; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5 ; RUN: opt -passes='require,function(codegenprepare)' < %s -mtriple=aarch64-apple-ios -S | FileCheck -enable-var-scope %s --check-prefix=OPTALL --check-prefixes=OPT,NONSTRESS ; RUN: opt -passes='require,function(codegenprepare)' < %s -mtriple=aarch64-apple-ios -S -stress-cgp-ext-ld-promotion | FileCheck -enable-var-scope %s --check-prefixes=OPTALL,OPT,STRESS ; RUN: opt -passes='require,function(codegenprepare)' < %s -mtriple=aarch64-apple-ios -S -disable-cgp-ext-ld-promotion | FileCheck -enable-var-scope %s --check-prefixes=OPTALL,DISABLE ; CodeGenPrepare should move the zext into the block with the load ; so that SelectionDAG can select it with the load. define void @foo(ptr %p, ptr %q) { ; OPTALL-LABEL: define void @foo( ; OPTALL-SAME: ptr [[P:%.*]], ptr [[Q:%.*]]) { ; OPTALL-NEXT: [[ENTRY:.*:]] ; OPTALL-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; OPTALL-NEXT: [[S:%.*]] = zext i8 [[T]] to i32 ; OPTALL-NEXT: [[A:%.*]] = icmp slt i8 [[T]], 20 ; OPTALL-NEXT: br i1 [[A]], label %[[TRUE:.*]], label %[[FALSE:.*]] ; OPTALL: [[TRUE]]: ; OPTALL-NEXT: store i32 [[S]], ptr [[Q]], align 4 ; OPTALL-NEXT: ret void ; OPTALL: [[FALSE]]: ; OPTALL-NEXT: ret void ; entry: %t = load i8, ptr %p %a = icmp slt i8 %t, 20 br i1 %a, label %true, label %false true: %s = zext i8 %t to i32 store i32 %s, ptr %q ret void false: ret void } ; Check that we manage to form a zextload is an operation with only one ; argument to explicitly extend is in the way. ; Make sure the operation is not promoted when the promotion pass is disabled. define void @promoteOneArg(ptr %p, ptr %q) { ; OPT-LABEL: define void @promoteOneArg( ; OPT-SAME: ptr [[P:%.*]], ptr [[Q:%.*]]) { ; OPT-NEXT: [[ENTRY:.*:]] ; OPT-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; OPT-NEXT: [[PROMOTED:%.*]] = zext i8 [[T]] to i32 ; OPT-NEXT: [[ADD:%.*]] = add nuw i32 [[PROMOTED]], 2 ; OPT-NEXT: [[A:%.*]] = icmp slt i8 [[T]], 20 ; OPT-NEXT: br i1 [[A]], label %[[TRUE:.*]], label %[[FALSE:.*]] ; OPT: [[TRUE]]: ; OPT-NEXT: store i32 [[ADD]], ptr [[Q]], align 4 ; OPT-NEXT: ret void ; OPT: [[FALSE]]: ; OPT-NEXT: ret void ; ; DISABLE-LABEL: define void @promoteOneArg( ; DISABLE-SAME: ptr [[P:%.*]], ptr [[Q:%.*]]) { ; DISABLE-NEXT: [[ENTRY:.*:]] ; DISABLE-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; DISABLE-NEXT: [[ADD:%.*]] = add nuw i8 [[T]], 2 ; DISABLE-NEXT: [[A:%.*]] = icmp slt i8 [[T]], 20 ; DISABLE-NEXT: br i1 [[A]], label %[[TRUE:.*]], label %[[FALSE:.*]] ; DISABLE: [[TRUE]]: ; DISABLE-NEXT: [[S:%.*]] = zext i8 [[ADD]] to i32 ; DISABLE-NEXT: store i32 [[S]], ptr [[Q]], align 4 ; DISABLE-NEXT: ret void ; DISABLE: [[FALSE]]: ; DISABLE-NEXT: ret void ; entry: %t = load i8, ptr %p %add = add nuw i8 %t, 2 %a = icmp slt i8 %t, 20 br i1 %a, label %true, label %false true: %s = zext i8 %add to i32 store i32 %s, ptr %q ret void false: ret void } ; Check that we manage to form a sextload is an operation with only one ; argument to explicitly extend is in the way. ; Version with sext. define void @promoteOneArgSExt(ptr %p, ptr %q) { ; OPT-LABEL: define void @promoteOneArgSExt( ; OPT-SAME: ptr [[P:%.*]], ptr [[Q:%.*]]) { ; OPT-NEXT: [[ENTRY:.*:]] ; OPT-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; OPT-NEXT: [[PROMOTED:%.*]] = sext i8 [[T]] to i32 ; OPT-NEXT: [[ADD:%.*]] = add nsw i32 [[PROMOTED]], 2 ; OPT-NEXT: [[A:%.*]] = icmp slt i8 [[T]], 20 ; OPT-NEXT: br i1 [[A]], label %[[TRUE:.*]], label %[[FALSE:.*]] ; OPT: [[TRUE]]: ; OPT-NEXT: store i32 [[ADD]], ptr [[Q]], align 4 ; OPT-NEXT: ret void ; OPT: [[FALSE]]: ; OPT-NEXT: ret void ; ; DISABLE-LABEL: define void @promoteOneArgSExt( ; DISABLE-SAME: ptr [[P:%.*]], ptr [[Q:%.*]]) { ; DISABLE-NEXT: [[ENTRY:.*:]] ; DISABLE-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; DISABLE-NEXT: [[ADD:%.*]] = add nsw i8 [[T]], 2 ; DISABLE-NEXT: [[A:%.*]] = icmp slt i8 [[T]], 20 ; DISABLE-NEXT: br i1 [[A]], label %[[TRUE:.*]], label %[[FALSE:.*]] ; DISABLE: [[TRUE]]: ; DISABLE-NEXT: [[S:%.*]] = sext i8 [[ADD]] to i32 ; DISABLE-NEXT: store i32 [[S]], ptr [[Q]], align 4 ; DISABLE-NEXT: ret void ; DISABLE: [[FALSE]]: ; DISABLE-NEXT: ret void ; entry: %t = load i8, ptr %p %add = add nsw i8 %t, 2 %a = icmp slt i8 %t, 20 br i1 %a, label %true, label %false true: %s = sext i8 %add to i32 store i32 %s, ptr %q ret void false: ret void } ; Check that we manage to form a zextload is an operation with two ; arguments to explicitly extend is in the way. ; Extending %add will create two extensions: ; 1. One for %b. ; 2. One for %t. ; #1 will not be removed as we do not know anything about %b. ; #2 may not be merged with the load because %t is used in a comparison. ; Since two extensions may be emitted in the end instead of one before the ; transformation, the regular heuristic does not apply the optimization. define void @promoteTwoArgZext(ptr %p, ptr %q, i8 %b) { ; NONSTRESS-LABEL: define void @promoteTwoArgZext( ; NONSTRESS-SAME: ptr [[P:%.*]], ptr [[Q:%.*]], i8 [[B:%.*]]) { ; NONSTRESS-NEXT: [[ENTRY:.*:]] ; NONSTRESS-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; NONSTRESS-NEXT: [[ADD:%.*]] = add nuw i8 [[T]], [[B]] ; NONSTRESS-NEXT: [[A:%.