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/InstSimplify/redundant-null-check-in-uad...

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

268 lines
12 KiB
LLVM
Raw Normal View History

[NFC][InstSimplify] Some tests for dropping null check after uadd.with.overflow of non-null (PR43246) https://rise4fun.com/Alive/WRzq Name: C <= Y && Y != 0 --> C <= Y iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = and i1 %y_is_nonnull, %no_overflow => %r = %no_overflow Name: C <= Y || Y != 0 --> Y != 0 iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = or i1 %y_is_nonnull, %no_overflow => %r = %y_is_nonnull Name: C > Y || Y == 0 --> C > Y iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = or i1 %y_is_null, %overflow => %r = %overflow Name: C > Y && Y == 0 --> Y == 0 iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = and i1 %y_is_null, %overflow => %r = %y_is_null https://bugs.llvm.org/show_bug.cgi?id=43246 llvm-svn: 371339
2019-09-09 01:50:40 +08:00
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt %s -instsimplify -S | FileCheck %s
; Here we have add some offset to a non-null pointer,
; and check that the result does not overflow and is not a null pointer.
; But since the base pointer is already non-null, and we check for overflow,
; that will already catch the get null pointer,
; so the separate null check is redundant and can be dropped.
define i1 @t0(i8* nonnull %base, i64 %offset) {
; CHECK-LABEL: @t0(
; CHECK-NEXT: [[BASE_INT:%.*]] = ptrtoint i8* [[BASE:%.*]] to i64
; CHECK-NEXT: [[ADJUSTED:%.*]] = add i64 [[BASE_INT]], [[OFFSET:%.*]]
; CHECK-NEXT: [[NO_OVERFLOW_DURING_ADJUSTMENT:%.*]] = icmp uge i64 [[ADJUSTED]], [[BASE_INT]]
[InstSimplify] simplifyUnsignedRangeCheck(): if we know that X != 0, handle more cases (PR43246) Summary: This is motivated by D67122 sanitizer check enhancement. That patch seemingly worsens `-fsanitize=pointer-overflow` overhead from 25% to 50%, which strongly implies missing folds. In this particular case, given ``` char* test(char& base, unsigned long offset) { return &base + offset; } ``` it will end up producing something like https://godbolt.org/z/LK5-iH which after optimizations reduces down to roughly ``` define i1 @t0(i8* nonnull %base, i64 %offset) { %base_int = ptrtoint i8* %base to i64 %adjusted = add i64 %base_int, %offset %non_null_after_adjustment = icmp ne i64 %adjusted, 0 %no_overflow_during_adjustment = icmp uge i64 %adjusted, %base_int %res = and i1 %non_null_after_adjustment, %no_overflow_during_adjustment ret i1 %res } ``` Without D67122 there was no `%non_null_after_adjustment`, and in this particular case we can get rid of the overhead: Here we add some offset to a non-null pointer, and check that the result does not overflow and is not a null pointer. But since the base pointer is already non-null, and we check for overflow, that overflow check will already catch the null pointer, so the separate null check is redundant and can be dropped. Alive proofs: https://rise4fun.com/Alive/WRzq There are more patterns of "unsigned-add-with-overflow", they are not handled here, but this is the main pattern, that we currently consider canonical, so it makes sense to handle it. https://bugs.llvm.org/show_bug.cgi?id=43246 Reviewers: spatel, nikic, vsk Reviewed By: spatel Subscribers: hiraditya, llvm-commits, reames Tags: #llvm Differential Revision: https://reviews.llvm.org/D67332 llvm-svn: 371349
2019-09-09 04:14:15 +08:00
; CHECK-NEXT: ret i1 [[NO_OVERFLOW_DURING_ADJUSTMENT]]
[NFC][InstSimplify] Some tests for dropping null check after uadd.with.overflow of non-null (PR43246) https://rise4fun.com/Alive/WRzq Name: C <= Y && Y != 0 --> C <= Y iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = and i1 %y_is_nonnull, %no_overflow => %r = %no_overflow Name: C <= Y || Y != 0 --> Y != 0 iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = or i1 %y_is_nonnull, %no_overflow => %r = %y_is_nonnull Name: C > Y || Y == 0 --> C > Y iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = or i1 %y_is_null, %overflow => %r = %overflow Name: C > Y && Y == 0 --> Y == 0 iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = and i1 %y_is_null, %overflow => %r = %y_is_null https://bugs.llvm.org/show_bug.cgi?id=43246 llvm-svn: 371339
2019-09-09 01:50:40 +08:00
;
%base_int = ptrtoint i8* %base to i64
%adjusted = add i64 %base_int, %offset
%non_null_after_adjustment = icmp ne i64 %adjusted, 0
%no_overflow_during_adjustment = icmp uge i64 %adjusted, %base_int
%res = and i1 %non_null_after_adjustment, %no_overflow_during_adjustment
ret i1 %res
}
define i1 @t1(i8* nonnull %base, i64 %offset) {
; CHECK-LABEL: @t1(
; CHECK-NEXT: [[BASE_INT:%.*]] = ptrtoint i8* [[BASE:%.*]] to i64
; CHECK-NEXT: [[ADJUSTED:%.*]] = add i64 [[BASE_INT]], [[OFFSET:%.*]]
; CHECK-NEXT: [[NO_OVERFLOW_DURING_ADJUSTMENT:%.*]] = icmp ule i64 [[BASE_INT]], [[ADJUSTED]]
[InstSimplify] simplifyUnsignedRangeCheck(): if we know that X != 0, handle more cases (PR43246) Summary: This is motivated by D67122 sanitizer check enhancement. That patch seemingly worsens `-fsanitize=pointer-overflow` overhead from 25% to 50%, which strongly implies missing folds. In this particular case, given ``` char* test(char& base, unsigned long offset) { return &base + offset; } ``` it will end up producing something like https://godbolt.org/z/LK5-iH which after optimizations reduces down to roughly ``` define i1 @t0(i8* nonnull %base, i64 %offset) { %base_int = ptrtoint i8* %base to i64 %adjusted = add i64 %base_int, %offset %non_null_after_adjustment = icmp ne i64 %adjusted, 0 %no_overflow_during_adjustment = icmp uge i64 %adjusted, %base_int %res = and i1 %non_null_after_adjustment, %no_overflow_during_adjustment ret i1 %res } ``` Without D67122 there was no `%non_null_after_adjustment`, and in this particular case we can get rid of the overhead: Here we add some offset to a non-null pointer, and check that the result does not overflow and is not a null pointer. But since the base pointer is already non-null, and we check for overflow, that overflow check will already catch the null pointer, so the separate null check is redundant and can be dropped. Alive proofs: https://rise4fun.com/Alive/WRzq There are more patterns of "unsigned-add-with-overflow", they are not handled here, but this is the main pattern, that we currently consider canonical, so it makes sense to handle it. https://bugs.llvm.org/show_bug.cgi?id=43246 Reviewers: spatel, nikic, vsk Reviewed By: spatel Subscribers: hiraditya, llvm-commits, reames Tags: #llvm Differential Revision: https://reviews.llvm.org/D67332 llvm-svn: 371349
2019-09-09 04:14:15 +08:00
; CHECK-NEXT: ret i1 [[NO_OVERFLOW_DURING_ADJUSTMENT]]
[NFC][InstSimplify] Some tests for dropping null check after uadd.with.overflow of non-null (PR43246) https://rise4fun.com/Alive/WRzq Name: C <= Y && Y != 0 --> C <= Y iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = and i1 %y_is_nonnull, %no_overflow => %r = %no_overflow Name: C <= Y || Y != 0 --> Y != 0 iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = or i1 %y_is_nonnull, %no_overflow => %r = %y_is_nonnull Name: C > Y || Y == 0 --> C > Y iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = or i1 %y_is_null, %overflow => %r = %overflow Name: C > Y && Y == 0 --> Y == 0 iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = and i1 %y_is_null, %overflow => %r = %y_is_null https://bugs.llvm.org/show_bug.cgi?id=43246 llvm-svn: 371339
