[Pipelines] Move IPSCCP after inliner pipeline #96620
Conversation
Moving the Interprocedural Constant Propagation (IPSCCP) pass to run after the inliner pipeline can enhance optimization effectiveness. Performance uplift for SPEC2017:548.exchange2_r on rv64gc is over 40%.
Please update these tests. |
|
This patch causes some significant performance regressions on llvm-test-suite (rv64gc-O3-thinlto):
|
llvmbot
added
clang
|
@llvm/pr-subscribers-llvm-transforms @llvm/pr-subscribers-clang Author: SiHuaN (sihuan) ChangesThis patch significantly improves the performance of LLVM for the SPEC2017:548.exchange2_r benchmark, with a performance uplift of over 40% on the rv64gc. During our investigation into the significant performance disparity between GCC and LLVM on the SPEC2017:548.exchange2_r benchmark on RISC-V, we identified that the primary difference stems from constant propagation optimization. $ objdump -D exchange2_r_gcc | grep "digits_2.*:$"
000000000001d480 <__brute_force_MOD_digits_2.isra.0>:
000000000001f0f6 <__brute_force_MOD_digits_2.constprop.7.isra.0>:
000000000001fdd0 <__brute_force_MOD_digits_2.constprop.6.isra.0>:
0000000000020900 <__brute_force_MOD_digits_2.constprop.5.isra.0>:
00000000000211c4 <__brute_force_MOD_digits_2.constprop.4.isra.0>:
0000000000022002 <__brute_force_MOD_digits_2.constprop.3.isra.0>:
0000000000022d6a <__brute_force_MOD_digits_2.constprop.2.isra.0>:
0000000000023898 <__brute_force_MOD_digits_2.constprop.1.isra.0>:However, in LLVM, this function is not split: $ objdump -D exchange2_r_llvm | grep "digits_2.*:$"
00000000000115a0 <_QMbrute_forcePdigits_2>:By applying this patch, LLVM now exhibits similar behavior, resulting in a substantial performance uplift. $ objdump -D exchange2_r_patched_llvm | grep "digits_2.*:$"
0000000000011ab0 <_QMbrute_forcePdigits_2>:
0000000000018a4e <_QMbrute_forcePdigits_2.specialized.1>:
0000000000019820 <_QMbrute_forcePdigits_2.specialized.2>:
000000000001a436 <_QMbrute_forcePdigits_2.specialized.3>:
000000000001ae78 <_QMbrute_forcePdigits_2.specialized.4>:
000000000001ba8e <_QMbrute_forcePdigits_2.specialized.5>:
000000000001c7e6 <_QMbrute_forcePdigits_2.specialized.6>:
000000000001d072 <_QMbrute_forcePdigits_2.specialized.7>:
000000000001dad0 <_QMbrute_forcePdigits_2.specialized.8>:And we used
Additionally, I performed tests on x86_64, yielding similar results:
Full diff: https://github.com/llvm/llvm-project/pull/96620.diff 12 Files Affected:
diff --git a/clang/test/CodeGen/attr-counted-by.c b/clang/test/CodeGen/attr-counted-by.c
index 79922eb4159f1..8d0e39d0e3dad 100644
--- a/clang/test/CodeGen/attr-counted-by.c
+++ b/clang/test/CodeGen/attr-counted-by.c
@@ -639,7 +639,7 @@ void test6(struct anon_struct *p, int index) {
p->array[index] = __builtin_dynamic_object_size(p->array, 1);
}