*]] = icmp slt i8 [[T]], 20 ; NONSTRESS-NEXT: br i1 [[A]], label %[[TRUE:.*]], label %[[FALSE:.*]] ; NONSTRESS: [[TRUE]]: ; NONSTRESS-NEXT: [[S:%.*]] = zext i8 [[ADD]] to i32 ; NONSTRESS-NEXT: store i32 [[S]], ptr [[Q]], align 4 ; NONSTRESS-NEXT: ret void ; NONSTRESS: [[FALSE]]: ; NONSTRESS-NEXT: ret void ; ; STRESS-LABEL: define void @promoteTwoArgZext( ; STRESS-SAME: ptr [[P:%.*]], ptr [[Q:%.*]], i8 [[B:%.*]]) { ; STRESS-NEXT: [[ENTRY:.*:]] ; STRESS-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; STRESS-NEXT: [[PROMOTED:%.*]] = zext i8 [[T]] to i32 ; STRESS-NEXT: [[PROMOTED1:%.*]] = zext i8 [[B]] to i32 ; STRESS-NEXT: [[ADD:%.*]] = add nuw i32 [[PROMOTED]], [[PROMOTED1]] ; STRESS-NEXT: [[A:%.*]] = icmp slt i8 [[T]], 20 ; STRESS-NEXT: br i1 [[A]], label %[[TRUE:.*]], label %[[FALSE:.*]] ; STRESS: [[TRUE]]: ; STRESS-NEXT: store i32 [[ADD]], ptr [[Q]], align 4 ; STRESS-NEXT: ret void ; STRESS: [[FALSE]]: ; STRESS-NEXT: ret void ; ; DISABLE-LABEL: define void @promoteTwoArgZext( ; DISABLE-SAME: ptr [[P:%.*]], ptr [[Q:%.*]], i8 [[B:%.*]]) { ; DISABLE-NEXT: [[ENTRY:.*:]] ; DISABLE-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; DISABLE-NEXT: [[ADD:%.*]] = add nuw i8 [[T]], [[B]] ; DISABLE-NEXT: [[A:%.*]] = icmp slt i8 [[T]], 20 ; DISABLE-NEXT: br i1 [[A]], label %[[TRUE:.*]], label %[[FALSE:.*]] ; DISABLE: [[TRUE]]: ; DISABLE-NEXT: [[S:%.*]] = zext i8 [[ADD]] to i32 ; DISABLE-NEXT: store i32 [[S]], ptr [[Q]], align 4 ; DISABLE-NEXT: ret void ; DISABLE: [[FALSE]]: ; DISABLE-NEXT: ret void ; entry: %t = load i8, ptr %p %add = add nuw i8 %t, %b %a = icmp slt i8 %t, 20 br i1 %a, label %true, label %false true: %s = zext i8 %add to i32 store i32 %s, ptr %q ret void false: ret void } ; Check that we manage to form a sextload is an operation with two ; arguments to explicitly extend is in the way. ; Version with sext. define void @promoteTwoArgSExt(ptr %p, ptr %q, i8 %b) { ; NONSTRESS-LABEL: define void @promoteTwoArgSExt( ; NONSTRESS-SAME: ptr [[P:%.*]], ptr [[Q:%.*]], i8 [[B:%.*]]) { ; NONSTRESS-NEXT: [[ENTRY:.*:]] ; NONSTRESS-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; NONSTRESS-NEXT: [[ADD:%.*]] = add nsw i8 [[T]], [[B]] ; NONSTRESS-NEXT: [[A:%.*]] = icmp slt i8 [[T]], 20 ; NONSTRESS-NEXT: br i1 [[A]], label %[[TRUE:.*]], label %[[FALSE:.*]] ; NONSTRESS: [[TRUE]]: ; NONSTRESS-NEXT: [[S:%.*]] = sext i8 [[ADD]] to i32 ; NONSTRESS-NEXT: store i32 [[S]], ptr [[Q]], align 4 ; NONSTRESS-NEXT: ret void ; NONSTRESS: [[FALSE]]: ; NONSTRESS-NEXT: ret void ; ; STRESS-LABEL: define void @promoteTwoArgSExt( ; STRESS-SAME: ptr [[P:%.*]], ptr [[Q:%.*]], i8 [[B:%.*]]) { ; STRESS-NEXT: [[ENTRY:.*:]] ; STRESS-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; STRESS-NEXT: [[PROMOTED:%.*]] = sext i8 [[T]] to i32 ; STRESS-NEXT: [[PROMOTED1:%.*]] = sext i8 [[B]] to i32 ; STRESS-NEXT: [[ADD:%.*]] = add nsw i32 [[PROMOTED]], [[PROMOTED1]] ; STRESS-NEXT: [[A:%.*]] = icmp slt i8 [[T]], 20 ; STRESS-NEXT: br i1 [[A]], label %[[TRUE:.*]], label %[[FALSE:.*]] ; STRESS: [[TRUE]]: ; STRESS-NEXT: store i32 [[ADD]], ptr [[Q]], align 4 ; STRESS-NEXT: ret void ; STRESS: [[FALSE]]: ; STRESS-NEXT: ret void ; ; DISABLE-LABEL: define void @promoteTwoArgSExt( ; DISABLE-SAME: ptr [[P:%.*]], ptr [[Q:%.*]], i8 [[B:%.*]]) { ; DISABLE-NEXT: [[ENTRY:.*:]] ; DISABLE-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; DISABLE-NEXT: [[ADD:%.*]] = add nsw i8 [[T]], [[B]] ; DISABLE-NEXT: [[A:%.*]] = icmp slt i8 [[T]], 20 ; DISABLE-NEXT: br i1 [[A]], label %[[TRUE:.*]], label %[[FALSE:.*]] ; DISABLE: [[TRUE]]: ; DISABLE-NEXT: [[S:%.*]] = sext i8 [[ADD]] to i32 ; DISABLE-NEXT: store i32 [[S]], ptr [[Q]], align 4 ; DISABLE-NEXT: ret void ; DISABLE: [[FALSE]]: ; DISABLE-NEXT: ret void ; entry: %t = load i8, ptr %p %add = add nsw i8 %t, %b %a = icmp slt i8 %t, 20 br i1 %a, label %true, label %false true: %s = sext i8 %add to i32 store i32 %s, ptr %q ret void false: ret void } ; Check that we do not a zextload if we need to introduce more than ; one additional extension. define void @promoteThreeArgZext(ptr %p, ptr %q, i8 %b, i8 %c) { ; NONSTRESS-LABEL: define void @promoteThreeArgZext( ; NONSTRESS-SAME: ptr [[P:%.*]], ptr [[Q:%.*]], i8 [[B:%.*]], i8 [[C:%.*]]) { ; NONSTRESS-NEXT: [[ENTRY:.*:]] ; NONSTRESS-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; NONSTRESS-NEXT: [[TMP:%.*]] = add nuw i8 [[T]], [[B]] ; NONSTRESS-NEXT: [[ADD:%.*]] = add nuw i8 [[TMP]], [[C]] ; NONSTRESS-NEXT: [[A:%.*]] = icmp slt i8 [[T]], 20 ; NONSTRESS-NEXT: br i1 [[A]], label %[[TRUE:.*]], label %[[FALSE:.*]] ; NONSTRESS: [[TRUE]]: ; NONSTRESS-NEXT: [[S:%.*]] = zext i8 [[ADD]] to i32 ; NONSTRESS-NEXT: store i32 [[S]], ptr [[Q]], align 4 ; NONSTRESS-NEXT: ret void ; NONSTRESS: [[FALSE]]: ; NONSTRESS-NEXT: ret void ; ; STRESS-LABEL: define void @promoteThreeArgZext( ; STRESS-SAME: ptr [[P:%.*]], ptr [[Q:%.*]], i8 [[B:%.*]], i8 [[C:%.*]]) { ; STRESS-NEXT: [[ENTRY:.*:]] ; STRESS-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; STRESS-NEXT: [[PROMOTED2:%.*]] = zext i8 [[T]] to i32 ; STRESS-NEXT: [[PROMOTED3:%.*]] = zext i8 [[B]] to i32 ; STRESS-NEXT: [[TMP:%.