[llvm] 8441a8e - [MergeICmps] Add test for call before first load (NFC)
Nikita Popov via llvm-commits
llvm-commits at lists.llvm.org
Thu Aug 26 12:24:34 PDT 2021
Author: Nikita Popov
Date: 2021-08-26T21:24:22+02:00
New Revision: 8441a8eea8007b9eaaaabf76055949180a702d6d
URL: https://github.com/llvm/llvm-project/commit/8441a8eea8007b9eaaaabf76055949180a702d6d
DIFF: https://github.com/llvm/llvm-project/commit/8441a8eea8007b9eaaaabf76055949180a702d6d.diff
LOG: [MergeICmps] Add test for call before first load (NFC)
If a clobbering call happens before all loads, that shouldn't
block the transform.
Added:
Modified:
llvm/test/Transforms/MergeICmps/X86/split-block-does-work.ll
Removed:
################################################################################
diff --git a/llvm/test/Transforms/MergeICmps/X86/split-block-does-work.ll b/llvm/test/Transforms/MergeICmps/X86/split-block-does-work.ll
index 1e341b92918d9..2fa8d9f709cce 100644
--- a/llvm/test/Transforms/MergeICmps/X86/split-block-does-work.ll
+++ b/llvm/test/Transforms/MergeICmps/X86/split-block-does-work.ll
@@ -3,13 +3,13 @@
%S = type { i32, i32, i32, i32 }
-declare void @foo(...) inaccessiblememonly
+declare void @foo(...)
; We can split %entry and create a memcmp(16 bytes).
define zeroext i1 @opeq1(
; X86-LABEL: @opeq1(
; X86-NEXT: "entry+land.rhs.i+land.rhs.i.2+land.rhs.i.3":
-; X86-NEXT: call void (...) @foo()
+; X86-NEXT: call void (...) @foo() #[[ATTR2:[0-9]+]]
; X86-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[S:%.*]], %S* [[A:%.*]], i64 0, i32 0
; X86-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[S]], %S* [[B:%.*]], i64 0, i32 0
; X86-NEXT: [[CSTR:%.*]] = bitcast i32* [[TMP0]] to i8*
@@ -29,7 +29,7 @@ entry:
%first1.i = getelementptr inbounds %S, %S* %b, i64 0, i32 0
%1 = load i32, i32* %first1.i, align 4
; Does other work.
- call void (...) @foo()
+ call void (...) @foo() inaccessiblememonly
%cmp.i = icmp eq i32 %0, %1
br i1 %cmp.i, label %land.rhs.i, label %opeq1.exit
@@ -71,7 +71,7 @@ define zeroext i1 @opeq1_discontiguous(
; X86-NEXT: [[TMP0:%.*]] = load i32, i32* [[FIRST_I]], align 4
; X86-NEXT: [[FIRST1_I:%.*]] = getelementptr inbounds [[S]], %S* [[B:%.*]], i64 0, i32 0
; X86-NEXT: [[TMP1:%.*]] = load i32, i32* [[FIRST1_I]], align 4
-; X86-NEXT: call void (...) @foo()
+; X86-NEXT: call void (...) @foo() #[[ATTR2]]
; X86-NEXT: [[CMP_I:%.*]] = icmp eq i32 [[TMP0]], [[TMP1]]
; X86-NEXT: br i1 [[CMP_I]], label [[LAND_RHS_I:%.*]], label [[OPEQ1_EXIT:%.*]]
; X86: land.rhs.i:
@@ -108,7 +108,7 @@ entry:
%first1.i = getelementptr inbounds %S, %S* %b, i64 0, i32 0
%1 = load i32, i32* %first1.i, align 4
; Does other work.
- call void (...) @foo()
+ call void (...) @foo() inaccessiblememonly
%cmp.i = icmp eq i32 %0, %1
br i1 %cmp.i, label %land.rhs.i, label %opeq1.exit
@@ -140,3 +140,131 @@ opeq1.exit:
%8 = phi i1 [ false, %entry ], [ false, %land.rhs.i] , [ false, %land.rhs.i.2 ], [ %cmp4.i, %land.rhs.i.3 ]
ret i1 %8
}
+
+; TODO: The call happen before the loads, so it cannot clobber them.