-// SANITIZE-WITH-ATTR-LABEL: define dso_local i64 @test6_bdos(
+// SANITIZE-WITH-ATTR-LABEL: define dso_local range(i64 0, -3) i64 @test6_bdos(
// SANITIZE-WITH-ATTR-SAME: ptr nocapture noundef readonly [[P:%.*]]) local_unnamed_addr #[[ATTR2]] {
// SANITIZE-WITH-ATTR-NEXT: entry:
// SANITIZE-WITH-ATTR-NEXT: [[DOT_COUNTED_BY_GEP:%.*]] = getelementptr inbounds i8, ptr [[P]], i64 8
@@ -649,7 +649,7 @@ void test6(struct anon_struct *p, int index) {
// SANITIZE-WITH-ATTR-NEXT: [[TMP1:%.*]] = select i1 [[DOTINV]], i64 0, i64 [[TMP0]]
// SANITIZE-WITH-ATTR-NEXT: ret i64 [[TMP1]]
//
-// NO-SANITIZE-WITH-ATTR-LABEL: define dso_local i64 @test6_bdos(
+// NO-SANITIZE-WITH-ATTR-LABEL: define dso_local range(i64 0, -3) i64 @test6_bdos(
// NO-SANITIZE-WITH-ATTR-SAME: ptr nocapture noundef readonly [[P:%.*]]) local_unnamed_addr #[[ATTR2]] {
// NO-SANITIZE-WITH-ATTR-NEXT: entry:
// NO-SANITIZE-WITH-ATTR-NEXT: [[DOT_COUNTED_BY_GEP:%.*]] = getelementptr inbounds i8, ptr [[P]], i64 8
diff --git a/llvm/lib/Passes/PassBuilderPipelines.cpp b/llvm/lib/Passes/PassBuilderPipelines.cpp
index 926515c9508a9..5659c116e9c95 100644
--- a/llvm/lib/Passes/PassBuilderPipelines.cpp
+++ b/llvm/lib/Passes/PassBuilderPipelines.cpp
@@ -1204,6 +1204,12 @@ PassBuilder::buildModuleSimplificationPipeline(OptimizationLevel Level,
else
MPM.addPass(buildInlinerPipeline(Level, Phase));
+ MPM.addPass(IPSCCPPass(
+ IPSCCPOptions(/*AllowFuncSpec=*/
+ Level != OptimizationLevel::Os &&
+ Level != OptimizationLevel::Oz &&
+ !isLTOPreLink(Phase))));
+
// Remove any dead arguments exposed by cleanups, constant folding globals,
// and argument promotion.
MPM.addPass(DeadArgumentEliminationPass());
diff --git a/llvm/test/Other/new-pm-defaults.ll b/llvm/test/Other/new-pm-defaults.ll
index 489aed40c190b..32fd2e2d3ffbf 100644
--- a/llvm/test/Other/new-pm-defaults.ll
+++ b/llvm/test/Other/new-pm-defaults.ll
@@ -229,6 +229,7 @@
; CHECK-O-NEXT: Running pass: InvalidateAnalysisPass<{{.*}}ShouldNotRunFunctionPassesAnalysis
; CHECK-O-NEXT: Invalidating analysis: ShouldNotRunFunctionPassesAnalysis
; CHECK-O-NEXT: Invalidating analysis: InlineAdvisorAnalysis
+; CHECK-O-NEXT: Running pass: IPSCCPPass
; CHECK-O-NEXT: Running pass: DeadArgumentEliminationPass
; CHECK-O-NEXT: Running pass: CoroCleanupPass
; CHECK-O-NEXT: Running pass: GlobalOptPass
diff --git a/llvm/test/Other/new-pm-thinlto-postlink-defaults.ll b/llvm/test/Other/new-pm-thinlto-postlink-defaults.ll
index 064362eabbf83..40468e917bc40 100644
--- a/llvm/test/Other/new-pm-thinlto-postlink-defaults.ll
+++ b/llvm/test/Other/new-pm-thinlto-postlink-defaults.ll
@@ -1,4 +1,4 @@
-; The IR below was crafted so as:
+ ; The IR below was crafted so as:
; 1) To have a loop, so we create a loop pass manager
; 2) To be "immutable" in the sense that no pass in the standard
; pipeline will modify it.