*]] = add nuw i32 [[PROMOTED2]], [[PROMOTED3]] ; STRESS-NEXT: [[PROMOTED1:%.*]] = zext i8 [[C]] to i32 ; STRESS-NEXT: [[ADD:%.*]] = add nuw i32 [[TMP]], [[PROMOTED1]] ; STRESS-NEXT: [[A:%.*]] = icmp slt i8 [[T]], 20 ; STRESS-NEXT: br i1 [[A]], label %[[TRUE:.*]], label %[[FALSE:.*]] ; STRESS: [[TRUE]]: ; STRESS-NEXT: store i32 [[ADD]], ptr [[Q]], align 4 ; STRESS-NEXT: ret void ; STRESS: [[FALSE]]: ; STRESS-NEXT: ret void ; ; DISABLE-LABEL: define void @promoteThreeArgZext( ; DISABLE-SAME: ptr [[P:%.*]], ptr [[Q:%.*]], i8 [[B:%.*]], i8 [[C:%.*]]) { ; DISABLE-NEXT: [[ENTRY:.*:]] ; DISABLE-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; DISABLE-NEXT: [[TMP:%.*]] = add nuw i8 [[T]], [[B]] ; DISABLE-NEXT: [[ADD:%.*]] = add nuw i8 [[TMP]], [[C]] ; DISABLE-NEXT: [[A:%.*]] = icmp slt i8 [[T]], 20 ; DISABLE-NEXT: br i1 [[A]], label %[[TRUE:.*]], label %[[FALSE:.*]] ; DISABLE: [[TRUE]]: ; DISABLE-NEXT: [[S:%.*]] = zext i8 [[ADD]] to i32 ; DISABLE-NEXT: store i32 [[S]], ptr [[Q]], align 4 ; DISABLE-NEXT: ret void ; DISABLE: [[FALSE]]: ; DISABLE-NEXT: ret void ; entry: %t = load i8, ptr %p %tmp = add nuw i8 %t, %b %add = add nuw i8 %tmp, %c %a = icmp slt i8 %t, 20 br i1 %a, label %true, label %false true: %s = zext i8 %add to i32 store i32 %s, ptr %q ret void false: ret void } ; Check that we manage to form a zextload after promoting and merging ; two extensions. define void @promoteMergeExtArgZExt(ptr %p, ptr %q, i16 %b) { ; NONSTRESS-LABEL: define void @promoteMergeExtArgZExt( ; NONSTRESS-SAME: ptr [[P:%.*]], ptr [[Q:%.*]], i16 [[B:%.*]]) { ; NONSTRESS-NEXT: [[ENTRY:.*:]] ; NONSTRESS-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; NONSTRESS-NEXT: [[EXT:%.*]] = zext i8 [[T]] to i16 ; NONSTRESS-NEXT: [[ADD:%.*]] = add nuw i16 [[EXT]], [[B]] ; NONSTRESS-NEXT: [[A:%.*]] = icmp slt i8 [[T]], 20 ; NONSTRESS-NEXT: br i1 [[A]], label %[[TRUE:.*]], label %[[FALSE:.*]] ; NONSTRESS: [[TRUE]]: ; NONSTRESS-NEXT: [[S:%.*]] = zext i16 [[ADD]] to i32 ; NONSTRESS-NEXT: store i32 [[S]], ptr [[Q]], align 4 ; NONSTRESS-NEXT: ret void ; NONSTRESS: [[FALSE]]: ; NONSTRESS-NEXT: ret void ; ; STRESS-LABEL: define void @promoteMergeExtArgZExt( ; STRESS-SAME: ptr [[P:%.*]], ptr [[Q:%.*]], i16 [[B:%.*]]) { ; STRESS-NEXT: [[ENTRY:.*:]] ; STRESS-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; STRESS-NEXT: [[PROMOTED2:%.*]] = zext i8 [[T]] to i32 ; STRESS-NEXT: [[PROMOTED1:%.*]] = zext i16 [[B]] to i32 ; STRESS-NEXT: [[ADD:%.*]] = add nuw i32 [[PROMOTED2]], [[PROMOTED1]] ; STRESS-NEXT: [[A:%.*]] = icmp slt i8 [[T]], 20 ; STRESS-NEXT: br i1 [[A]], label %[[TRUE:.*]], label %[[FALSE:.*]] ; STRESS: [[TRUE]]: ; STRESS-NEXT: store i32 [[ADD]], ptr [[Q]], align 4 ; STRESS-NEXT: ret void ; STRESS: [[FALSE]]: ; STRESS-NEXT: ret void ; ; DISABLE-LABEL: define void @promoteMergeExtArgZExt( ; DISABLE-SAME: ptr [[P:%.*]], ptr [[Q:%.*]], i16 [[B:%.*]]) { ; DISABLE-NEXT: [[ENTRY:.*:]] ; DISABLE-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; DISABLE-NEXT: [[EXT:%.*]] = zext i8 [[T]] to i16 ; DISABLE-NEXT: [[ADD:%.*]] = add nuw i16 [[EXT]], [[B]] ; DISABLE-NEXT: [[A:%.*]] = icmp slt i8 [[T]], 20 ; DISABLE-NEXT: br i1 [[A]], label %[[TRUE:.*]], label %[[FALSE:.*]] ; DISABLE: [[TRUE]]: ; DISABLE-NEXT: [[S:%.*]] = zext i16 [[ADD]] to i32 ; DISABLE-NEXT: store i32 [[S]], ptr [[Q]], align 4 ; DISABLE-NEXT: ret void ; DISABLE: [[FALSE]]: ; DISABLE-NEXT: ret void ; entry: %t = load i8, ptr %p %ext = zext i8 %t to i16 %add = add nuw i16 %ext, %b %a = icmp slt i8 %t, 20 br i1 %a, label %true, label %false true: %s = zext i16 %add to i32 store i32 %s, ptr %q ret void false: ret void } ; Check that we manage to form a sextload after promoting and merging ; two extensions. ; Version with sext. define void @promoteMergeExtArgSExt(ptr %p, ptr %q, i16 %b) { ; NONSTRESS-LABEL: define void @promoteMergeExtArgSExt( ; NONSTRESS-SAME: ptr [[P:%.*]], ptr [[Q:%.*]], i16 [[B:%.*]]) { ; NONSTRESS-NEXT: [[ENTRY:.*:]] ; NONSTRESS-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; NONSTRESS-NEXT: [[EXT:%.*]] = zext i8 [[T]] to i16 ; NONSTRESS-NEXT: [[ADD:%.*]] = add nsw i16 [[EXT]], [[B]] ; NONSTRESS-NEXT: [[A:%.*]] = icmp slt i8 [[T]], 20 ; NONSTRESS-NEXT: br i1 [[A]], label %[[TRUE:.*]], label %[[FALSE:.*]] ; NONSTRESS: [[TRUE]]: ; NONSTRESS-NEXT: [[S:%.*]] = sext i16 [[ADD]] to i32 ; NONSTRESS-NEXT: store i32 [[S]], ptr [[Q]], align 4 ; NONSTRESS-NEXT: ret void ; NONSTRESS: [[FALSE]]: ; NONSTRESS-NEXT: ret void ; ; STRESS-LABEL: define void @promoteMergeExtArgSExt( ; STRESS-SAME: ptr [[P:%.*]], ptr [[Q:%.*]], i16 [[B:%.*]]) { ; STRESS-NEXT: [[ENTRY:.*:]] ; STRESS-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; STRESS-NEXT: [[PROMOTED2:%.*]] = zext i8 [[T]] to i32 ; STRESS-NEXT: [[PROMOTED1:%.*]] = sext i16 [[B]] to i32 ; STRESS-NEXT: [[ADD:%.*]] = add nsw i32 [[PROMOTED2]], [[PROMOTED1]] ; STRESS-NEXT: [[A:%.*]] = icmp slt i8 [[T]], 20 ; STRESS-NEXT: br i1 [[A]], label %[[TRUE:.*]], label %[[FALSE:.*]] ; STRESS: [[TRUE]]: ; STRESS-NEXT: store i32 [[ADD]], ptr [[Q]], align 4 ; STRESS-NEXT: ret void ; STRESS: [[FALSE]]: ; STRESS-NEXT: ret void ; ; DISABLE-LABEL: define void @promoteMergeExtArgSExt( ; DISABLE-SAME: ptr [[P:%.*]], ptr [[Q:%.