+define zeroext i1 @opeq1_call_before_loads(
+; X86-LABEL: @opeq1_call_before_loads(
+; X86-NEXT: entry:
+; X86-NEXT: call void (...) @foo()
+; X86-NEXT: [[FIRST_I:%.*]] = getelementptr inbounds [[S:%.*]], %S* [[A:%.*]], i64 0, i32 0
+; X86-NEXT: [[TMP0:%.*]] = load i32, i32* [[FIRST_I]], align 4
+; X86-NEXT: [[FIRST1_I:%.*]] = getelementptr inbounds [[S]], %S* [[B:%.*]], i64 0, i32 0
+; X86-NEXT: [[TMP1:%.*]] = load i32, i32* [[FIRST1_I]], align 4
+; X86-NEXT: [[CMP_I:%.*]] = icmp eq i32 [[TMP0]], [[TMP1]]
+; X86-NEXT: br i1 [[CMP_I]], label %"land.rhs.i+land.rhs.i.2+land.rhs.i.3", label [[OPEQ1_EXIT:%.*]]
+; X86: "land.rhs.i+land.rhs.i.2+land.rhs.i.3":
+; X86-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[S]], %S* [[A]], i64 0, i32 1
+; X86-NEXT: [[TMP3:%.*]] = getelementptr inbounds [[S]], %S* [[B]], i64 0, i32 1
+; X86-NEXT: [[CSTR:%.*]] = bitcast i32* [[TMP2]] to i8*
+; X86-NEXT: [[CSTR1:%.*]] = bitcast i32* [[TMP3]] to i8*
+; X86-NEXT: [[MEMCMP:%.*]] = call i32 @memcmp(i8* [[CSTR]], i8* [[CSTR1]], i64 12)
+; X86-NEXT: [[TMP4:%.*]] = icmp eq i32 [[MEMCMP]], 0
+; X86-NEXT: br label [[OPEQ1_EXIT]]
+; X86: opeq1.exit:
+; X86-NEXT: [[TMP5:%.*]] = phi i1 [ false, [[ENTRY:%.*]] ], [ [[TMP4]], %"land.rhs.i+land.rhs.i.2+land.rhs.i.3" ]
+; X86-NEXT: ret i1 [[TMP5]]
+;
+; Make sure this call is moved to the beginning of the entry block.
+ %S* nocapture readonly dereferenceable(16) %a,
+ %S* nocapture readonly dereferenceable(16) %b) local_unnamed_addr nofree nosync {
+entry:
+ call void (...) @foo()
+ %first.i = getelementptr inbounds %S, %S* %a, i64 0, i32 0
+ %0 = load i32, i32* %first.i, align 4
+ %first1.i = getelementptr inbounds %S, %S* %b, i64 0, i32 0
+ %1 = load i32, i32* %first1.i, align 4
+ %cmp.i = icmp eq i32 %0, %1
+ br i1 %cmp.i, label %land.rhs.i, label %opeq1.exit
+
+land.rhs.i:
+ %second.i = getelementptr inbounds %S, %S* %a, i64 0, i32 1
+ %2 = load i32, i32* %second.i, align 4
+ %second2.i = getelementptr inbounds %S, %S* %b, i64 0, i32 1
+ %3 = load i32, i32* %second2.i, align 4
+ %cmp2.i = icmp eq i32 %2, %3
+ br i1 %cmp2.i, label %land.rhs.i.2, label %opeq1.exit
+
+land.rhs.i.2:
+ %third.i = getelementptr inbounds %S, %S* %a, i64 0, i32 2
+ %4 = load i32, i32* %third.i, align 4
+ %third2.i = getelementptr inbounds %S, %S* %b, i64 0, i32 2
+ %5 = load i32, i32* %third2.i, align 4
+ %cmp3.i = icmp eq i32 %4, %5
+ br i1 %cmp3.i, label %land.rhs.i.3, label %opeq1.exit
+
+land.rhs.i.3:
+ %fourth.i = getelementptr inbounds %S, %S* %a, i64 0, i32 3
+ %6 = load i32, i32* %fourth.i, align 4
+ %fourth2.i = getelementptr inbounds %S, %S* %b, i64 0, i32 3
+ %7 = load i32, i32* %fourth2.i, align 4
+ %cmp4.i = icmp eq i32 %6, %7
+ br label %opeq1.exit
+
+opeq1.exit:
+ %8 = phi i1 [ false, %entry ], [ false, %land.rhs.i] , [ false, %land.rhs.i.2 ], [ %cmp4.i, %land.rhs.i.3 ]
+ ret i1 %8
+}
+
+; Call happens after the loads, and may clobber them.