@@ -156,6 +156,7 @@
; CHECK-O-NEXT: Running pass: InvalidateAnalysisPass<{{.*}}ShouldNotRunFunctionPassesAnalysis
; CHECK-O-NEXT: Invalidating analysis: ShouldNotRunFunctionPassesAnalysis
; CHECK-O-NEXT: Invalidating analysis: InlineAdvisorAnalysis
+; CHECK-O-NEXT: Running pass: IPSCCPPass
; CHECK-O-NEXT: Running pass: DeadArgumentEliminationPass
; CHECK-O-NEXT: Running pass: CoroCleanupPass
; CHECK-POSTLINK-O-NEXT: Running pass: GlobalOptPass
diff --git a/llvm/test/Other/new-pm-thinlto-postlink-pgo-defaults.ll b/llvm/test/Other/new-pm-thinlto-postlink-pgo-defaults.ll
index 19a44867e434a..b4b63d8cb3d6f 100644
--- a/llvm/test/Other/new-pm-thinlto-postlink-pgo-defaults.ll
+++ b/llvm/test/Other/new-pm-thinlto-postlink-pgo-defaults.ll
@@ -140,6 +140,7 @@
; CHECK-O-NEXT: Running pass: InvalidateAnalysisPass<{{.*}}ShouldNotRunFunctionPassesAnalysis
; CHECK-O-NEXT: Invalidating analysis: ShouldNotRunFunctionPassesAnalysis
; CHECK-O-NEXT: Invalidating analysis: InlineAdvisorAnalysis
+; CHECK-O-NEXT: Running pass: IPSCCPPass
; CHECK-O-NEXT: Running pass: DeadArgumentEliminationPass
; CHECK-O-NEXT: Running pass: CoroCleanupPass
; CHECK-O-NEXT: Running pass: GlobalOptPass
diff --git a/llvm/test/Other/new-pm-thinlto-postlink-samplepgo-defaults.ll b/llvm/test/Other/new-pm-thinlto-postlink-samplepgo-defaults.ll
index ac80a31d8fd4b..72753a169d288 100644
--- a/llvm/test/Other/new-pm-thinlto-postlink-samplepgo-defaults.ll
+++ b/llvm/test/Other/new-pm-thinlto-postlink-samplepgo-defaults.ll
@@ -148,6 +148,7 @@
; CHECK-O-NEXT: Running pass: InvalidateAnalysisPass<{{.*}}ShouldNotRunFunctionPassesAnalysis
; CHECK-O-NEXT: Invalidating analysis: ShouldNotRunFunctionPassesAnalysis
; CHECK-O-NEXT: Invalidating analysis: InlineAdvisorAnalysis
+; CHECK-O-NEXT: Running pass: IPSCCPPass
; CHECK-O-NEXT: Running pass: DeadArgumentEliminationPass
; CHECK-O-NEXT: Running pass: CoroCleanupPass
; CHECK-O-NEXT: Running pass: GlobalOptPass
diff --git a/llvm/test/Other/new-pm-thinlto-prelink-defaults.ll b/llvm/test/Other/new-pm-thinlto-prelink-defaults.ll
index 42ef49f8f7c7e..896feeb801f5f 100644
--- a/llvm/test/Other/new-pm-thinlto-prelink-defaults.ll
+++ b/llvm/test/Other/new-pm-thinlto-prelink-defaults.ll
@@ -188,6 +188,7 @@
; CHECK-O-NEXT: Running pass: InvalidateAnalysisPass<{{.*}}ShouldNotRunFunctionPassesAnalysis
; CHECK-O-NEXT: Invalidating analysis: ShouldNotRunFunctionPassesAnalysis
; CHECK-O-NEXT: Invalidating analysis: InlineAdvisorAnalysis
+; CHECK-O-NEXT: Running pass: IPSCCPPass
; CHECK-O-NEXT: Running pass: DeadArgumentEliminationPass
; CHECK-O-NEXT: Running pass: CoroCleanupPass
; CHECK-O-NEXT: Running pass: GlobalOptPass
diff --git a/llvm/test/Other/new-pm-thinlto-prelink-pgo-defaults.ll b/llvm/test/Other/new-pm-thinlto-prelink-pgo-defaults.ll
index e74f88c1a3bf9..687e9d08ee31d 100644
--- a/llvm/test/Other/new-pm-thinlto-prelink-pgo-defaults.ll
+++ b/llvm/test/Other/new-pm-thinlto-prelink-pgo-defaults.ll
@@ -187,6 +187,7 @@