*]], i16 [[B:%.*]]) { ; DISABLE-NEXT: [[ENTRY:.*:]] ; DISABLE-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; DISABLE-NEXT: [[EXT:%.*]] = zext i8 [[T]] to i16 ; DISABLE-NEXT: [[ADD:%.*]] = add nsw i16 [[EXT]], [[B]] ; DISABLE-NEXT: [[A:%.*]] = icmp slt i8 [[T]], 20 ; DISABLE-NEXT: br i1 [[A]], label %[[TRUE:.*]], label %[[FALSE:.*]] ; DISABLE: [[TRUE]]: ; DISABLE-NEXT: [[S:%.*]] = sext i16 [[ADD]] to i32 ; DISABLE-NEXT: store i32 [[S]], ptr [[Q]], align 4 ; DISABLE-NEXT: ret void ; DISABLE: [[FALSE]]: ; DISABLE-NEXT: ret void ; entry: %t = load i8, ptr %p %ext = zext i8 %t to i16 %add = add nsw i16 %ext, %b %a = icmp slt i8 %t, 20 br i1 %a, label %true, label %false true: %s = sext i16 %add to i32 store i32 %s, ptr %q ret void false: ret void } ; Check that we manage to catch all the extload opportunities that are exposed ; by the different iterations of codegen prepare. ; Moreover, check that we do not promote more than we need to. ; Here is what is happening in this test (not necessarly in this order): ; 1. We try to promote the operand of %sextadd. ; a. This creates one sext of %ld2 and one of %zextld ; b. The sext of %ld2 can be combine with %ld2, so we remove one sext but ; introduced one. This is fine with the current heuristic: neutral. ; => We have one zext of %zextld left and we created one sext of %ld2. ; 2. We try to promote the operand of %sextaddza. ; a. This creates one sext of %zexta and one of %zextld ; b. The sext of %zexta can be combined with the zext of %a. ; c. The sext of %zextld leads to %ld and can be combined with it. This is ; done by promoting %zextld. This is fine with the current heuristic: ; neutral. ; => We have created a new zext of %ld and we created one sext of %zexta. ; 3. We try to promote the operand of %sextaddb. ; a. This creates one sext of %b and one of %zextld ; b. The sext of %b is a dead-end, nothing to be done. ; c. Same thing as 2.c. happens. ; => We have created a new zext of %ld and we created one sext of %b. ; 4. We try to promote the operand of the zext of %zextld introduced in #1. ; a. Same thing as 2.c. happens. ; b. %zextld does not have any other uses. It is dead coded. ; => We have created a new zext of %ld and we removed a zext of %zextld and ; a zext of %ld. ; Currently we do not try to reuse existing extensions, so in the end we have ; 3 identical zext of %ld. The extensions will be CSE'ed by SDag. define void @severalPromotions(ptr %addr1, ptr %addr2, i8 %a, i32 %b) { ; OPT-LABEL: define void @severalPromotions( ; OPT-SAME: ptr [[ADDR1:%.*]], ptr [[ADDR2:%.*]], i8 [[A:%.*]], i32 [[B:%.*]]) { ; OPT-NEXT: [[LD:%.*]] = load i8, ptr [[ADDR1]], align 1 ; OPT-NEXT: [[PROMOTED9:%.*]] = zext i8 [[LD]] to i64 ; OPT-NEXT: [[PROMOTED6:%.*]] = zext i8 [[LD]] to i64 ; OPT-NEXT: [[LD2:%.*]] = load i32, ptr [[ADDR2]], align 4 ; OPT-NEXT: [[PROMOTED:%.*]] = sext i32 [[LD2]] to i64 ; OPT-NEXT: [[PROMOTED2:%.*]] = zext i8 [[LD]] to i64 ; OPT-NEXT: [[ADD:%.*]] = add nsw i64 [[PROMOTED]], [[PROMOTED2]] ; OPT-NEXT: [[PROMOTED5:%.*]] = zext i8 [[A]] to i64 ; OPT-NEXT: [[ADDZA:%.*]] = add nsw i64 [[PROMOTED5]], [[PROMOTED6]] ; OPT-NEXT: [[PROMOTED7:%.*]] = sext i32 [[B]] to i64 ; OPT-NEXT: [[ADDB:%.*]] = add nsw i64 [[PROMOTED7]], [[PROMOTED9]] ; OPT-NEXT: call void @dummy(i64 [[ADD]], i64 [[ADDZA]], i64 [[ADDB]]) ; OPT-NEXT: ret void ; ; DISABLE-LABEL: define void @severalPromotions( ; DISABLE-SAME: ptr [[ADDR1:%.*]], ptr [[ADDR2:%.*]], i8 [[A:%.*]], i32 [[B:%.*]]) { ; DISABLE-NEXT: [[LD:%.*]] = load i8, ptr [[ADDR1]], align 1 ; DISABLE-NEXT: [[ZEXTLD:%.*]] = zext i8 [[LD]] to i32 ; DISABLE-NEXT: [[LD2:%.*]] = load i32, ptr [[ADDR2]], align 4 ; DISABLE-NEXT: [[ADD:%.*]] = add nsw i32 [[LD2]], [[ZEXTLD]] ; DISABLE-NEXT: [[SEXTADD:%.*]] = sext i32 [[ADD]] to i64 ; DISABLE-NEXT: [[ZEXTA:%.*]] = zext i8 [[A]] to i32 ; DISABLE-NEXT: [[ADDZA:%.*]] = add nsw i32 [[ZEXTA]], [[ZEXTLD]] ; DISABLE-NEXT: [[SEXTADDZA:%.*]] = sext i32 [[ADDZA]] to i64 ; DISABLE-NEXT: [[ADDB:%.*]] = add nsw i32 [[B]], [[ZEXTLD]] ; DISABLE-NEXT: [[SEXTADDB:%.*]] = sext i32 [[ADDB]] to i64 ; DISABLE-NEXT: call void @dummy(i64 [[SEXTADD]], i64 [[SEXTADDZA]], i64 [[SEXTADDB]]) ; DISABLE-NEXT: ret void ; %ld = load i8, ptr %addr1 %zextld = zext i8 %ld to i32 %ld2 = load i32, ptr %addr2 %add = add nsw i32 %ld2, %zextld %sextadd = sext i32 %add to i64 %zexta = zext i8 %a to i32 %addza = add nsw i32 %zexta, %zextld %sextaddza = sext i32 %addza to i64 %addb = add nsw i32 %b, %zextld %sextaddb = sext i32 %addb to i64 call void @dummy(i64 %sextadd, i64 %sextaddza, i64 %sextaddb) ret void } declare void @dummy(i64, i64, i64) ; Make sure we do not try to promote vector types since the type promotion ; helper does not support them for now. define void @vectorPromotion() { ; OPTALL-LABEL: define void @vectorPromotion() { ; OPTALL-NEXT: [[ENTRY:.*:]] ; OPTALL-NEXT: [[A:%.*]] = shl nuw nsw <2 x i32> zeroinitializer, splat (i32 8) ; OPTALL-NEXT: [[B:%.