [[FOR_END:%.*]]
; CHECK: for.end:
; CHECK-NEXT: ret void
;
[LoopFusion] Move instructions from FC0.Latch to FC1.Latch. Summary:This PR move instructions from FC0.Latch bottom up to the beginning of FC1.Latch as long as they are proven safe. To illustrate why this is beneficial, let's consider the following example: Before Fusion: header1: br header2 header2: br header2, latch1 latch1: br header1, preheader3 preheader3: br header3 header3: br header4 header4: br header4, latch3 latch3: br header3, exit3 After Fusion (before this PR): header1: br header2 header2: br header2, latch1 latch1: br header3 header3: br header4 header4: br header4, latch3 latch3: br header1, exit3 Note that preheader3 is removed during fusion before this PR. Notice that we cannot fuse loop2 with loop4 as there exists block latch1 in between. This PR move instructions from latch1 to beginning of latch3, and remove block latch1. LoopFusion is now able to fuse loop nest recursively. After Fusion (after this PR): header1: br header2 header2: br header3 header3: br header4 header4: br header2, latch3 latch3: br header1, exit3 Reviewer: kbarton, jdoerfert, Meinersbur, dmgreen, fhahn, hfinkel, bmahjour, etiotto Reviewed By: kbarton, Meinersbur Subscribers: hiraditya, llvm-commits Tag: LLVM Differential Revision: https://reviews.llvm.org/D71165
2019-12-18 05:44:06 +08:00
entry:
br label %for.first
for.first:
%i.first = phi i64 [ 0, %entry ], [ %inc.first, %for.first ]
%Ai.first = getelementptr inbounds i32, i32* %A, i64 %i.first
store i32 0, i32* %Ai.first, align 4
%inc.first = add nsw i64 %i.first, 1
%cmp.first = icmp slt i64 %inc.first, 100
br i1 %cmp.first, label %for.first, label %for.second.preheader
for.second.preheader:
br label %for.second
for.second:
%i.second = phi i64 [ %inc.second, %for.second.latch ], [ 0, %for.second.preheader ]
%Ai.second = getelementptr inbounds i32, i32* %A, i64 %i.second
%0 = load i32, i32* %Ai.second, align 4
%Bi = getelementptr inbounds i32, i32* %B, i64 %i.second
store i32 %0, i32* %Bi, align 4
br label %for.second.latch
for.second.latch:
%inc.second = add nsw i64 %i.second, 1
%cmp.second = icmp slt i64 %inc.second, 100
br i1 %cmp.second, label %for.second, label %for.end
for.end:
ret void
}
[LoopFusion] Move instructions from FC1.Preheader to FC0.Preheader when proven safe. Summary: Currently LoopFusion give up when the second loop nest preheader is not empty. For example: for (int i = 0; i < 100; ++i) {} x+=1; for (int i = 0; i < 100; ++i) {} The above example should be safe to fuse. This PR moves instructions in FC1 preheader (e.g. x+=1; ) to FC0 preheader, which then LoopFusion is able to fuse them. Reviewer: kbarton, Meinersbur, jdoerfert, dmgreen, fhahn, hfinkel, bmahjour, etiotto Reviewed By: jdoerfert Subscribers: hiraditya, llvm-commits Tag: LLVM Differential Revision: https://reviews.llvm.org/D71821
2020-01-29 23:02:21 +08:00
; Test that `%add` is moved in basic block entry, and the two loops for.first
; and for.second are fused.
define i32 @moveinsts_preheader(i32* %A, i32 %x) {
[LoopFusion] Regenerate test checks (NFC)
2020-11-03 05:42:06 +08:00
; CHECK-LABEL: @moveinsts_preheader(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[X:%.*]], 1
; CHECK-NEXT: br label [[FOR_FIRST:%.*]]
; CHECK: for.first:
; CHECK-NEXT: [[I:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INC_I:%.*]], [[FOR_FIRST]] ]
; CHECK-NEXT: [[J:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ [[INC_J:%.*]], [[FOR_FIRST]] ]
; CHECK-NEXT: [[AI:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[I]]
; CHECK-NEXT: store i32 0, i32* [[AI]], align 4
; CHECK-NEXT: [[INC_I]] = add nsw i64 [[I]], 1
; CHECK-NEXT: [[CMP_I:%.*]] = icmp slt i64 [[INC_I]], 100
; CHECK-NEXT: [[AJ:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[J]]
; CHECK-NEXT: store i32 2, i32* [[AJ]], align 4
; CHECK-NEXT: [[INC_J]] = add nsw i64 [[J]], 1
; CHECK-NEXT: [[CMP_J:%.*]] = icmp slt i64 [[INC_J]], 100
; CHECK-NEXT: br i1 [[CMP_J]], label [[FOR_FIRST]], label [[FOR_SECOND_EXIT:%.*]]
; CHECK: for.second.exit:
; CHECK-NEXT: ret i32 [[ADD]]