; CHECK-O-NEXT: Running pass: InvalidateAnalysisPass<{{.*}}ShouldNotRunFunctionPassesAnalysis
; CHECK-O-NEXT: Invalidating analysis: ShouldNotRunFunctionPassesAnalysis
; CHECK-O-NEXT: Invalidating analysis: InlineAdvisorAnalysis
+; CHECK-O-NEXT: Running pass: IPSCCPPass
; CHECK-O-NEXT: Running pass: DeadArgumentEliminationPass
; CHECK-O-NEXT: Running pass: CoroCleanupPass
; CHECK-O-NEXT: Running pass: GlobalOptPass
diff --git a/llvm/test/Other/new-pm-thinlto-prelink-samplepgo-defaults.ll b/llvm/test/Other/new-pm-thinlto-prelink-samplepgo-defaults.ll
index 210a4ef1f7664..0f87f3a5b1615 100644
--- a/llvm/test/Other/new-pm-thinlto-prelink-samplepgo-defaults.ll
+++ b/llvm/test/Other/new-pm-thinlto-prelink-samplepgo-defaults.ll
@@ -152,6 +152,7 @@
; CHECK-O-NEXT: Running pass: InvalidateAnalysisPass<{{.*}}ShouldNotRunFunctionPassesAnalysis
; CHECK-O-NEXT: Invalidating analysis: ShouldNotRunFunctionPassesAnalysis
; CHECK-O-NEXT: Invalidating analysis: InlineAdvisorAnalysis
+; CHECK-O-NEXT: Running pass: IPSCCPPass
; CHECK-O-NEXT: Running pass: DeadArgumentEliminationPass
; CHECK-O-NEXT: Running pass: CoroCleanupPass
; CHECK-O-NEXT: Running pass: GlobalOptPass
diff --git a/llvm/test/Transforms/PhaseOrdering/AArch64/constraint-elimination-placement.ll b/llvm/test/Transforms/PhaseOrdering/AArch64/constraint-elimination-placement.ll
index 335a850f1ec68..bd29141786ebb 100644
--- a/llvm/test/Transforms/PhaseOrdering/AArch64/constraint-elimination-placement.ll
+++ b/llvm/test/Transforms/PhaseOrdering/AArch64/constraint-elimination-placement.ll
@@ -105,29 +105,20 @@ define void @test2(ptr %this) #0 {
; CHECK-NEXT: i64 17, label [[IF_END_I31:%.*]]
; CHECK-NEXT: ]
; CHECK: if.end.i:
-; CHECK-NEXT: [[CALL8_I_I:%.*]] = tail call fastcc noundef i32 @test2_fn6()
-; CHECK-NEXT: [[TRUNC_I_I:%.*]] = trunc i32 [[CALL8_I_I]] to i8
-; CHECK-NEXT: [[CALL1_I1_I:%.*]] = tail call i1 @test2_fn4(i8 [[TRUNC_I_I]])
+; CHECK-NEXT: [[CALL1_I1_I:%.*]] = tail call i1 @test2_fn4(i8 0)
; CHECK-NEXT: [[TMP0:%.*]] = xor i1 [[CALL1_I1_I]], true
; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP0]])
; CHECK-NEXT: br label [[COMMON_RET]]
; CHECK: test2_fn2.exit12:
-; CHECK-NEXT: [[CALL8_I_I8:%.*]] = tail call fastcc noundef i32 @test2_fn6()
-; CHECK-NEXT: [[TRUNC_I_I9:%.*]] = trunc i32 [[CALL8_I_I8]] to i8
-; CHECK-NEXT: [[CALL1_I1_I10:%.*]] = tail call i1 @test2_fn4(i8 [[TRUNC_I_I9]])
+; CHECK-NEXT: [[CALL1_I1_I10:%.*]] = tail call i1 @test2_fn4(i8 0)
; CHECK-NEXT: [[TMP1:%.*]] = xor i1 [[CALL1_I1_I10]], true
; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP1]])
-; CHECK-NEXT: [[CMP4_I11:%.*]] = icmp eq i32 [[CALL8_I_I8]], 0
-; CHECK-NEXT: br i1 [[CMP4_I11]], label [[TEST2_FN2_EXIT24:%.*]], label [[COMMON_RET]]
-; CHECK: common.ret:
-; CHECK-NEXT: ret void
-; CHECK: test2_fn2.exit24:
; CHECK-NEXT: store i8 0, ptr [[THIS]], align 4
; CHECK-NEXT: br label [[COMMON_RET]]
+; CHECK: common.ret:
+; CHECK-NEXT: ret void