*]] = zext <2 x i32> [[A]] to <2 x i64> ; OPTALL-NEXT: ret void ; entry: %a = shl nuw nsw <2 x i32> zeroinitializer, %b = zext <2 x i32> %a to <2 x i64> ret void } @a = common global i32 0, align 4 @c = common global [2 x i32] zeroinitializer, align 4 ; Make sure we support promotion of operands that produces a Value as opposed ; to an instruction. ; This used to cause a crash. define i32 @promotionOfArgEndsUpInValue(ptr %addr) { ; OPT-LABEL: define i32 @promotionOfArgEndsUpInValue( ; OPT-SAME: ptr [[ADDR:%.*]]) { ; OPT-NEXT: [[ENTRY:.*:]] ; OPT-NEXT: [[VAL:%.*]] = load i16, ptr [[ADDR]], align 2 ; OPT-NEXT: [[PROMOTED:%.*]] = sext i16 [[VAL]] to i32 ; OPT-NEXT: [[CMP:%.*]] = icmp ne ptr getelementptr inbounds ([2 x i32], ptr @c, i64 0, i64 1), @a ; OPT-NEXT: [[PROMOTED2:%.*]] = zext i1 [[CMP]] to i32 ; OPT-NEXT: [[ADD:%.*]] = add nuw nsw i32 [[PROMOTED]], [[PROMOTED2]] ; OPT-NEXT: ret i32 [[ADD]] ; ; DISABLE-LABEL: define i32 @promotionOfArgEndsUpInValue( ; DISABLE-SAME: ptr [[ADDR:%.*]]) { ; DISABLE-NEXT: [[ENTRY:.*:]] ; DISABLE-NEXT: [[VAL:%.*]] = load i16, ptr [[ADDR]], align 2 ; DISABLE-NEXT: [[CMP:%.*]] = icmp ne ptr getelementptr inbounds ([2 x i32], ptr @c, i64 0, i64 1), @a ; DISABLE-NEXT: [[EXT:%.*]] = zext i1 [[CMP]] to i16 ; DISABLE-NEXT: [[ADD:%.*]] = add nuw nsw i16 [[VAL]], [[EXT]] ; DISABLE-NEXT: [[CONV3:%.*]] = sext i16 [[ADD]] to i32 ; DISABLE-NEXT: ret i32 [[CONV3]] ; entry: %val = load i16, ptr %addr %cmp = icmp ne ptr getelementptr inbounds ([2 x i32], ptr @c, i64 0, i64 1), @a %ext = zext i1 %cmp to i16 %add = add nuw nsw i16 %val, %ext %conv3 = sext i16 %add to i32 ret i32 %conv3 } ; Check that we see that one zext can be derived from the other for free. define void @promoteTwoArgZextWithSourceExtendedTwice(ptr %p, ptr %q, i32 %b, ptr %addr) { ; OPT-LABEL: define void @promoteTwoArgZextWithSourceExtendedTwice( ; OPT-SAME: ptr [[P:%.*]], ptr [[Q:%.*]], i32 [[B:%.*]], ptr [[ADDR:%.*]]) { ; OPT-NEXT: [[ENTRY:.*:]] ; OPT-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; OPT-NEXT: [[PROMOTED1:%.*]] = zext i8 [[T]] to i64 ; OPT-NEXT: [[ZEXTT:%.*]] = zext i8 [[T]] to i32 ; OPT-NEXT: [[ADD:%.*]] = add nuw i32 [[ZEXTT]], [[B]] ; OPT-NEXT: [[ADD2:%.*]] = add nuw i64 [[PROMOTED1]], 12 ; OPT-NEXT: store i32 [[ADD]], ptr [[ADDR]], align 4 ; OPT-NEXT: store i64 [[ADD2]], ptr [[Q]], align 8 ; OPT-NEXT: ret void ; ; DISABLE-LABEL: define void @promoteTwoArgZextWithSourceExtendedTwice( ; DISABLE-SAME: ptr [[P:%.*]], ptr [[Q:%.*]], i32 [[B:%.*]], ptr [[ADDR:%.*]]) { ; DISABLE-NEXT: [[ENTRY:.*:]] ; DISABLE-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; DISABLE-NEXT: [[ZEXTT:%.*]] = zext i8 [[T]] to i32 ; DISABLE-NEXT: [[ADD:%.*]] = add nuw i32 [[ZEXTT]], [[B]] ; DISABLE-NEXT: [[ADD2:%.*]] = add nuw i32 [[ZEXTT]], 12 ; DISABLE-NEXT: store i32 [[ADD]], ptr [[ADDR]], align 4 ; DISABLE-NEXT: [[S:%.*]] = zext i32 [[ADD2]] to i64 ; DISABLE-NEXT: store i64 [[S]], ptr [[Q]], align 8 ; DISABLE-NEXT: ret void ; entry: %t = load i8, ptr %p %zextt = zext i8 %t to i32 %add = add nuw i32 %zextt, %b %add2 = add nuw i32 %zextt, 12 store i32 %add, ptr %addr %s = zext i32 %add2 to i64 store i64 %s, ptr %q ret void } ; Check that we do not increase the cost of the code. ; The input has one free zext and one free sext. If we would have promoted ; all the way through the load we would end up with a free zext and a ; non-free sext (of %b). define void @doNotPromoteFreeSExtFromAddrMode(ptr %p, i32 %b, ptr %addr) { ; NONSTRESS-LABEL: define void @doNotPromoteFreeSExtFromAddrMode( ; NONSTRESS-SAME: ptr [[P:%.*]], i32 [[B:%.*]], ptr [[ADDR:%.*]]) { ; NONSTRESS-NEXT: [[ENTRY:.*:]] ; NONSTRESS-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; NONSTRESS-NEXT: [[ZEXTT:%.*]] = zext i8 [[T]] to i32 ; NONSTRESS-NEXT: [[ADD:%.*]] = add nsw i32 [[ZEXTT]], [[B]] ; NONSTRESS-NEXT: [[IDX64:%.*]] = sext i32 [[ADD]] to i64 ; NONSTRESS-NEXT: [[STADDR:%.*]] = getelementptr inbounds i32, ptr [[ADDR]], i64 [[IDX64]] ; NONSTRESS-NEXT: store i32 [[ADD]], ptr [[STADDR]], align 4 ; NONSTRESS-NEXT: ret void ; ; STRESS-LABEL: define void @doNotPromoteFreeSExtFromAddrMode( ; STRESS-SAME: ptr [[P:%.*]], i32 [[B:%.*]], ptr [[ADDR:%.*]]) { ; STRESS-NEXT: [[ENTRY:.*:]] ; STRESS-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; STRESS-NEXT: [[PROMOTED3:%.*]] = zext i8 [[T]] to i64 ; STRESS-NEXT: [[PROMOTED2:%.*]] = sext i32 [[B]] to i64 ; STRESS-NEXT: [[ADD:%.*]] = add nsw i64 [[PROMOTED3]], [[PROMOTED2]] ; STRESS-NEXT: [[PROMOTED:%.*]] = trunc i64 [[ADD]] to i32 ; STRESS-NEXT: [[STADDR:%.*]] = getelementptr inbounds i32, ptr [[ADDR]], i64 [[ADD]] ; STRESS-NEXT: store i32 [[PROMOTED]], ptr [[STADDR]], align 4 ; STRESS-NEXT: ret void ; ; DISABLE-LABEL: define void @doNotPromoteFreeSExtFromAddrMode( ; DISABLE-SAME: ptr [[P:%.*]], i32 [[B:%.*]], ptr [[ADDR:%.*]]) { ; DISABLE-NEXT: [[ENTRY:.*:]] ; DISABLE-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; DISABLE-NEXT: [[ZEXTT:%.*]] = zext i8 [[T]] to i32 ; DISABLE-NEXT: [[ADD:%.*]] = add nsw i32 [[ZEXTT]], [[B]] ; DISABLE-NEXT: [[IDX64:%.*]] = sext i32 [[ADD]] to i64 ; DISABLE-NEXT: [[STADDR:%.