;
[LoopFusion] Move instructions from FC1.Preheader to FC0.Preheader when proven safe. Summary: Currently LoopFusion give up when the second loop nest preheader is not empty. For example: for (int i = 0; i < 100; ++i) {} x+=1; for (int i = 0; i < 100; ++i) {} The above example should be safe to fuse. This PR moves instructions in FC1 preheader (e.g. x+=1; ) to FC0 preheader, which then LoopFusion is able to fuse them. Reviewer: kbarton, Meinersbur, jdoerfert, dmgreen, fhahn, hfinkel, bmahjour, etiotto Reviewed By: jdoerfert Subscribers: hiraditya, llvm-commits Tag: LLVM Differential Revision: https://reviews.llvm.org/D71821
2020-01-29 23:02:21 +08:00
entry:
br label %for.first
for.first:
%i = phi i64 [ 0, %entry ], [ %inc.i, %for.first ]
%Ai = getelementptr inbounds i32, i32* %A, i64 %i
store i32 0, i32* %Ai, align 4
%inc.i = add nsw i64 %i, 1
%cmp.i = icmp slt i64 %inc.i, 100
br i1 %cmp.i, label %for.first, label %for.first.exit
for.first.exit:
%add = add nsw i32 %x, 1
br label %for.second
for.second:
%j = phi i64 [ 0, %for.first.exit ], [ %inc.j, %for.second ]
%Aj = getelementptr inbounds i32, i32* %A, i64 %j
store i32 2, i32* %Aj, align 4
%inc.j = add nsw i64 %j, 1
%cmp.j = icmp slt i64 %inc.j, 100
br i1 %cmp.j, label %for.second, label %for.second.exit
for.second.exit:
ret i32 %add
}
; Test that `%add` cannot be moved to basic block entry, as it uses %i, which
; defined after basic block entry. And the two loops for.first and for.second
; are not fused.
define i64 @unsafe_preheader(i32* %A, i64 %x) {
[LoopFusion] Regenerate test checks (NFC)
2020-11-03 05:42:06 +08:00
; CHECK-LABEL: @unsafe_preheader(
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[FOR_FIRST:%.*]]
; CHECK: for.first:
; CHECK-NEXT: [[I:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INC_I:%.*]], [[FOR_FIRST]] ]
; CHECK-NEXT: [[AI:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[I]]
; CHECK-NEXT: store i32 0, i32* [[AI]], align 4
; CHECK-NEXT: [[INC_I]] = add nsw i64 [[I]], 1
; CHECK-NEXT: [[CMP_I:%.*]] = icmp slt i64 [[INC_I]], 100
; CHECK-NEXT: br i1 [[CMP_I]], label [[FOR_FIRST]], label [[FOR_FIRST_EXIT:%.*]]
; CHECK: for.first.exit:
; CHECK-NEXT: [[ADD:%.*]] = add nsw i64 [[X:%.*]], [[I]]
; CHECK-NEXT: br label [[FOR_SECOND:%.*]]
; CHECK: for.second:
; CHECK-NEXT: [[J:%.*]] = phi i64 [ 0, [[FOR_FIRST_EXIT]] ], [ [[INC_J:%.*]], [[FOR_SECOND]] ]
; CHECK-NEXT: [[AJ:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[J]]
; CHECK-NEXT: store i32 2, i32* [[AJ]], align 4
; CHECK-NEXT: [[INC_J]] = add nsw i64 [[J]], 1
; CHECK-NEXT: [[CMP_J:%.*]] = icmp slt i64 [[INC_J]], 100
; CHECK-NEXT: br i1 [[CMP_J]], label [[FOR_SECOND]], label [[FOR_SECOND_EXIT:%.*]]
; CHECK: for.second.exit:
; CHECK-NEXT: ret i64 [[ADD]]
;
[LoopFusion] Move instructions from FC1.Preheader to FC0.Preheader when proven safe. Summary: Currently LoopFusion give up when the second loop nest preheader is not empty. For example: for (int i = 0; i < 100; ++i) {} x+=1; for (int i = 0; i < 100; ++i) {} The above example should be safe to fuse. This PR moves instructions in FC1 preheader (e.g. x+=1; ) to FC0 preheader, which then LoopFusion is able to fuse them. Reviewer: kbarton, Meinersbur, jdoerfert, dmgreen, fhahn, hfinkel, bmahjour, etiotto Reviewed By: jdoerfert Subscribers: hiraditya, llvm-commits Tag: LLVM Differential Revision: https://reviews.llvm.org/D71821
2020-01-29 23:02:21 +08:00
entry:
br label %for.first
for.first:
%i = phi i64 [ 0, %entry ], [ %inc.i, %for.first ]
%Ai = getelementptr inbounds i32, i32* %A, i64 %i
store i32 0, i32* %Ai, align 4
%inc.i = add nsw i64 %i, 1
%cmp.i = icmp slt i64 %inc.i, 100
br i1 %cmp.i, label %for.first, label %for.first.exit
for.first.exit:
%add = add nsw i64 %x, %i
br label %for.second
for.second:
%j = phi i64 [ 0, %for.first.exit ], [ %inc.j, %for.second ]
%Aj = getelementptr inbounds i32, i32* %A, i64 %j
store i32 2, i32* %Aj, align 4
%inc.j = add nsw i64 %j, 1
%cmp.j = icmp slt i64 %inc.j, 100
br i1 %cmp.j, label %for.second, label %for.second.exit
for.second.exit:
ret i64 %add
}