; CHECK: if.end.i31:
-; CHECK-NEXT: [[CALL8_I_I32:%.*]] = tail call fastcc noundef i32 @test2_fn6()
-; CHECK-NEXT: [[TRUNC_I_I33:%.*]] = trunc i32 [[CALL8_I_I32]] to i8
-; CHECK-NEXT: [[CALL1_I1_I34:%.*]] = tail call i1 @test2_fn4(i8 [[TRUNC_I_I33]])
+; CHECK-NEXT: [[CALL1_I1_I34:%.*]] = tail call i1 @test2_fn4(i8 0)
; CHECK-NEXT: [[TMP2:%.*]] = xor i1 [[CALL1_I1_I34]], true
; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP2]])
; CHECK-NEXT: br label [[COMMON_RET]]
@@ -162,16 +153,12 @@ define i1 @test2_fn2(ptr %__rhs) #0 {
; CHECK-NEXT: [[CMP2_NOT:%.*]] = icmp eq i64 [[CALL]], [[COND_I]]
; CHECK-NEXT: br i1 [[CMP2_NOT]], label [[IF_END:%.*]], label [[CLEANUP:%.*]]
; CHECK: if.end:
-; CHECK-NEXT: [[CALL8_I:%.*]] = tail call fastcc noundef i32 @test2_fn6()
-; CHECK-NEXT: [[TRUNC_I:%.*]] = trunc i32 [[CALL8_I]] to i8
-; CHECK-NEXT: [[CALL1_I1:%.*]] = tail call i1 @test2_fn4(i8 [[TRUNC_I]])
+; CHECK-NEXT: [[CALL1_I1:%.*]] = tail call i1 @test2_fn4(i8 0)
; CHECK-NEXT: [[TMP0:%.*]] = xor i1 [[CALL1_I1]], true
; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP0]])
-; CHECK-NEXT: [[CMP4:%.*]] = icmp eq i32 [[CALL8_I]], 0
; CHECK-NEXT: br label [[CLEANUP]]
; CHECK: cleanup:
-; CHECK-NEXT: [[RETVAL_0:%.*]] = phi i1 [ [[CMP4]], [[IF_END]] ], [ false, [[ENTRY:%.*]] ]
-; CHECK-NEXT: ret i1 [[RETVAL_0]]
+; CHECK-NEXT: ret i1 [[CMP2_NOT]]
;
entry:
%call = call i64 @strlen(ptr %__rhs)
@@ -231,11 +218,6 @@ entry:
}
define internal i32 @test2_fn6() {
-; CHECK-LABEL: define internal fastcc noundef i32 @test2_fn6(
-; CHECK-SAME: ) unnamed_addr #[[ATTR5]] {
-; CHECK-NEXT: entry:
-; CHECK-NEXT: ret i32 0
-;
entry:
%call = call i32 @memcmp(ptr @.str.79, ptr @.str.79, i64 2)
ret i32 %call
diff --git a/llvm/test/Transforms/PhaseOrdering/dce-after-argument-promotion.ll b/llvm/test/Transforms/PhaseOrdering/dce-after-argument-promotion.ll
index c33fcfbe6ed97..cc144a700d63a 100644
--- a/llvm/test/Transforms/PhaseOrdering/dce-after-argument-promotion.ll
+++ b/llvm/test/Transforms/PhaseOrdering/dce-after-argument-promotion.ll
@@ -9,10 +9,9 @@
define internal void @f(ptr byval(%struct.ss) align 8 %b, ptr byval(i32) align 4 %X) noinline nounwind {
; CHECK-LABEL: define {{[^@]+}}@f
-; CHECK-SAME: (i32 [[B_0:%.*]]){{[^#]*}} #[[ATTR0:[0-9]+]] {
+; CHECK-SAME: (){{[^#]*}} #[[ATTR0:[0-9]+]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[TEMP:%.*]] = add i32 [[B_0]], 1
-; CHECK-NEXT: store i32 [[TEMP]], ptr [[DUMMY]], align 4
+; CHECK-NEXT: store i32 2, ptr [[DUMMY]], align 4
; CHECK-NEXT: ret void
;
entry:
@@ -27,7 +26,7 @@ define i32 @test(ptr %X) {
; CHECK-LABEL: define {{[^@]+}}@test
; CHECK-SAME: (ptr {{[^%]*}} [[X:%.*]]){{[^#]*}} #[[ATTR1:[0-9]+]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: tail call {{.*}}void @f(i32 1)
+; CHECK-NEXT: tail call {{.*}}void @f()
; CHECK-NEXT: ret i32 0
;
entry:
diff --git a/llvm/test/Transforms/PhaseOrdering/deletion-of-loops-that-became-side-effect-free.ll b/llvm/test/Transforms/PhaseOrdering/deletion-of-loops-that-became-side-effect-free.ll