*]] = getelementptr inbounds i32, ptr [[ADDR]], i64 [[IDX64]] ; DISABLE-NEXT: store i32 [[ADD]], ptr [[STADDR]], align 4 ; DISABLE-NEXT: ret void ; entry: %t = load i8, ptr %p %zextt = zext i8 %t to i32 %add = add nsw i32 %zextt, %b %idx64 = sext i32 %add to i64 %staddr = getelementptr inbounds i32, ptr %addr, i64 %idx64 store i32 %add, ptr %staddr ret void } ; Check that we do not increase the cost of the code. ; The input has one free zext and one free sext. If we would have promoted ; all the way through the load we would end up with a free zext and a ; non-free sext (of %b). define void @doNotPromoteFreeSExtFromAddrMode64(ptr %p, i32 %b, ptr %addr, i64 %stuff) { ; NONSTRESS-LABEL: define void @doNotPromoteFreeSExtFromAddrMode64( ; NONSTRESS-SAME: ptr [[P:%.*]], i32 [[B:%.*]], ptr [[ADDR:%.*]], i64 [[STUFF:%.*]]) { ; NONSTRESS-NEXT: [[ENTRY:.*:]] ; NONSTRESS-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; NONSTRESS-NEXT: [[ZEXTT:%.*]] = zext i8 [[T]] to i32 ; NONSTRESS-NEXT: [[ADD:%.*]] = add nsw i32 [[ZEXTT]], [[B]] ; NONSTRESS-NEXT: [[IDX64:%.*]] = sext i32 [[ADD]] to i64 ; NONSTRESS-NEXT: [[STADDR:%.*]] = getelementptr inbounds i64, ptr [[ADDR]], i64 [[IDX64]] ; NONSTRESS-NEXT: store i64 [[STUFF]], ptr [[STADDR]], align 8 ; NONSTRESS-NEXT: ret void ; ; STRESS-LABEL: define void @doNotPromoteFreeSExtFromAddrMode64( ; STRESS-SAME: ptr [[P:%.*]], i32 [[B:%.*]], ptr [[ADDR:%.*]], i64 [[STUFF:%.*]]) { ; STRESS-NEXT: [[ENTRY:.*:]] ; STRESS-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; STRESS-NEXT: [[PROMOTED2:%.*]] = zext i8 [[T]] to i64 ; STRESS-NEXT: [[PROMOTED1:%.*]] = sext i32 [[B]] to i64 ; STRESS-NEXT: [[ADD:%.*]] = add nsw i64 [[PROMOTED2]], [[PROMOTED1]] ; STRESS-NEXT: [[STADDR:%.*]] = getelementptr inbounds i64, ptr [[ADDR]], i64 [[ADD]] ; STRESS-NEXT: store i64 [[STUFF]], ptr [[STADDR]], align 8 ; STRESS-NEXT: ret void ; ; DISABLE-LABEL: define void @doNotPromoteFreeSExtFromAddrMode64( ; DISABLE-SAME: ptr [[P:%.*]], i32 [[B:%.*]], ptr [[ADDR:%.*]], i64 [[STUFF:%.*]]) { ; DISABLE-NEXT: [[ENTRY:.*:]] ; DISABLE-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; DISABLE-NEXT: [[ZEXTT:%.*]] = zext i8 [[T]] to i32 ; DISABLE-NEXT: [[ADD:%.*]] = add nsw i32 [[ZEXTT]], [[B]] ; DISABLE-NEXT: [[IDX64:%.*]] = sext i32 [[ADD]] to i64 ; DISABLE-NEXT: [[STADDR:%.*]] = getelementptr inbounds i64, ptr [[ADDR]], i64 [[IDX64]] ; DISABLE-NEXT: store i64 [[STUFF]], ptr [[STADDR]], align 8 ; DISABLE-NEXT: ret void ; entry: %t = load i8, ptr %p %zextt = zext i8 %t to i32 %add = add nsw i32 %zextt, %b %idx64 = sext i32 %add to i64 %staddr = getelementptr inbounds i64, ptr %addr, i64 %idx64 store i64 %stuff, ptr %staddr ret void } ; Check that we do not increase the cost of the code. ; The input has one free zext and one free sext. If we would have promoted ; all the way through the load we would end up with a free zext and a ; non-free sext (of %b). define void @doNotPromoteFreeSExtFromAddrMode128(ptr %p, i32 %b, ptr %addr, i128 %stuff) { ; NONSTRESS-LABEL: define void @doNotPromoteFreeSExtFromAddrMode128( ; NONSTRESS-SAME: ptr [[P:%.*]], i32 [[B:%.*]], ptr [[ADDR:%.*]], i128 [[STUFF:%.*]]) { ; NONSTRESS-NEXT: [[ENTRY:.*:]] ; NONSTRESS-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; NONSTRESS-NEXT: [[ZEXTT:%.*]] = zext i8 [[T]] to i32 ; NONSTRESS-NEXT: [[ADD:%.*]] = add nsw i32 [[ZEXTT]], [[B]] ; NONSTRESS-NEXT: [[IDX64:%.*]] = sext i32 [[ADD]] to i64 ; NONSTRESS-NEXT: [[STADDR:%.*]] = getelementptr inbounds i128, ptr [[ADDR]], i64 [[IDX64]] ; NONSTRESS-NEXT: store i128 [[STUFF]], ptr [[STADDR]], align 16 ; NONSTRESS-NEXT: ret void ; ; STRESS-LABEL: define void @doNotPromoteFreeSExtFromAddrMode128( ; STRESS-SAME: ptr [[P:%.*]], i32 [[B:%.*]], ptr [[ADDR:%.*]], i128 [[STUFF:%.*]]) { ; STRESS-NEXT: [[ENTRY:.*:]] ; STRESS-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; STRESS-NEXT: [[PROMOTED2:%.*]] = zext i8 [[T]] to i64 ; STRESS-NEXT: [[PROMOTED1:%.*]] = sext i32 [[B]] to i64 ; STRESS-NEXT: [[ADD:%.*]] = add nsw i64 [[PROMOTED2]], [[PROMOTED1]] ; STRESS-NEXT: [[STADDR:%.*]] = getelementptr inbounds i128, ptr [[ADDR]], i64 [[ADD]] ; STRESS-NEXT: store i128 [[STUFF]], ptr [[STADDR]], align 16 ; STRESS-NEXT: ret void ; ; DISABLE-LABEL: define void @doNotPromoteFreeSExtFromAddrMode128( ; DISABLE-SAME: ptr [[P:%.*]], i32 [[B:%.*]], ptr [[ADDR:%.*]], i128 [[STUFF:%.*]]) { ; DISABLE-NEXT: [[ENTRY:.*:]] ; DISABLE-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; DISABLE-NEXT: [[ZEXTT:%.*]] = zext i8 [[T]] to i32 ; DISABLE-NEXT: [[ADD:%.*]] = add nsw i32 [[ZEXTT]], [[B]] ; DISABLE-NEXT: [[IDX64:%.*]] = sext i32 [[ADD]] to i64 ; DISABLE-NEXT: [[STADDR:%.*]] = getelementptr inbounds i128, ptr [[ADDR]], i64 [[IDX64]] ; DISABLE-NEXT: store i128 [[STUFF]], ptr [[STADDR]], align 16 ; DISABLE-NEXT: ret void ; entry: %t = load i8, ptr %p %zextt = zext i8 %t to i32 %add = add nsw i32 %zextt, %b %idx64 = sext i32 %add to i64 %staddr = getelementptr inbounds i128, ptr %addr, i64 %idx64 store i128 %stuff, ptr %staddr ret void } ; Check that we do not increase the cost of the code. ; The input has one free zext and one free sext. If we would have promoted ; all the way through the load we would end up with a free zext and a ; non-free sext (of %b). define void @promoteSExtFromAddrMode256(ptr %p, i32 %b, ptr %addr, i256 %stuff) { ; OPT-LABEL: define void @promoteSExtFromAddrMode256( ; OPT-SAME: ptr [[P:%.*]], i32 [[B:%.*]], ptr [[ADDR:%.