+define zeroext i1 @opeq1_call_after_loads(
+; X86-LABEL: @opeq1_call_after_loads(
+; X86-NEXT: entry:
+; X86-NEXT: [[FIRST_I:%.*]] = getelementptr inbounds [[S:%.*]], %S* [[A:%.*]], i64 0, i32 0
+; X86-NEXT: [[TMP0:%.*]] = load i32, i32* [[FIRST_I]], align 4
+; X86-NEXT: [[FIRST1_I:%.*]] = getelementptr inbounds [[S]], %S* [[B:%.*]], i64 0, i32 0
+; X86-NEXT: [[TMP1:%.*]] = load i32, i32* [[FIRST1_I]], align 4
+; X86-NEXT: call void (...) @foo()
+; X86-NEXT: [[CMP_I:%.*]] = icmp eq i32 [[TMP0]], [[TMP1]]
+; X86-NEXT: br i1 [[CMP_I]], label %"land.rhs.i+land.rhs.i.2+land.rhs.i.3", label [[OPEQ1_EXIT:%.*]]
+; X86: "land.rhs.i+land.rhs.i.2+land.rhs.i.3":
+; X86-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[S]], %S* [[A]], i64 0, i32 1
+; X86-NEXT: [[TMP3:%.*]] = getelementptr inbounds [[S]], %S* [[B]], i64 0, i32 1
+; X86-NEXT: [[CSTR:%.*]] = bitcast i32* [[TMP2]] to i8*
+; X86-NEXT: [[CSTR1:%.*]] = bitcast i32* [[TMP3]] to i8*
+; X86-NEXT: [[MEMCMP:%.*]] = call i32 @memcmp(i8* [[CSTR]], i8* [[CSTR1]], i64 12)
+; X86-NEXT: [[TMP4:%.*]] = icmp eq i32 [[MEMCMP]], 0
+; X86-NEXT: br label [[OPEQ1_EXIT]]
+; X86: opeq1.exit:
+; X86-NEXT: [[TMP5:%.*]] = phi i1 [ false, [[ENTRY:%.*]] ], [ [[TMP4]], %"land.rhs.i+land.rhs.i.2+land.rhs.i.3" ]
+; X86-NEXT: ret i1 [[TMP5]]
+;
+; Make sure this call is moved to the beginning of the entry block.
+ %S* nocapture readonly dereferenceable(16) %a,
+ %S* nocapture readonly dereferenceable(16) %b) local_unnamed_addr nofree nosync {
+entry:
+ %first.i = getelementptr inbounds %S, %S* %a, i64 0, i32 0
+ %0 = load i32, i32* %first.i, align 4
+ %first1.i = getelementptr inbounds %S, %S* %b, i64 0, i32 0
+ %1 = load i32, i32* %first1.i, align 4
+ call void (...) @foo()
+ %cmp.i = icmp eq i32 %0, %1
+ br i1 %cmp.i, label %land.rhs.i, label %opeq1.exit
+
+land.rhs.i:
+ %second.i = getelementptr inbounds %S, %S* %a, i64 0, i32 1
+ %2 = load i32, i32* %second.i, align 4
+ %second2.i = getelementptr inbounds %S, %S* %b, i64 0, i32 1
+ %3 = load i32, i32* %second2.i, align 4
+ %cmp2.i = icmp eq i32 %2, %3
+ br i1 %cmp2.i, label %land.rhs.i.2, label %opeq1.exit
+
+land.rhs.i.2:
+ %third.i = getelementptr inbounds %S, %S* %a, i64 0, i32 2
+ %4 = load i32, i32* %third.i, align 4
+ %third2.i = getelementptr inbounds %S, %S* %b, i64 0, i32 2
+ %5 = load i32, i32* %third2.i, align 4
+ %cmp3.i = icmp eq i32 %4, %5
+ br i1 %cmp3.i, label %land.rhs.i.3, label %opeq1.exit
+
+land.rhs.i.3:
+ %fourth.i = getelementptr inbounds %S, %S* %a, i64 0, i32 3
+ %6 = load i32, i32* %fourth.i, align 4
+ %fourth2.i = getelementptr inbounds %S, %S* %b, i64 0, i32 3
+ %7 = load i32, i32* %fourth2.i, align 4
+ %cmp4.i = icmp eq i32 %6, %7
+ br label %opeq1.exit
+
+opeq1.exit:
+ %8 = phi i1 [ false, %entry ], [ false, %land.rhs.i] , [ false, %land.rhs.i.2 ], [ %cmp4.i, %land.rhs.i.3 ]
+ ret i1 %8
+}
More information about the llvm-commits
mailing list