2019-09-09 01:50:40 +08:00
;
%base_int = ptrtoint i8* %base to i64
%adjusted = add i64 %base_int, %offset
%non_null_after_adjustment = icmp ne i64 %adjusted, 0
%no_overflow_during_adjustment = icmp ule i64 %base_int, %adjusted ; swapped
%res = and i1 %non_null_after_adjustment, %no_overflow_during_adjustment
ret i1 %res
}
define i1 @t2(i8* nonnull %base, i64 %offset) {
; CHECK-LABEL: @t2(
; CHECK-NEXT: [[BASE_INT:%.*]] = ptrtoint i8* [[BASE:%.*]] to i64
; CHECK-NEXT: [[ADJUSTED:%.*]] = add i64 [[BASE_INT]], [[OFFSET:%.*]]
; CHECK-NEXT: [[NO_OVERFLOW_DURING_ADJUSTMENT:%.*]] = icmp uge i64 [[ADJUSTED]], [[BASE_INT]]
[InstSimplify] simplifyUnsignedRangeCheck(): if we know that X != 0, handle more cases (PR43246) Summary: This is motivated by D67122 sanitizer check enhancement. That patch seemingly worsens `-fsanitize=pointer-overflow` overhead from 25% to 50%, which strongly implies missing folds. In this particular case, given ``` char* test(char& base, unsigned long offset) { return &base + offset; } ``` it will end up producing something like https://godbolt.org/z/LK5-iH which after optimizations reduces down to roughly ``` define i1 @t0(i8* nonnull %base, i64 %offset) { %base_int = ptrtoint i8* %base to i64 %adjusted = add i64 %base_int, %offset %non_null_after_adjustment = icmp ne i64 %adjusted, 0 %no_overflow_during_adjustment = icmp uge i64 %adjusted, %base_int %res = and i1 %non_null_after_adjustment, %no_overflow_during_adjustment ret i1 %res } ``` Without D67122 there was no `%non_null_after_adjustment`, and in this particular case we can get rid of the overhead: Here we add some offset to a non-null pointer, and check that the result does not overflow and is not a null pointer. But since the base pointer is already non-null, and we check for overflow, that overflow check will already catch the null pointer, so the separate null check is redundant and can be dropped. Alive proofs: https://rise4fun.com/Alive/WRzq There are more patterns of "unsigned-add-with-overflow", they are not handled here, but this is the main pattern, that we currently consider canonical, so it makes sense to handle it. https://bugs.llvm.org/show_bug.cgi?id=43246 Reviewers: spatel, nikic, vsk Reviewed By: spatel Subscribers: hiraditya, llvm-commits, reames Tags: #llvm Differential Revision: https://reviews.llvm.org/D67332 llvm-svn: 371349
2019-09-09 04:14:15 +08:00
; CHECK-NEXT: ret i1 [[NO_OVERFLOW_DURING_ADJUSTMENT]]
[NFC][InstSimplify] Some tests for dropping null check after uadd.with.overflow of non-null (PR43246) https://rise4fun.com/Alive/WRzq Name: C <= Y && Y != 0 --> C <= Y iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = and i1 %y_is_nonnull, %no_overflow => %r = %no_overflow Name: C <= Y || Y != 0 --> Y != 0 iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = or i1 %y_is_nonnull, %no_overflow => %r = %y_is_nonnull Name: C > Y || Y == 0 --> C > Y iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = or i1 %y_is_null, %overflow => %r = %overflow Name: C > Y && Y == 0 --> Y == 0 iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = and i1 %y_is_null, %overflow => %r = %y_is_null https://bugs.llvm.org/show_bug.cgi?id=43246 llvm-svn: 371339
2019-09-09 01:50:40 +08:00
;
%base_int = ptrtoint i8* %base to i64
%adjusted = add i64 %base_int, %offset
%non_null_after_adjustment = icmp ne i64 %adjusted, 0
%no_overflow_during_adjustment = icmp uge i64 %adjusted, %base_int
%res = and i1 %no_overflow_during_adjustment, %non_null_after_adjustment ; swapped
ret i1 %res
}
define i1 @t3(i8* nonnull %base, i64 %offset) {
; CHECK-LABEL: @t3(
; CHECK-NEXT: [[BASE_INT:%.*]] = ptrtoint i8* [[BASE:%.*]] to i64
; CHECK-NEXT: [[ADJUSTED:%.*]] = add i64 [[BASE_INT]], [[OFFSET:%.*]]
; CHECK-NEXT: [[NO_OVERFLOW_DURING_ADJUSTMENT:%.*]] = icmp ule i64 [[BASE_INT]], [[ADJUSTED]]
[InstSimplify] simplifyUnsignedRangeCheck(): if we know that X != 0, handle more cases (PR43246) Summary: This is motivated by D67122 sanitizer check enhancement. That patch seemingly worsens `-fsanitize=pointer-overflow` overhead from 25% to 50%, which strongly implies missing folds. In this particular case, given ``` char* test(char& base, unsigned long offset) { return &base + offset; } ``` it will end up producing something like https://godbolt.org/z/LK5-iH which after optimizations reduces down to roughly ``` define i1 @t0(i8* nonnull %base, i64 %offset) { %base_int = ptrtoint i8* %base to i64 %adjusted = add i64 %base_int, %offset %non_null_after_adjustment = icmp ne i64 %adjusted, 0 %no_overflow_during_adjustment = icmp uge i64 %adjusted, %base_int %res = and i1 %non_null_after_adjustment, %no_overflow_during_adjustment ret i1 %res } ``` Without D67122 there was no `%non_null_after_adjustment`, and in this particular case we can get rid of the overhead: Here we add some offset to a non-null pointer, and check that the result does not overflow and is not a null pointer. But since the base pointer is already non-null, and we check for overflow, that overflow check will already catch the null pointer, so the separate null check is redundant and can be dropped. Alive proofs: https://rise4fun.com/Alive/WRzq There are more patterns of "unsigned-add-with-overflow", they are not handled here, but this is the main pattern, that we currently consider canonical, so it makes sense to handle it. https://bugs.llvm.org/show_bug.cgi?id=43246 Reviewers: spatel, nikic, vsk Reviewed By: spatel Subscribers: hiraditya, llvm-commits, reames Tags: #llvm Differential Revision: https://reviews.llvm.org/D67332 llvm-svn: 371349
2019-09-09 04:14:15 +08:00
; CHECK-NEXT: ret i1 [[NO_OVERFLOW_DURING_ADJUSTMENT]]
[NFC][InstSimplify] Some tests for dropping null check after uadd.with.overflow of non-null (PR43246) https://rise4fun.com/Alive/WRzq Name: C <= Y && Y != 0 --> C <= Y iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = and i1 %y_is_nonnull, %no_overflow => %r = %no_overflow Name: C <= Y || Y != 0 --> Y != 0 iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = or i1 %y_is_nonnull, %no_overflow => %r = %y_is_nonnull Name: C > Y || Y == 0 --> C > Y iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = or i1 %y_is_null, %overflow => %r = %overflow Name: C > Y && Y == 0 --> Y == 0 iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = and i1 %y_is_null, %overflow => %r = %y_is_null https://bugs.llvm.org/show_bug.cgi?id=43246 llvm-svn: 371339
2019-09-09 01:50:40 +08:00
;
%base_int = ptrtoint i8* %base to i64
%adjusted = add i64 %base_int, %offset
%non_null_after_adjustment = icmp ne i64 %adjusted, 0
%no_overflow_during_adjustment = icmp ule i64 %base_int, %adjusted ; swapped
%res = and i1 %no_overflow_during_adjustment, %non_null_after_adjustment ; swapped
ret i1 %res
}