index 689f4a9798a75..09c7b1cebef8e 100644
--- a/llvm/test/Transforms/PhaseOrdering/deletion-of-loops-that-became-side-effect-free.ll
+++ b/llvm/test/Transforms/PhaseOrdering/deletion-of-loops-that-became-side-effect-free.ll
@@ -1,7 +1,7 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
-; RUN: opt -passes='default<O3>' -S < %s | FileCheck %s --check-prefixes=ALL,O3
-; RUN: opt -passes='default<O2>' -S < %s | FileCheck %s --check-prefixes=ALL,O2
-; RUN: opt -passes='default<O1>' -S < %s | FileCheck %s --check-prefixes=ALL,O1
+; RUN: opt -passes='default<O3>' -S < %s | FileCheck %s
+; RUN: opt -passes='default<O2>' -S < %s | FileCheck %s
+; RUN: opt -passes='default<O1>' -S < %s | FileCheck %s
; All these tests should optimize to a single comparison
; of the original argument with null. There should be no loops.
@@ -9,10 +9,10 @@
%struct.node = type { ptr, i32 }
define dso_local zeroext i1 @is_not_empty_variant1(ptr %p) {
-; ALL-LABEL: @is_not_empty_variant1(
-; ALL-NEXT: entry:
-; ALL-NEXT: [[TOBOOL_NOT3_I:%.*]] = icmp ne ptr [[P:%.*]], null
-; ALL-NEXT: ret i1 [[TOBOOL_NOT3_I]]
+; CHECK-LABEL: @is_not_empty_variant1(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[TOBOOL_NOT3_I:%.*]] = icmp ne ptr [[P:%.*]], null
+; CHECK-NEXT: ret i1 [[TOBOOL_NOT3_I]]
;
entry:
%p.addr = alloca ptr, align 8
@@ -51,10 +51,10 @@ while.end:
}
define dso_local zeroext i1 @is_not_empty_variant2(ptr %p) {
-; ALL-LABEL: @is_not_empty_variant2(
-; ALL-NEXT: entry:
-; ALL-NEXT: [[TOBOOL_NOT4_I:%.*]] = icmp ne ptr [[P:%.*]], null
-; ALL-NEXT: ret i1 [[TOBOOL_NOT4_I]]
+; CHECK-LABEL: @is_not_empty_variant2(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[TOBOOL_NOT4_I:%.*]] = icmp ne ptr [[P:%.*]], null
+; CHECK-NEXT: ret i1 [[TOBOOL_NOT4_I]]
;
entry:
%p.addr = alloca ptr, align 8
@@ -96,35 +96,10 @@ while.end:
}
define dso_local zeroext i1 @is_not_empty_variant3(ptr %p) {
-; O3-LABEL: @is_not_empty_variant3(
-; O3-NEXT: entry:
-; O3-NEXT: [[TOBOOL_NOT4_I:%.*]] = icmp ne ptr [[P:%.*]], null
-; O3-NEXT: ret i1 [[TOBOOL_NOT4_I]]
-;
-; O2-LABEL: @is_not_empty_variant3(
-; O2-NEXT: entry:
-; O2-NEXT: [[TOBOOL_NOT4_I:%.*]] = icmp ne ptr [[P:%.*]], null
-; O2-NEXT: ret i1 [[TOBOOL_NOT4_I]]
-;
-; O1-LABEL: @is_not_empty_variant3(
-; O1-NEXT: entry:
-; O1-NEXT: [[TOBOOL_NOT4_I:%.*]] = icmp eq ptr [[P:%.*]], null
-; O1-NEXT: br i1 [[TOBOOL_NOT4_I]], label [[COUNT_NODES_VARIANT3_EXIT:%.*]], label [[WHILE_BODY_I:%.*]]
-; O1: while.body.i:
-; O1-NEXT: [[SIZE_06_I:%.*]] = phi i64 [ [[INC_I:%.*]], [[WHILE_BODY_I]] ], [ 0, [[ENTRY:%.*]] ]
-; O1-NEXT: [[P_ADDR_05_I:%.*]] = phi ptr [ [[TMP0:%.*]], [[WHILE_BODY_I]] ], [ [[P]], [[ENTRY]] ]
-; O1-NEXT: [[CMP_I:%.*]] = icmp ne i64 [[SIZE_06_I]], -1
-; O1-NEXT: call void @llvm.assume(i1 [[CMP_I]])
-; O1-NEXT: [[TMP0]] = load ptr, ptr [[P_ADDR_05_I]], align 8
-; O1-NEXT: [[INC_I]] = add i64 [[SIZE_06_I]], 1