*]], i256 [[STUFF:%.*]]) { ; OPT-NEXT: [[ENTRY:.*:]] ; OPT-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; OPT-NEXT: [[PROMOTED2:%.*]] = zext i8 [[T]] to i64 ; OPT-NEXT: [[PROMOTED1:%.*]] = sext i32 [[B]] to i64 ; OPT-NEXT: [[ADD:%.*]] = add nsw i64 [[PROMOTED2]], [[PROMOTED1]] ; OPT-NEXT: [[STADDR:%.*]] = getelementptr inbounds i256, ptr [[ADDR]], i64 [[ADD]] ; OPT-NEXT: store i256 [[STUFF]], ptr [[STADDR]], align 16 ; OPT-NEXT: ret void ; ; DISABLE-LABEL: define void @promoteSExtFromAddrMode256( ; DISABLE-SAME: ptr [[P:%.*]], i32 [[B:%.*]], ptr [[ADDR:%.*]], i256 [[STUFF:%.*]]) { ; DISABLE-NEXT: [[ENTRY:.*:]] ; DISABLE-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; DISABLE-NEXT: [[ZEXTT:%.*]] = zext i8 [[T]] to i32 ; DISABLE-NEXT: [[ADD:%.*]] = add nsw i32 [[ZEXTT]], [[B]] ; DISABLE-NEXT: [[IDX64:%.*]] = sext i32 [[ADD]] to i64 ; DISABLE-NEXT: [[STADDR:%.*]] = getelementptr inbounds i256, ptr [[ADDR]], i64 [[IDX64]] ; DISABLE-NEXT: store i256 [[STUFF]], ptr [[STADDR]], align 16 ; DISABLE-NEXT: ret void ; entry: %t = load i8, ptr %p %zextt = zext i8 %t to i32 %add = add nsw i32 %zextt, %b %idx64 = sext i32 %add to i64 %staddr = getelementptr inbounds i256, ptr %addr, i64 %idx64 store i256 %stuff, ptr %staddr ret void } ; Check that we do not increase the cost of the code. ; The input has one free zext and one free zext. ; When we promote all the way through the load, we end up with ; a free zext and a non-free zext (of %b). ; However, the current target lowering says zext i32 to i64 is free ; so the promotion happens because the cost did not change and may ; expose more opportunities. ; This would need to be fixed at some point. ; ; This transformation should really happen only for stress mode. define void @doNotPromoteFreeZExtFromAddrMode(ptr %p, i32 %b, ptr %addr) { ; NONSTRESS-LABEL: define void @doNotPromoteFreeZExtFromAddrMode( ; NONSTRESS-SAME: ptr [[P:%.*]], i32 [[B:%.*]], ptr [[ADDR:%.*]]) { ; NONSTRESS-NEXT: [[ENTRY:.*:]] ; NONSTRESS-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; NONSTRESS-NEXT: [[ZEXTT:%.*]] = zext i8 [[T]] to i32 ; NONSTRESS-NEXT: [[ADD:%.*]] = add nuw i32 [[ZEXTT]], [[B]] ; NONSTRESS-NEXT: [[IDX64:%.*]] = zext i32 [[ADD]] to i64 ; NONSTRESS-NEXT: [[STADDR:%.*]] = getelementptr inbounds i32, ptr [[ADDR]], i64 [[IDX64]] ; NONSTRESS-NEXT: store i32 [[ADD]], ptr [[STADDR]], align 4 ; NONSTRESS-NEXT: ret void ; ; STRESS-LABEL: define void @doNotPromoteFreeZExtFromAddrMode( ; STRESS-SAME: ptr [[P:%.*]], i32 [[B:%.*]], ptr [[ADDR:%.*]]) { ; STRESS-NEXT: [[ENTRY:.*:]] ; STRESS-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; STRESS-NEXT: [[PROMOTED3:%.*]] = zext i8 [[T]] to i64 ; STRESS-NEXT: [[PROMOTED2:%.*]] = zext i32 [[B]] to i64 ; STRESS-NEXT: [[ADD:%.*]] = add nuw i64 [[PROMOTED3]], [[PROMOTED2]] ; STRESS-NEXT: [[PROMOTED:%.*]] = trunc i64 [[ADD]] to i32 ; STRESS-NEXT: [[STADDR:%.*]] = getelementptr inbounds i32, ptr [[ADDR]], i64 [[ADD]] ; STRESS-NEXT: store i32 [[PROMOTED]], ptr [[STADDR]], align 4 ; STRESS-NEXT: ret void ; ; DISABLE-LABEL: define void @doNotPromoteFreeZExtFromAddrMode( ; DISABLE-SAME: ptr [[P:%.*]], i32 [[B:%.*]], ptr [[ADDR:%.*]]) { ; DISABLE-NEXT: [[ENTRY:.*:]] ; DISABLE-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; DISABLE-NEXT: [[ZEXTT:%.*]] = zext i8 [[T]] to i32 ; DISABLE-NEXT: [[ADD:%.*]] = add nuw i32 [[ZEXTT]], [[B]] ; DISABLE-NEXT: [[IDX64:%.*]] = zext i32 [[ADD]] to i64 ; DISABLE-NEXT: [[STADDR:%.*]] = getelementptr inbounds i32, ptr [[ADDR]], i64 [[IDX64]] ; DISABLE-NEXT: store i32 [[ADD]], ptr [[STADDR]], align 4 ; DISABLE-NEXT: ret void ; entry: %t = load i8, ptr %p %zextt = zext i8 %t to i32 %add = add nuw i32 %zextt, %b %idx64 = zext i32 %add to i64 %staddr = getelementptr inbounds i32, ptr %addr, i64 %idx64 store i32 %add, ptr %staddr ret void } define i64 @doNotPromoteFreeSExtFromShift(ptr %p, i32 %b) { ; NONSTRESS-LABEL: define i64 @doNotPromoteFreeSExtFromShift( ; NONSTRESS-SAME: ptr [[P:%.*]], i32 [[B:%.*]]) { ; NONSTRESS-NEXT: [[ENTRY:.*:]] ; NONSTRESS-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; NONSTRESS-NEXT: [[ZEXTT:%.*]] = zext i8 [[T]] to i32 ; NONSTRESS-NEXT: [[ADD:%.*]] = add nsw i32 [[ZEXTT]], [[B]] ; NONSTRESS-NEXT: [[IDX64:%.*]] = sext i32 [[ADD]] to i64 ; NONSTRESS-NEXT: [[STADDR:%.*]] = shl i64 [[IDX64]], 12 ; NONSTRESS-NEXT: ret i64 [[STADDR]] ; ; STRESS-LABEL: define i64 @doNotPromoteFreeSExtFromShift( ; STRESS-SAME: ptr [[P:%.*]], i32 [[B:%.*]]) { ; STRESS-NEXT: [[ENTRY:.*:]] ; STRESS-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; STRESS-NEXT: [[PROMOTED2:%.*]] = zext i8 [[T]] to i64 ; STRESS-NEXT: [[PROMOTED1:%.*]] = sext i32 [[B]] to i64 ; STRESS-NEXT: [[ADD:%.*]] = add nsw i64 [[PROMOTED2]], [[PROMOTED1]] ; STRESS-NEXT: [[STADDR:%.*]] = shl i64 [[ADD]], 12 ; STRESS-NEXT: ret i64 [[STADDR]] ; ; DISABLE-LABEL: define i64 @doNotPromoteFreeSExtFromShift( ; DISABLE-SAME: ptr [[P:%.*]], i32 [[B:%.*]]) { ; DISABLE-NEXT: [[ENTRY:.*:]] ; DISABLE-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; DISABLE-NEXT: [[ZEXTT:%.*]] = zext i8 [[T]] to i32 ; DISABLE-NEXT: [[ADD:%.*]] = add nsw i32 [[ZEXTT]], [[B]] ; DISABLE-NEXT: [[IDX64:%.*]] = sext i32 [[ADD]] to i64 ; DISABLE-NEXT: [[STADDR:%.*]] = shl i64 [[IDX64]], 12 ; DISABLE-NEXT: ret i64 [[STADDR]] ; entry: %t = load i8, ptr %p %zextt = zext i8 %t to i32 %add = add nsw i32 %zextt, %b %idx64 = sext i32 %add to i64 %staddr = shl i64 %idx64, 12 ret i64 %staddr } ; Same comment as doNotPromoteFreeZExtFromAddrMode. ; ; This transformation should really happen only for stress mode. define i64 @doNotPromoteFreeZExtFromShift(ptr %p, i32 %b) { ; NONSTRESS-LABEL: define i64 @doNotPromoteFreeZExtFromShift( ; NONSTRESS-SAME: ptr [[P:%.*]], i32 [[B:%.*]]) { ; NONSTRESS-NEXT: [[ENTRY:.*:]] ; NONSTRESS-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; NONSTRESS-NEXT: [[ZEXTT:%.*]] = zext i8 [[T]] to i32 ; NONSTRESS-NEXT: [[ADD:%.*]] = add nuw i32 [[ZEXTT]], [[B]] ; NONSTRESS-NEXT: [[IDX64:%.*]] = zext i32 [[ADD]] to i64 ; NONSTRESS-NEXT: [[STADDR:%.*]] = shl i64 [[IDX64]], 12 ; NONSTRESS-NEXT: ret i64 [[STADDR]] ; ; STRESS-LABEL: define i64 @doNotPromoteFreeZExtFromShift( ; STRESS-SAME: ptr [[P:%.*]], i32 [[B:%.*]]) { ; STRESS-NEXT: [[ENTRY:.*:]] ; STRESS-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; STRESS-NEXT: [[PROMOTED2:%.*]] = zext i8 [[T]] to i64 ; STRESS-NEXT: [[PROMOTED1:%.*]] = zext i32 [[B]] to i64 ; STRESS-NEXT: [[ADD:%.*]] = add nuw i64 [[PROMOTED2]], [[PROMOTED1]] ; STRESS-NEXT: [[STADDR:%.*]] = shl i64 [[ADD]], 12 ; STRESS-NEXT: ret i64 [[STADDR]] ; ; DISABLE-LABEL: define i64 @doNotPromoteFreeZExtFromShift( ; DISABLE-SAME: ptr [[P:%.*]], i32 [[B:%.*]]) { ; DISABLE-NEXT: [[ENTRY:.*:]] ; DISABLE-NEXT: [[T:%.*]] = load i8, ptr [[P]], align 1 ; DISABLE-NEXT: [[ZEXTT:%.*]] = zext i8 [[T]] to i32 ; DISABLE-NEXT: [[ADD:%.*]] = add nuw i32 [[ZEXTT]], [[B]] ; DISABLE-NEXT: [[IDX64:%.*]] = zext i32 [[ADD]] to i64 ; DISABLE-NEXT: [[STADDR:%.*]] = shl i64 [[IDX64]], 12 ; DISABLE-NEXT: ret i64 [[STADDR]] ; entry: %t = load i8, ptr %p %zextt = zext i8 %t to i32 %add = add nuw i32 %zextt, %b %idx64 = zext i32 %add to i64 %staddr = shl i64 %idx64, 12 ret i64 %staddr } ; The input has one free zext and one non-free sext. ; When we promote all the way through to the load, we end up with ; a free zext, a free sext (%ld1), and a non-free sext (of %cst). ; However, we when generate load pair and the free sext(%ld1) becomes ; non-free. So technically, we trade a non-free sext to two non-free ; sext. ; This would need to be fixed at some point. ; ; This transformation should really happen only for stress mode. define i64 @doNotPromoteBecauseOfPairedLoad(ptr %p, i32 %cst) { ; OPT-LABEL: define i64 @doNotPromoteBecauseOfPairedLoad( ; OPT-SAME: ptr [[P:%.*]], i32 [[CST:%.*]]) { ; OPT-NEXT: [[LD0:%.*]] = load i32, ptr [[P]], align 4 ; OPT-NEXT: [[IDXLD1:%.*]] = getelementptr inbounds i32, ptr [[P]], i64 1 ; OPT-NEXT: [[LD1:%.*]] = load i32, ptr [[IDXLD1]], align 4 ; OPT-NEXT: [[PROMOTED:%.*]] = sext i32 [[LD1]] to i64 ; OPT-NEXT: [[PROMOTED1:%.*]] = sext i32 [[CST]] to i64 ; OPT-NEXT: [[RES:%.*]] = add nsw i64 [[PROMOTED]], [[PROMOTED1]] ; OPT-NEXT: [[ZEXTLD0:%.*]] = zext i32 [[LD0]] to i64 ; OPT-NEXT: [[FINAL:%.*]] = add i64 [[RES]], [[ZEXTLD0]] ; OPT-NEXT: ret i64 [[FINAL]] ; ; DISABLE-LABEL: define i64 @doNotPromoteBecauseOfPairedLoad( ; DISABLE-SAME: ptr [[P:%.*]], i32 [[CST:%.*]]) { ; DISABLE-NEXT: [[LD0:%.*]] = load i32, ptr [[P]], align 4 ; DISABLE-NEXT: [[IDXLD1:%.*]] = getelementptr inbounds i32, ptr [[P]], i64 1 ; DISABLE-NEXT: [[LD1:%.*]] = load i32, ptr [[IDXLD1]], align 4 ; DISABLE-NEXT: [[RES:%.*]] = add nsw i32 [[LD1]], [[CST]] ; DISABLE-NEXT: [[SEXTRES:%.*]] = sext i32 [[RES]] to i64 ; DISABLE-NEXT: [[ZEXTLD0:%.*]] = zext i32 [[LD0]] to i64 ; DISABLE-NEXT: [[FINAL:%.*]] = add i64 [[SEXTRES]], [[ZEXTLD0]] ; DISABLE-NEXT: ret i64 [[FINAL]] ; %ld0 = load i32, ptr %p %idxLd1 = getelementptr inbounds i32, ptr %p, i64 1 %ld1 = load i32, ptr %idxLd1 %res = add nsw i32 %ld1, %cst %sextres = sext i32 %res to i64 %zextLd0 = zext i32 %ld0 to i64 %final = add i64 %sextres, %zextLd0 ret i64 %final } define i64 @promoteZextShl(i1 %c, ptr %P) { ; OPT-LABEL: define i64 @promoteZextShl( ; OPT-SAME: i1 [[C:%.*]], ptr [[P:%.*]]) { ; OPT-NEXT: [[ENTRY:.*:]] ; OPT-NEXT: [[LD:%.*]] = load i16, ptr [[P]], align 2 ; OPT-NEXT: [[PROMOTED1:%.*]] = zext i16 [[LD]] to i64 ; OPT-NEXT: br i1 [[C]], label %[[END:.*]], label %[[IF_THEN:.*]] ; OPT: [[IF_THEN]]: ; OPT-NEXT: [[SHL2:%.*]] = shl nsw i64 [[PROMOTED1]], 1 ; OPT-NEXT: ret i64 [[SHL2]] ; OPT: [[END]]: ; OPT-NEXT: ret i64 0 ; ; DISABLE-LABEL: define i64 @promoteZextShl( ; DISABLE-SAME: i1 [[C:%.*]], ptr [[P:%.*]]) { ; DISABLE-NEXT: [[ENTRY:.*:]] ; DISABLE-NEXT: [[LD:%.*]] = load i16, ptr [[P]], align 2 ; DISABLE-NEXT: [[Z:%.*]] = zext i16 [[LD]] to i32 ; DISABLE-NEXT: br i1 [[C]], label %[[END:.*]], label %[[IF_THEN:.*]] ; DISABLE: [[IF_THEN]]: ; DISABLE-NEXT: [[SHL2:%.*]] = shl nsw i32 [[Z]], 1 ; DISABLE-NEXT: [[R:%.*]] = sext i32 [[SHL2]] to i64 ; DISABLE-NEXT: ret i64 [[R]] ; DISABLE: [[END]]: ; DISABLE-NEXT: ret i64 0 ; entry: %ld = load i16, ptr %P br i1 %c, label %end, label %if.then if.then: %z = zext i16 %ld to i32 %shl2 = shl nsw i32 %z, 1 %r = sext i32 %shl2 to i64 ret i64 %r end: ret i64 0 }