[InstSimplify] simplifyUnsignedRangeCheck(): if we know that X != 0, handle more cases (PR43246) Summary: This is motivated by D67122 sanitizer check enhancement. That patch seemingly worsens `-fsanitize=pointer-overflow` overhead from 25% to 50%, which strongly implies missing folds. In this particular case, given ``` char* test(char& base, unsigned long offset) { return &base + offset; } ``` it will end up producing something like https://godbolt.org/z/LK5-iH which after optimizations reduces down to roughly ``` define i1 @t0(i8* nonnull %base, i64 %offset) { %base_int = ptrtoint i8* %base to i64 %adjusted = add i64 %base_int, %offset %non_null_after_adjustment = icmp ne i64 %adjusted, 0 %no_overflow_during_adjustment = icmp uge i64 %adjusted, %base_int %res = and i1 %non_null_after_adjustment, %no_overflow_during_adjustment ret i1 %res } ``` Without D67122 there was no `%non_null_after_adjustment`, and in this particular case we can get rid of the overhead: Here we add some offset to a non-null pointer, and check that the result does not overflow and is not a null pointer. But since the base pointer is already non-null, and we check for overflow, that overflow check will already catch the null pointer, so the separate null check is redundant and can be dropped. Alive proofs: https://rise4fun.com/Alive/WRzq There are more patterns of "unsigned-add-with-overflow", they are not handled here, but this is the main pattern, that we currently consider canonical, so it makes sense to handle it. https://bugs.llvm.org/show_bug.cgi?id=43246 Reviewers: spatel, nikic, vsk Reviewed By: spatel Subscribers: hiraditya, llvm-commits, reames Tags: #llvm Differential Revision: https://reviews.llvm.org/D67332 llvm-svn: 371349
2019-09-09 04:14:15 +08:00
; If the joining operator was 'or', i.e. we check that either we produced non-null
[NFC][InstSimplify] Some tests for dropping null check after uadd.with.overflow of non-null (PR43246) https://rise4fun.com/Alive/WRzq Name: C <= Y && Y != 0 --> C <= Y iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = and i1 %y_is_nonnull, %no_overflow => %r = %no_overflow Name: C <= Y || Y != 0 --> Y != 0 iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = or i1 %y_is_nonnull, %no_overflow => %r = %y_is_nonnull Name: C > Y || Y == 0 --> C > Y iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = or i1 %y_is_null, %overflow => %r = %overflow Name: C > Y && Y == 0 --> Y == 0 iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = and i1 %y_is_null, %overflow => %r = %y_is_null https://bugs.llvm.org/show_bug.cgi?id=43246 llvm-svn: 371339
2019-09-09 01:50:40 +08:00
; pointer, or no overflow happened, then the overflow check itself is redundant.
define i1 @t4(i8* nonnull %base, i64 %offset) {
; CHECK-LABEL: @t4(
; CHECK-NEXT: [[BASE_INT:%.*]] = ptrtoint i8* [[BASE:%.*]] to i64
; CHECK-NEXT: [[ADJUSTED:%.*]] = add i64 [[BASE_INT]], [[OFFSET:%.*]]
; CHECK-NEXT: [[NON_NULL_AFTER_ADJUSTMENT:%.*]] = icmp ne i64 [[ADJUSTED]], 0
[InstSimplify] simplifyUnsignedRangeCheck(): if we know that X != 0, handle more cases (PR43246) Summary: This is motivated by D67122 sanitizer check enhancement. That patch seemingly worsens `-fsanitize=pointer-overflow` overhead from 25% to 50%, which strongly implies missing folds. In this particular case, given ``` char* test(char& base, unsigned long offset) { return &base + offset; } ``` it will end up producing something like https://godbolt.org/z/LK5-iH which after optimizations reduces down to roughly ``` define i1 @t0(i8* nonnull %base, i64 %offset) { %base_int = ptrtoint i8* %base to i64 %adjusted = add i64 %base_int, %offset %non_null_after_adjustment = icmp ne i64 %adjusted, 0 %no_overflow_during_adjustment = icmp uge i64 %adjusted, %base_int %res = and i1 %non_null_after_adjustment, %no_overflow_during_adjustment ret i1 %res } ``` Without D67122 there was no `%non_null_after_adjustment`, and in this particular case we can get rid of the overhead: Here we add some offset to a non-null pointer, and check that the result does not overflow and is not a null pointer. But since the base pointer is already non-null, and we check for overflow, that overflow check will already catch the null pointer, so the separate null check is redundant and can be dropped. Alive proofs: https://rise4fun.com/Alive/WRzq There are more patterns of "unsigned-add-with-overflow", they are not handled here, but this is the main pattern, that we currently consider canonical, so it makes sense to handle it. https://bugs.llvm.org/show_bug.cgi?id=43246 Reviewers: spatel, nikic, vsk Reviewed By: spatel Subscribers: hiraditya, llvm-commits, reames Tags: #llvm Differential Revision: https://reviews.llvm.org/D67332 llvm-svn: 371349
2019-09-09 04:14:15 +08:00
; CHECK-NEXT: ret i1 [[NON_NULL_AFTER_ADJUSTMENT]]
[NFC][InstSimplify] Some tests for dropping null check after uadd.with.overflow of non-null (PR43246) https://rise4fun.com/Alive/WRzq Name: C <= Y && Y != 0 --> C <= Y iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = and i1 %y_is_nonnull, %no_overflow => %r = %no_overflow Name: C <= Y || Y != 0 --> Y != 0 iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = or i1 %y_is_nonnull, %no_overflow => %r = %y_is_nonnull Name: C > Y || Y == 0 --> C > Y iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = or i1 %y_is_null, %overflow => %r = %overflow Name: C > Y && Y == 0 --> Y == 0 iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = and i1 %y_is_null, %overflow => %r = %y_is_null https://bugs.llvm.org/show_bug.cgi?id=43246 llvm-svn: 371339
2019-09-09 01:50:40 +08:00
;
%base_int = ptrtoint i8* %base to i64
%adjusted = add i64 %base_int, %offset
%non_null_after_adjustment = icmp ne i64 %adjusted, 0
%no_overflow_during_adjustment = icmp uge i64 %adjusted, %base_int
%res = or i1 %non_null_after_adjustment, %no_overflow_during_adjustment
ret i1 %res
}
define i1 @t5(i8* nonnull %base, i64 %offset) {
; CHECK-LABEL: @t5(
; CHECK-NEXT: [[BASE_INT:%.*]] = ptrtoint i8* [[BASE:%.*]] to i64
; CHECK-NEXT: [[ADJUSTED:%.*]] = add i64 [[BASE_INT]], [[OFFSET:%.*]]
; CHECK-NEXT: [[NON_NULL_AFTER_ADJUSTMENT:%.*]] = icmp ne i64 [[ADJUSTED]], 0
[InstSimplify] simplifyUnsignedRangeCheck(): if we know that X != 0, handle more cases (PR43246) Summary: This is motivated by D67122 sanitizer check enhancement. That patch seemingly worsens `-fsanitize=pointer-overflow` overhead from 25% to 50%, which strongly implies missing folds. In this particular case, given ``` char* test(char& base, unsigned long offset) { return &base + offset; } ``` it will end up producing something like https://godbolt.org/z/LK5-iH which after optimizations reduces down to roughly ``` define i1 @t0(i8* nonnull %base, i64 %offset) { %base_int = ptrtoint i8* %base to i64 %adjusted = add i64 %base_int, %offset %non_null_after_adjustment = icmp ne i64 %adjusted, 0 %no_overflow_during_adjustment = icmp uge i64 %adjusted, %base_int %res = and i1 %non_null_after_adjustment, %no_overflow_during_adjustment ret i1 %res } ``` Without D67122 there was no `%non_null_after_adjustment`, and in this particular case we can get rid of the overhead: Here we add some offset to a non-null pointer, and check that the result does not overflow and is not a null pointer. But since the base pointer is already non-null, and we check for overflow, that overflow check will already catch the null pointer, so the separate null check is redundant and can be dropped. Alive proofs: https://rise4fun.com/Alive/WRzq There are more patterns of "unsigned-add-with-overflow", they are not handled here, but this is the main pattern, that we currently consider canonical, so it makes sense to handle it. https://bugs.llvm.org/show_bug.cgi?id=43246 Reviewers: spatel, nikic, vsk Reviewed By: spatel Subscribers: hiraditya, llvm-commits, reames Tags: #llvm Differential Revision: https://reviews.llvm.org/D67332 llvm-svn: 371349
2019-09-09 04:14:15 +08:00
; CHECK-NEXT: ret i1 [[NON_NULL_AFTER_ADJUSTMENT]]
[NFC][InstSimplify] Some tests for dropping null check after uadd.with.overflow of non-null (PR43246) https://rise4fun.com/Alive/WRzq Name: C <= Y && Y != 0 --> C <= Y iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = and i1 %y_is_nonnull, %no_overflow => %r = %no_overflow Name: C <= Y || Y != 0 --> Y != 0 iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = or i1 %y_is_nonnull, %no_overflow => %r = %y_is_nonnull Name: C > Y || Y == 0 --> C > Y iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = or i1 %y_is_null, %overflow => %r = %overflow Name: C > Y && Y == 0 --> Y == 0 iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = and i1 %y_is_null, %overflow => %r = %y_is_null https://bugs.llvm.org/show_bug.cgi?id=43246 llvm-svn: 371339
2019-09-09 01:50:40 +08:00
;
%base_int = ptrtoint i8* %base to i64
%adjusted = add i64 %base_int, %offset
%non_null_after_adjustment = icmp ne i64 %adjusted, 0
%no_overflow_during_adjustment = icmp ule i64 %base_int, %adjusted ; swapped
%res = or i1 %non_null_after_adjustment, %no_overflow_during_adjustment
ret i1 %res
}
define i1 @t6(i8* nonnull %base, i64 %offset) {
; CHECK-LABEL: @t6(
; CHECK-NEXT: [[BASE_INT:%.*]] = ptrtoint i8* [[BASE:%.*]] to i64
; CHECK-NEXT: [[ADJUSTED:%.*]] = add i64 [[BASE_INT]], [[OFFSET:%.*]]
; CHECK-NEXT: [[NON_NULL_AFTER_ADJUSTMENT:%.*]] = icmp ne i64 [[ADJUSTED]], 0
[InstSimplify] simplifyUnsignedRangeCheck(): if we know that X != 0, handle more cases (PR43246) Summary: This is motivated by D67122 sanitizer check enhancement. That patch seemingly worsens `-fsanitize=pointer-overflow` overhead from 25% to 50%, which strongly implies missing folds. In this particular case, given ``` char* test(char& base, unsigned long offset) { return &base + offset; } ``` it will end up producing something like https://godbolt.org/z/LK5-iH which after optimizations reduces down to roughly ``` define i1 @t0(i8* nonnull %base, i64 %offset) { %base_int = ptrtoint i8* %base to i64 %adjusted = add i64 %base_int, %offset %non_null_after_adjustment = icmp ne i64 %adjusted, 0 %no_overflow_during_adjustment = icmp uge i64 %adjusted, %base_int %res = and i1 %non_null_after_adjustment, %no_overflow_during_adjustment ret i1 %res } ``` Without D67122 there was no `%non_null_after_adjustment`, and in this particular case we can get rid of the overhead: Here we add some offset to a non-null pointer, and check that the result does not overflow and is not a null pointer. But since the base pointer is already non-null, and we check for overflow, that overflow check will already catch the null pointer, so the separate null check is redundant and can be dropped. Alive proofs: https://rise4fun.com/Alive/WRzq There are more patterns of "unsigned-add-with-overflow", they are not handled here, but this is the main pattern, that we currently consider canonical, so it makes sense to handle it. https://bugs.llvm.org/show_bug.cgi?id=43246 Reviewers: spatel, nikic, vsk Reviewed By: spatel Subscribers: hiraditya, llvm-commits, reames Tags: #llvm Differential Revision: https://reviews.llvm.org/D67332 llvm-svn: 371349
2019-09-09 04:14:15 +08:00
; CHECK-NEXT: ret i1 [[NON_NULL_AFTER_ADJUSTMENT]]
[NFC][InstSimplify] Some tests for dropping null check after uadd.with.overflow of non-null (PR43246) https://rise4fun.com/Alive/WRzq Name: C <= Y && Y != 0 --> C <= Y iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = and i1 %y_is_nonnull, %no_overflow => %r = %no_overflow Name: C <= Y || Y != 0 --> Y != 0 iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = or i1 %y_is_nonnull, %no_overflow => %r = %y_is_nonnull Name: C > Y || Y == 0 --> C > Y iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = or i1 %y_is_null, %overflow => %r = %overflow Name: C > Y && Y == 0 --> Y == 0 iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = and i1 %y_is_null, %overflow => %r = %y_is_null https://bugs.llvm.org/show_bug.cgi?id=43246 llvm-svn: 371339
2019-09-09 01:50:40 +08:00
;
%base_int = ptrtoint i8* %base to i64
%adjusted = add i64 %base_int, %offset
%non_null_after_adjustment = icmp ne i64 %adjusted, 0
%no_overflow_during_adjustment = icmp uge i64 %adjusted, %base_int
%res = or i1 %no_overflow_during_adjustment, %non_null_after_adjustment ; swapped
ret i1 %res
}
define i1 @t7(i8* nonnull %base, i64 %offset) {
; CHECK-LABEL: @t7(
; CHECK-NEXT: [[BASE_INT:%.*]] = ptrtoint i8* [[BASE:%.*]] to i64
; CHECK-NEXT: [[ADJUSTED:%.*]] = add i64 [[BASE_INT]], [[OFFSET:%.*]]
; CHECK-NEXT: [[NON_NULL_AFTER_ADJUSTMENT:%.*]] = icmp ne i64 [[ADJUSTED]], 0
[InstSimplify] simplifyUnsignedRangeCheck(): if we know that X != 0, handle more cases (PR43246) Summary: This is motivated by D67122 sanitizer check enhancement. That patch seemingly worsens `-fsanitize=pointer-overflow` overhead from 25% to 50%, which strongly implies missing folds. In this particular case, given ``` char* test(char& base, unsigned long offset) { return &base + offset; } ``` it will end up producing something like https://godbolt.org/z/LK5-iH which after optimizations reduces down to roughly ``` define i1 @t0(i8* nonnull %base, i64 %offset) { %base_int = ptrtoint i8* %base to i64 %adjusted = add i64 %base_int, %offset %non_null_after_adjustment = icmp ne i64 %adjusted, 0 %no_overflow_during_adjustment = icmp uge i64 %adjusted, %base_int %res = and i1 %non_null_after_adjustment, %no_overflow_during_adjustment ret i1 %res } ``` Without D67122 there was no `%non_null_after_adjustment`, and in this particular case we can get rid of the overhead: Here we add some offset to a non-null pointer, and check that the result does not overflow and is not a null pointer. But since the base pointer is already non-null, and we check for overflow, that overflow check will already catch the null pointer, so the separate null check is redundant and can be dropped. Alive proofs: https://rise4fun.com/Alive/WRzq There are more patterns of "unsigned-add-with-overflow", they are not handled here, but this is the main pattern, that we currently consider canonical, so it makes sense to handle it. https://bugs.llvm.org/show_bug.cgi?id=43246 Reviewers: spatel, nikic, vsk Reviewed By: spatel Subscribers: hiraditya, llvm-commits, reames Tags: #llvm Differential Revision: https://reviews.llvm.org/D67332 llvm-svn: 371349
2019-09-09 04:14:15 +08:00
; CHECK-NEXT: ret i1 [[NON_NULL_AFTER_ADJUSTMENT]]
[NFC][InstSimplify] Some tests for dropping null check after uadd.with.overflow of non-null (PR43246) https://rise4fun.com/Alive/WRzq Name: C <= Y && Y != 0 --> C <= Y iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = and i1 %y_is_nonnull, %no_overflow => %r = %no_overflow Name: C <= Y || Y != 0 --> Y != 0 iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = or i1 %y_is_nonnull, %no_overflow => %r = %y_is_nonnull Name: C > Y || Y == 0 --> C > Y iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = or i1 %y_is_null, %overflow => %r = %overflow Name: C > Y && Y == 0 --> Y == 0 iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = and i1 %y_is_null, %overflow => %r = %y_is_null https://bugs.llvm.org/show_bug.cgi?id=43246 llvm-svn: 371339
2019-09-09 01:50:40 +08:00
;
%base_int = ptrtoint i8* %base to i64
%adjusted = add i64 %base_int, %offset
%non_null_after_adjustment = icmp ne i64 %adjusted, 0
%no_overflow_during_adjustment = icmp ule i64 %base_int, %adjusted ; swapped
%res = or i1 %no_overflow_during_adjustment, %non_null_after_adjustment ; swapped
ret i1 %res
}
; Or, we could be checking the reverse condition, that we either get null pointer,
; or overflow happens, then again, the standalone null check is redundant and
; can be dropped.
define i1 @t8(i8* nonnull %base, i64 %offset) {
; CHECK-LABEL: @t8(
; CHECK-NEXT: [[BASE_INT:%.*]] = ptrtoint i8* [[BASE:%.*]] to i64
; CHECK-NEXT: [[ADJUSTED:%.*]] = add i64 [[BASE_INT]], [[OFFSET:%.*]]
; CHECK-NEXT: [[NO_OVERFLOW_DURING_ADJUSTMENT:%.*]] = icmp ult i64 [[ADJUSTED]], [[BASE_INT]]
[InstSimplify] simplifyUnsignedRangeCheck(): if we know that X != 0, handle more cases (PR43246) Summary: This is motivated by D67122 sanitizer check enhancement. That patch seemingly worsens `-fsanitize=pointer-overflow` overhead from 25% to 50%, which strongly implies missing folds. In this particular case, given ``` char* test(char& base, unsigned long offset) { return &base + offset; } ``` it will end up producing something like https://godbolt.org/z/LK5-iH which after optimizations reduces down to roughly ``` define i1 @t0(i8* nonnull %base, i64 %offset) { %base_int = ptrtoint i8* %base to i64 %adjusted = add i64 %base_int, %offset %non_null_after_adjustment = icmp ne i64 %adjusted, 0 %no_overflow_during_adjustment = icmp uge i64 %adjusted, %base_int %res = and i1 %non_null_after_adjustment, %no_overflow_during_adjustment ret i1 %res } ``` Without D67122 there was no `%non_null_after_adjustment`, and in this particular case we can get rid of the overhead: Here we add some offset to a non-null pointer, and check that the result does not overflow and is not a null pointer. But since the base pointer is already non-null, and we check for overflow, that overflow check will already catch the null pointer, so the separate null check is redundant and can be dropped. Alive proofs: https://rise4fun.com/Alive/WRzq There are more patterns of "unsigned-add-with-overflow", they are not handled here, but this is the main pattern, that we currently consider canonical, so it makes sense to handle it. https://bugs.llvm.org/show_bug.cgi?id=43246 Reviewers: spatel, nikic, vsk Reviewed By: spatel Subscribers: hiraditya, llvm-commits, reames Tags: #llvm Differential Revision: https://reviews.llvm.org/D67332 llvm-svn: 371349
2019-09-09 04:14:15 +08:00
; CHECK-NEXT: ret i1 [[NO_OVERFLOW_DURING_ADJUSTMENT]]
[NFC][InstSimplify] Some tests for dropping null check after uadd.with.overflow of non-null (PR43246) https://rise4fun.com/Alive/WRzq Name: C <= Y && Y != 0 --> C <= Y iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = and i1 %y_is_nonnull, %no_overflow => %r = %no_overflow Name: C <= Y || Y != 0 --> Y != 0 iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = or i1 %y_is_nonnull, %no_overflow => %r = %y_is_nonnull Name: C > Y || Y == 0 --> C > Y iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = or i1 %y_is_null, %overflow => %r = %overflow Name: C > Y && Y == 0 --> Y == 0 iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = and i1 %y_is_null, %overflow => %r = %y_is_null https://bugs.llvm.org/show_bug.cgi?id=43246 llvm-svn: 371339
2019-09-09 01:50:40 +08:00
;
%base_int = ptrtoint i8* %base to i64
%adjusted = add i64 %base_int, %offset
%non_null_after_adjustment = icmp eq i64 %adjusted, 0
%no_overflow_during_adjustment = icmp ult i64 %adjusted, %base_int
%res = or i1 %non_null_after_adjustment, %no_overflow_during_adjustment
ret i1 %res
}
define i1 @t9(i8* nonnull %base, i64 %offset) {
; CHECK-LABEL: @t9(
; CHECK-NEXT: [[BASE_INT:%.*]] = ptrtoint i8* [[BASE:%.*]] to i64
; CHECK-NEXT: [[ADJUSTED:%.*]] = add i64 [[BASE_INT]], [[OFFSET:%.*]]
; CHECK-NEXT: [[NO_OVERFLOW_DURING_ADJUSTMENT:%.*]] = icmp ugt i64 [[BASE_INT]], [[ADJUSTED]]
[InstSimplify] simplifyUnsignedRangeCheck(): if we know that X != 0, handle more cases (PR43246) Summary: This is motivated by D67122 sanitizer check enhancement. That patch seemingly worsens `-fsanitize=pointer-overflow` overhead from 25% to 50%, which strongly implies missing folds. In this particular case, given ``` char* test(char& base, unsigned long offset) { return &base + offset; } ``` it will end up producing something like https://godbolt.org/z/LK5-iH which after optimizations reduces down to roughly ``` define i1 @t0(i8* nonnull %base, i64 %offset) { %base_int = ptrtoint i8* %base to i64 %adjusted = add i64 %base_int, %offset %non_null_after_adjustment = icmp ne i64 %adjusted, 0 %no_overflow_during_adjustment = icmp uge i64 %adjusted, %base_int %res = and i1 %non_null_after_adjustment, %no_overflow_during_adjustment ret i1 %res } ``` Without D67122 there was no `%non_null_after_adjustment`, and in this particular case we can get rid of the overhead: Here we add some offset to a non-null pointer, and check that the result does not overflow and is not a null pointer. But since the base pointer is already non-null, and we check for overflow, that overflow check will already catch the null pointer, so the separate null check is redundant and can be dropped. Alive proofs: https://rise4fun.com/Alive/WRzq There are more patterns of "unsigned-add-with-overflow", they are not handled here, but this is the main pattern, that we currently consider canonical, so it makes sense to handle it. https://bugs.llvm.org/show_bug.cgi?id=43246 Reviewers: spatel, nikic, vsk Reviewed By: spatel Subscribers: hiraditya, llvm-commits, reames Tags: #llvm Differential Revision: https://reviews.llvm.org/D67332 llvm-svn: 371349
2019-09-09 04:14:15 +08:00
; CHECK-NEXT: ret i1 [[NO_OVERFLOW_DURING_ADJUSTMENT]]
[NFC][InstSimplify] Some tests for dropping null check after uadd.with.overflow of non-null (PR43246) https://rise4fun.com/Alive/WRzq Name: C <= Y && Y != 0 --> C <= Y iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = and i1 %y_is_nonnull, %no_overflow => %r = %no_overflow Name: C <= Y || Y != 0 --> Y != 0 iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = or i1 %y_is_nonnull, %no_overflow => %r = %y_is_nonnull Name: C > Y || Y == 0 --> C > Y iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = or i1 %y_is_null, %overflow => %r = %overflow Name: C > Y && Y == 0 --> Y == 0 iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = and i1 %y_is_null, %overflow => %r = %y_is_null https://bugs.llvm.org/show_bug.cgi?id=43246 llvm-svn: 371339
2019-09-09 01:50:40 +08:00
;
%base_int = ptrtoint i8* %base to i64
%adjusted = add i64 %base_int, %offset
%non_null_after_adjustment = icmp eq i64 %adjusted, 0
%no_overflow_during_adjustment = icmp ugt i64 %base_int, %adjusted ; swapped
%res = or i1 %non_null_after_adjustment, %no_overflow_during_adjustment
ret i1 %res
}
define i1 @t10(i8* nonnull %base, i64 %offset) {
; CHECK-LABEL: @t10(
; CHECK-NEXT: [[BASE_INT:%.*]] = ptrtoint i8* [[BASE:%.*]] to i64
; CHECK-NEXT: [[ADJUSTED:%.*]] = add i64 [[BASE_INT]], [[OFFSET:%.*]]
; CHECK-NEXT: [[NO_OVERFLOW_DURING_ADJUSTMENT:%.*]] = icmp ult i64 [[ADJUSTED]], [[BASE_INT]]
[InstSimplify] simplifyUnsignedRangeCheck(): if we know that X != 0, handle more cases (PR43246) Summary: This is motivated by D67122 sanitizer check enhancement. That patch seemingly worsens `-fsanitize=pointer-overflow` overhead from 25% to 50%, which strongly implies missing folds. In this particular case, given ``` char* test(char& base, unsigned long offset) { return &base + offset; } ``` it will end up producing something like https://godbolt.org/z/LK5-iH which after optimizations reduces down to roughly ``` define i1 @t0(i8* nonnull %base, i64 %offset) { %base_int = ptrtoint i8* %base to i64 %adjusted = add i64 %base_int, %offset %non_null_after_adjustment = icmp ne i64 %adjusted, 0 %no_overflow_during_adjustment = icmp uge i64 %adjusted, %base_int %res = and i1 %non_null_after_adjustment, %no_overflow_during_adjustment ret i1 %res } ``` Without D67122 there was no `%non_null_after_adjustment`, and in this particular case we can get rid of the overhead: Here we add some offset to a non-null pointer, and check that the result does not overflow and is not a null pointer. But since the base pointer is already non-null, and we check for overflow, that overflow check will already catch the null pointer, so the separate null check is redundant and can be dropped. Alive proofs: https://rise4fun.com/Alive/WRzq There are more patterns of "unsigned-add-with-overflow", they are not handled here, but this is the main pattern, that we currently consider canonical, so it makes sense to handle it. https://bugs.llvm.org/show_bug.cgi?id=43246 Reviewers: spatel, nikic, vsk Reviewed By: spatel Subscribers: hiraditya, llvm-commits, reames Tags: #llvm Differential Revision: https://reviews.llvm.org/D67332 llvm-svn: 371349
2019-09-09 04:14:15 +08:00
; CHECK-NEXT: ret i1 [[NO_OVERFLOW_DURING_ADJUSTMENT]]
[NFC][InstSimplify] Some tests for dropping null check after uadd.with.overflow of non-null (PR43246) https://rise4fun.com/Alive/WRzq Name: C <= Y && Y != 0 --> C <= Y iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = and i1 %y_is_nonnull, %no_overflow => %r = %no_overflow Name: C <= Y || Y != 0 --> Y != 0 iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = or i1 %y_is_nonnull, %no_overflow => %r = %y_is_nonnull Name: C > Y || Y == 0 --> C > Y iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = or i1 %y_is_null, %overflow => %r = %overflow Name: C > Y && Y == 0 --> Y == 0 iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = and i1 %y_is_null, %overflow => %r = %y_is_null https://bugs.llvm.org/show_bug.cgi?id=43246 llvm-svn: 371339
2019-09-09 01:50:40 +08:00
;
%base_int = ptrtoint i8* %base to i64
%adjusted = add i64 %base_int, %offset
%non_null_after_adjustment = icmp eq i64 %adjusted, 0
%no_overflow_during_adjustment = icmp ult i64 %adjusted, %base_int
%res = or i1 %no_overflow_during_adjustment, %non_null_after_adjustment ; swapped
ret i1 %res
}
define i1 @t11(i8* nonnull %base, i64 %offset) {
; CHECK-LABEL: @t11(
; CHECK-NEXT: [[BASE_INT:%.*]] = ptrtoint i8* [[BASE:%.*]] to i64
; CHECK-NEXT: [[ADJUSTED:%.*]] = add i64 [[BASE_INT]], [[OFFSET:%.*]]
; CHECK-NEXT: [[NO_OVERFLOW_DURING_ADJUSTMENT:%.*]] = icmp ugt i64 [[BASE_INT]], [[ADJUSTED]]
[InstSimplify] simplifyUnsignedRangeCheck(): if we know that X != 0, handle more cases (PR43246) Summary: This is motivated by D67122 sanitizer check enhancement. That patch seemingly worsens `-fsanitize=pointer-overflow` overhead from 25% to 50%, which strongly implies missing folds. In this particular case, given ``` char* test(char& base, unsigned long offset) { return &base + offset; } ``` it will end up producing something like https://godbolt.org/z/LK5-iH which after optimizations reduces down to roughly ``` define i1 @t0(i8* nonnull %base, i64 %offset) { %base_int = ptrtoint i8* %base to i64 %adjusted = add i64 %base_int, %offset %non_null_after_adjustment = icmp ne i64 %adjusted, 0 %no_overflow_during_adjustment = icmp uge i64 %adjusted, %base_int %res = and i1 %non_null_after_adjustment, %no_overflow_during_adjustment ret i1 %res } ``` Without D67122 there was no `%non_null_after_adjustment`, and in this particular case we can get rid of the overhead: Here we add some offset to a non-null pointer, and check that the result does not overflow and is not a null pointer. But since the base pointer is already non-null, and we check for overflow, that overflow check will already catch the null pointer, so the separate null check is redundant and can be dropped. Alive proofs: https://rise4fun.com/Alive/WRzq There are more patterns of "unsigned-add-with-overflow", they are not handled here, but this is the main pattern, that we currently consider canonical, so it makes sense to handle it. https://bugs.llvm.org/show_bug.cgi?id=43246 Reviewers: spatel, nikic, vsk Reviewed By: spatel Subscribers: hiraditya, llvm-commits, reames Tags: #llvm Differential Revision: https://reviews.llvm.org/D67332 llvm-svn: 371349
2019-09-09 04:14:15 +08:00
; CHECK-NEXT: ret i1 [[NO_OVERFLOW_DURING_ADJUSTMENT]]
[NFC][InstSimplify] Some tests for dropping null check after uadd.with.overflow of non-null (PR43246) https://rise4fun.com/Alive/WRzq Name: C <= Y && Y != 0 --> C <= Y iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = and i1 %y_is_nonnull, %no_overflow => %r = %no_overflow Name: C <= Y || Y != 0 --> Y != 0 iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = or i1 %y_is_nonnull, %no_overflow => %r = %y_is_nonnull Name: C > Y || Y == 0 --> C > Y iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = or i1 %y_is_null, %overflow => %r = %overflow Name: C > Y && Y == 0 --> Y == 0 iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = and i1 %y_is_null, %overflow => %r = %y_is_null https://bugs.llvm.org/show_bug.cgi?id=43246 llvm-svn: 371339
2019-09-09 01:50:40 +08:00
;
%base_int = ptrtoint i8* %base to i64
%adjusted = add i64 %base_int, %offset
%non_null_after_adjustment = icmp eq i64 %adjusted, 0
%no_overflow_during_adjustment = icmp ugt i64 %base_int, %adjusted ; swapped
%res = or i1 %no_overflow_during_adjustment, %non_null_after_adjustment ; swapped
ret i1 %res
}
; If the joining operator was 'and', i.e. we check that we both get null pointer
; AND overflow happens, then the overflow check is redundant.
define i1 @t12(i8* nonnull %base, i64 %offset) {
; CHECK-LABEL: @t12(
; CHECK-NEXT: [[BASE_INT:%.*]] = ptrtoint i8* [[BASE:%.*]] to i64
; CHECK-NEXT: [[ADJUSTED:%.*]] = add i64 [[BASE_INT]], [[OFFSET:%.*]]
; CHECK-NEXT: [[NON_NULL_AFTER_ADJUSTMENT:%.*]] = icmp eq i64 [[ADJUSTED]], 0
[InstSimplify] simplifyUnsignedRangeCheck(): if we know that X != 0, handle more cases (PR43246) Summary: This is motivated by D67122 sanitizer check enhancement. That patch seemingly worsens `-fsanitize=pointer-overflow` overhead from 25% to 50%, which strongly implies missing folds. In this particular case, given ``` char* test(char& base, unsigned long offset) { return &base + offset; } ``` it will end up producing something like https://godbolt.org/z/LK5-iH which after optimizations reduces down to roughly ``` define i1 @t0(i8* nonnull %base, i64 %offset) { %base_int = ptrtoint i8* %base to i64 %adjusted = add i64 %base_int, %offset %non_null_after_adjustment = icmp ne i64 %adjusted, 0 %no_overflow_during_adjustment = icmp uge i64 %adjusted, %base_int %res = and i1 %non_null_after_adjustment, %no_overflow_during_adjustment ret i1 %res } ``` Without D67122 there was no `%non_null_after_adjustment`, and in this particular case we can get rid of the overhead: Here we add some offset to a non-null pointer, and check that the result does not overflow and is not a null pointer. But since the base pointer is already non-null, and we check for overflow, that overflow check will already catch the null pointer, so the separate null check is redundant and can be dropped. Alive proofs: https://rise4fun.com/Alive/WRzq There are more patterns of "unsigned-add-with-overflow", they are not handled here, but this is the main pattern, that we currently consider canonical, so it makes sense to handle it. https://bugs.llvm.org/show_bug.cgi?id=43246 Reviewers: spatel, nikic, vsk Reviewed By: spatel Subscribers: hiraditya, llvm-commits, reames Tags: #llvm Differential Revision: https://reviews.llvm.org/D67332 llvm-svn: 371349
2019-09-09 04:14:15 +08:00
; CHECK-NEXT: ret i1 [[NON_NULL_AFTER_ADJUSTMENT]]
[NFC][InstSimplify] Some tests for dropping null check after uadd.with.overflow of non-null (PR43246) https://rise4fun.com/Alive/WRzq Name: C <= Y && Y != 0 --> C <= Y iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = and i1 %y_is_nonnull, %no_overflow => %r = %no_overflow Name: C <= Y || Y != 0 --> Y != 0 iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = or i1 %y_is_nonnull, %no_overflow => %r = %y_is_nonnull Name: C > Y || Y == 0 --> C > Y iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = or i1 %y_is_null, %overflow => %r = %overflow Name: C > Y && Y == 0 --> Y == 0 iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = and i1 %y_is_null, %overflow => %r = %y_is_null https://bugs.llvm.org/show_bug.cgi?id=43246 llvm-svn: 371339
2019-09-09 01:50:40 +08:00
;
%base_int = ptrtoint i8* %base to i64
%adjusted = add i64 %base_int, %offset
%non_null_after_adjustment = icmp eq i64 %adjusted, 0
%no_overflow_during_adjustment = icmp ult i64 %adjusted, %base_int
%res = and i1 %non_null_after_adjustment, %no_overflow_during_adjustment
ret i1 %res
}
define i1 @t13(i8* nonnull %base, i64 %offset) {
; CHECK-LABEL: @t13(
; CHECK-NEXT: [[BASE_INT:%.*]] = ptrtoint i8* [[BASE:%.*]] to i64
; CHECK-NEXT: [[ADJUSTED:%.*]] = add i64 [[BASE_INT]], [[OFFSET:%.*]]
; CHECK-NEXT: [[NON_NULL_AFTER_ADJUSTMENT:%.*]] = icmp eq i64 [[ADJUSTED]], 0
[InstSimplify] simplifyUnsignedRangeCheck(): if we know that X != 0, handle more cases (PR43246) Summary: This is motivated by D67122 sanitizer check enhancement. That patch seemingly worsens `-fsanitize=pointer-overflow` overhead from 25% to 50%, which strongly implies missing folds. In this particular case, given ``` char* test(char& base, unsigned long offset) { return &base + offset; } ``` it will end up producing something like https://godbolt.org/z/LK5-iH which after optimizations reduces down to roughly ``` define i1 @t0(i8* nonnull %base, i64 %offset) { %base_int = ptrtoint i8* %base to i64 %adjusted = add i64 %base_int, %offset %non_null_after_adjustment = icmp ne i64 %adjusted, 0 %no_overflow_during_adjustment = icmp uge i64 %adjusted, %base_int %res = and i1 %non_null_after_adjustment, %no_overflow_during_adjustment ret i1 %res } ``` Without D67122 there was no `%non_null_after_adjustment`, and in this particular case we can get rid of the overhead: Here we add some offset to a non-null pointer, and check that the result does not overflow and is not a null pointer. But since the base pointer is already non-null, and we check for overflow, that overflow check will already catch the null pointer, so the separate null check is redundant and can be dropped. Alive proofs: https://rise4fun.com/Alive/WRzq There are more patterns of "unsigned-add-with-overflow", they are not handled here, but this is the main pattern, that we currently consider canonical, so it makes sense to handle it. https://bugs.llvm.org/show_bug.cgi?id=43246 Reviewers: spatel, nikic, vsk Reviewed By: spatel Subscribers: hiraditya, llvm-commits, reames Tags: #llvm Differential Revision: https://reviews.llvm.org/D67332 llvm-svn: 371349
2019-09-09 04:14:15 +08:00
; CHECK-NEXT: ret i1 [[NON_NULL_AFTER_ADJUSTMENT]]
[NFC][InstSimplify] Some tests for dropping null check after uadd.with.overflow of non-null (PR43246) https://rise4fun.com/Alive/WRzq Name: C <= Y && Y != 0 --> C <= Y iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = and i1 %y_is_nonnull, %no_overflow => %r = %no_overflow Name: C <= Y || Y != 0 --> Y != 0 iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = or i1 %y_is_nonnull, %no_overflow => %r = %y_is_nonnull Name: C > Y || Y == 0 --> C > Y iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = or i1 %y_is_null, %overflow => %r = %overflow Name: C > Y && Y == 0 --> Y == 0 iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = and i1 %y_is_null, %overflow => %r = %y_is_null https://bugs.llvm.org/show_bug.cgi?id=43246 llvm-svn: 371339
2019-09-09 01:50:40 +08:00
;
%base_int = ptrtoint i8* %base to i64
%adjusted = add i64 %base_int, %offset
%non_null_after_adjustment = icmp eq i64 %adjusted, 0
%no_overflow_during_adjustment = icmp ugt i64 %base_int, %adjusted ; swapped
%res = and i1 %non_null_after_adjustment, %no_overflow_during_adjustment
ret i1 %res
}
define i1 @t14(i8* nonnull %base, i64 %offset) {
; CHECK-LABEL: @t14(
; CHECK-NEXT: [[BASE_INT:%.*]] = ptrtoint i8* [[BASE:%.*]] to i64
; CHECK-NEXT: [[ADJUSTED:%.*]] = add i64 [[BASE_INT]], [[OFFSET:%.*]]
; CHECK-NEXT: [[NON_NULL_AFTER_ADJUSTMENT:%.*]] = icmp eq i64 [[ADJUSTED]], 0
[InstSimplify] simplifyUnsignedRangeCheck(): if we know that X != 0, handle more cases (PR43246) Summary: This is motivated by D67122 sanitizer check enhancement. That patch seemingly worsens `-fsanitize=pointer-overflow` overhead from 25% to 50%, which strongly implies missing folds. In this particular case, given ``` char* test(char& base, unsigned long offset) { return &base + offset; } ``` it will end up producing something like https://godbolt.org/z/LK5-iH which after optimizations reduces down to roughly ``` define i1 @t0(i8* nonnull %base, i64 %offset) { %base_int = ptrtoint i8* %base to i64 %adjusted = add i64 %base_int, %offset %non_null_after_adjustment = icmp ne i64 %adjusted, 0 %no_overflow_during_adjustment = icmp uge i64 %adjusted, %base_int %res = and i1 %non_null_after_adjustment, %no_overflow_during_adjustment ret i1 %res } ``` Without D67122 there was no `%non_null_after_adjustment`, and in this particular case we can get rid of the overhead: Here we add some offset to a non-null pointer, and check that the result does not overflow and is not a null pointer. But since the base pointer is already non-null, and we check for overflow, that overflow check will already catch the null pointer, so the separate null check is redundant and can be dropped. Alive proofs: https://rise4fun.com/Alive/WRzq There are more patterns of "unsigned-add-with-overflow", they are not handled here, but this is the main pattern, that we currently consider canonical, so it makes sense to handle it. https://bugs.llvm.org/show_bug.cgi?id=43246 Reviewers: spatel, nikic, vsk Reviewed By: spatel Subscribers: hiraditya, llvm-commits, reames Tags: #llvm Differential Revision: https://reviews.llvm.org/D67332 llvm-svn: 371349
2019-09-09 04:14:15 +08:00
; CHECK-NEXT: ret i1 [[NON_NULL_AFTER_ADJUSTMENT]]
[NFC][InstSimplify] Some tests for dropping null check after uadd.with.overflow of non-null (PR43246) https://rise4fun.com/Alive/WRzq Name: C <= Y && Y != 0 --> C <= Y iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = and i1 %y_is_nonnull, %no_overflow => %r = %no_overflow Name: C <= Y || Y != 0 --> Y != 0 iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = or i1 %y_is_nonnull, %no_overflow => %r = %y_is_nonnull Name: C > Y || Y == 0 --> C > Y iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = or i1 %y_is_null, %overflow => %r = %overflow Name: C > Y && Y == 0 --> Y == 0 iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = and i1 %y_is_null, %overflow => %r = %y_is_null https://bugs.llvm.org/show_bug.cgi?id=43246 llvm-svn: 371339
2019-09-09 01:50:40 +08:00
;
%base_int = ptrtoint i8* %base to i64
%adjusted = add i64 %base_int, %offset
%non_null_after_adjustment = icmp eq i64 %adjusted, 0
%no_overflow_during_adjustment = icmp ult i64 %adjusted, %base_int
%res = and i1 %no_overflow_during_adjustment, %non_null_after_adjustment ; swapped
ret i1 %res
}
define i1 @t15(i8* nonnull %base, i64 %offset) {
; CHECK-LABEL: @t15(
; CHECK-NEXT: [[BASE_INT:%.*]] = ptrtoint i8* [[BASE:%.*]] to i64
; CHECK-NEXT: [[ADJUSTED:%.*]] = add i64 [[BASE_INT]], [[OFFSET:%.*]]
; CHECK-NEXT: [[NON_NULL_AFTER_ADJUSTMENT:%.*]] = icmp eq i64 [[ADJUSTED]], 0
[InstSimplify] simplifyUnsignedRangeCheck(): if we know that X != 0, handle more cases (PR43246) Summary: This is motivated by D67122 sanitizer check enhancement. That patch seemingly worsens `-fsanitize=pointer-overflow` overhead from 25% to 50%, which strongly implies missing folds. In this particular case, given ``` char* test(char& base, unsigned long offset) { return &base + offset; } ``` it will end up producing something like https://godbolt.org/z/LK5-iH which after optimizations reduces down to roughly ``` define i1 @t0(i8* nonnull %base, i64 %offset) { %base_int = ptrtoint i8* %base to i64 %adjusted = add i64 %base_int, %offset %non_null_after_adjustment = icmp ne i64 %adjusted, 0 %no_overflow_during_adjustment = icmp uge i64 %adjusted, %base_int %res = and i1 %non_null_after_adjustment, %no_overflow_during_adjustment ret i1 %res } ``` Without D67122 there was no `%non_null_after_adjustment`, and in this particular case we can get rid of the overhead: Here we add some offset to a non-null pointer, and check that the result does not overflow and is not a null pointer. But since the base pointer is already non-null, and we check for overflow, that overflow check will already catch the null pointer, so the separate null check is redundant and can be dropped. Alive proofs: https://rise4fun.com/Alive/WRzq There are more patterns of "unsigned-add-with-overflow", they are not handled here, but this is the main pattern, that we currently consider canonical, so it makes sense to handle it. https://bugs.llvm.org/show_bug.cgi?id=43246 Reviewers: spatel, nikic, vsk Reviewed By: spatel Subscribers: hiraditya, llvm-commits, reames Tags: #llvm Differential Revision: https://reviews.llvm.org/D67332 llvm-svn: 371349
2019-09-09 04:14:15 +08:00
; CHECK-NEXT: ret i1 [[NON_NULL_AFTER_ADJUSTMENT]]
[NFC][InstSimplify] Some tests for dropping null check after uadd.with.overflow of non-null (PR43246) https://rise4fun.com/Alive/WRzq Name: C <= Y && Y != 0 --> C <= Y iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = and i1 %y_is_nonnull, %no_overflow => %r = %no_overflow Name: C <= Y || Y != 0 --> Y != 0 iff C != 0 Pre: C != 0 %y_is_nonnull = icmp ne i64 %y, 0 %no_overflow = icmp ule i64 C, %y %r = or i1 %y_is_nonnull, %no_overflow => %r = %y_is_nonnull Name: C > Y || Y == 0 --> C > Y iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = or i1 %y_is_null, %overflow => %r = %overflow Name: C > Y && Y == 0 --> Y == 0 iff C != 0 Pre: C != 0 %y_is_null = icmp eq i64 %y, 0 %overflow = icmp ugt i64 C, %y %r = and i1 %y_is_null, %overflow => %r = %y_is_null https://bugs.llvm.org/show_bug.cgi?id=43246 llvm-svn: 371339
2019-09-09 01:50:40 +08:00
;
%base_int = ptrtoint i8* %base to i64
%adjusted = add i64 %base_int, %offset
%non_null_after_adjustment = icmp eq i64 %adjusted, 0
%no_overflow_during_adjustment = icmp ugt i64 %base_int, %adjusted ; swapped
%res = and i1 %no_overflow_during_adjustment, %non_null_after_adjustment ; swapped
ret i1 %res
}
[NFC][InstSimplify] Add extra test for D67411 with @llvm.assume llvm-svn: 371629
2019-09-11 23:28:03 +08:00
declare void @llvm.assume(i1)
define i1 @t16(i64 %base, i64 %offset) {
; CHECK-LABEL: @t16(
; CHECK-NEXT: [[CMP:%.*]] = icmp slt i64 [[BASE:%.*]], 0
; CHECK-NEXT: call void @llvm.assume(i1 [[CMP]])
; CHECK-NEXT: [[ADJUSTED:%.*]] = add i64 [[BASE]], [[OFFSET:%.*]]
; CHECK-NEXT: [[NON_NULL_AFTER_ADJUSTMENT:%.*]] = icmp ne i64 [[ADJUSTED]], 0
; CHECK-NEXT: [[NO_OVERFLOW_DURING_ADJUSTMENT:%.*]] = icmp uge i64 [[ADJUSTED]], [[BASE]]
; CHECK-NEXT: [[RES:%.*]] = and i1 [[NON_NULL_AFTER_ADJUSTMENT]], [[NO_OVERFLOW_DURING_ADJUSTMENT]]
; CHECK-NEXT: ret i1 [[RES]]
;
%cmp = icmp slt i64 %base, 0
call void @llvm.assume(i1 %cmp)
%adjusted = add i64 %base, %offset
%non_null_after_adjustment = icmp ne i64 %adjusted, 0
%no_overflow_during_adjustment = icmp uge i64 %adjusted, %base
%res = and i1 %non_null_after_adjustment, %no_overflow_during_adjustment
ret i1 %res
}