-; O1-NEXT: [[TOBOOL_NOT_I:%.*]] = icmp eq ptr [[TMP0]], null
-; O1-NEXT: br i1 [[TOBOOL_NOT_I]], label [[COUNT_NODES_VARIANT3_EXIT_LOOPEXIT:%.*]], label [[WHILE_BODY_I]], !llvm.loop [[LOOP0:![0-9]+]]
-; O1: count_nodes_variant3.exit.loopexit:
-; O1-NEXT: [[PHI_CMP:%.*]] = icmp ne i64 [[INC_I]], 0
-; O1-NEXT: br label [[COUNT_NODES_VARIANT3_EXIT]]
-; O1: count_nodes_variant3.exit:
-; O1-NEXT: [[SIZE_0_LCSSA_I:%.*]] = phi i1 [ false, [[ENTRY]] ], [ [[PHI_CMP]], [[COUNT_NODES_VARIANT3_EXIT_LOOPEXIT]] ]
-; O1-NEXT: ret i1 [[SIZE_0_LCSSA_I]]
+; CHECK-LABEL: @is_not_empty_variant3(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[TOBOOL_NOT4_I:%.*]] = icmp ne ptr [[P:%.*]], null
+; CHECK-NEXT: ret i1 [[TOBOOL_NOT4_I]]
;
entry:
%p.addr = alloca ptr, align 8
|
It has been fixed. But this patch didn't show a positive net effect :( |
Does that mean it has a negative net effect, or its neutral (in which case the original motivating case should be enough). |
There was a problem hiding this comment.
Running IPSCCP twice seems like quite a heavy hammer, I'd expect a noticeable compile-time impact.
I'd recommend to try to extract a reproducer from your motivating use case and check why IPSCCP cannot perform the desired optimization before inlining. Note that we run SCCP after inlining I think, which is the non-IP version of IPSCCP
It is neutral.
But this patch may have an impact on compile time. |
|
Yes, this causes a massive compile-time regression: http://llvm-compile-time-tracker.com/compare.php?from=253a294b54a6096a0b66f840931dd0e345d70c4f&to=20869e8ed55c58696ce4e9dc3d18cd6af01ed49c&stat=instructions:u |
|
@sihuan Are you testing your patch on LTO? Before applying your patch, the function is split into two patches: With your patch, it exhibits the same behavior. Do you know how I can get the same behavior as gcc where digits_2 is split into eight parts? Are you testing with other options? |
|
@joshua-arch1
When using LTO, I don't think there is a problem that performance is worse than GCC, so this patch is not specific to LTO. |
Hi @sihuan, I have tried to apply yout initial design where IPSCCP is directly moved after inliner. It can work with LTO and |
This patch significantly improves the performance of LLVM for the SPEC2017:548.exchange2_r benchmark, with a performance uplift of over 40% on the rv64gc.
During our investigation into the significant performance disparity between GCC and LLVM on the SPEC2017:548.exchange2_r benchmark on RISC-V, we identified that the primary difference stems from constant propagation optimization.
In GCC, the hotspot function
digits_2is split into several parts:However, in LLVM, this function is not split:
By applying this patch, LLVM now exhibits similar behavior, resulting in a substantial performance uplift.
And we used
perf statto measure the instruction count forexchange2_r 0on rv64gc, as shown in the table below:Additionally, I performed tests on x86_64, yielding similar results: