
This patch canonicalizes getelementptr instructions with constant indices to use the `i8` source element type. This makes it easier for optimizations to recognize that two GEPs are identical, because they don't need to see past many different ways to express the same offset. This is a first step towards https://discourse.llvm.org/t/rfc-replacing-getelementptr-with-ptradd/68699. This is limited to constant GEPs only for now, as they have a clear canonical form, while we're not yet sure how exactly to deal with variable indices. The test llvm/test/Transforms/PhaseOrdering/switch_with_geps.ll gives two representative examples of the kind of optimization improvement we expect from this change. In the first test SimplifyCFG can now realize that all switch branches are actually the same. In the second test it can convert it into simple arithmetic. These are representative of common optimization failures we see in Rust. Fixes https://github.com/llvm/llvm-project/issues/69841.
166 lines
6.8 KiB
LLVM
166 lines
6.8 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
|
|
; Verify that calls to memcmp with arrays of elements larger than char
|
|
; are folded correctly.
|
|
; RUN: opt < %s -passes=instcombine -S -data-layout="E" | FileCheck %s --check-prefixes=BE
|
|
; RUN: opt < %s -passes=instcombine -S -data-layout="e" | FileCheck %s --check-prefixes=LE
|
|
|
|
declare i32 @memcmp(ptr, ptr, i64)
|
|
|
|
; BE representation: { 'a', 'b', 'c', ..., 'f', 'g', 'h' }
|
|
; LE representation: { 'b', 'a', 'd', ..., 'e', 'h', 'g' }
|
|
@ia16a = constant [4 x i16] [i16 24930, i16 25444, i16 25958, i16 26472]
|
|
|
|
; Same as the BE representation above except ending in "gg".
|
|
@i8a = constant [8 x i8] c"abcdefgg"
|
|
|
|
; Fold memcmp(ia16a, i8a, N) for N in [0, 8].
|
|
|
|
define void @fold_memcmp_ia16a_i8a(ptr %pcmp) {
|
|
; BE-LABEL: @fold_memcmp_ia16a_i8a(
|
|
; BE-NEXT: store i32 0, ptr [[PCMP:%.*]], align 4
|
|
; BE-NEXT: [[PSTOR1:%.*]] = getelementptr i8, ptr [[PCMP]], i64 4
|
|
; BE-NEXT: store i32 0, ptr [[PSTOR1]], align 4
|
|
; BE-NEXT: [[PSTOR2:%.*]] = getelementptr i8, ptr [[PCMP]], i64 8
|
|
; BE-NEXT: store i32 0, ptr [[PSTOR2]], align 4
|
|
; BE-NEXT: [[PSTOR3:%.*]] = getelementptr i8, ptr [[PCMP]], i64 12
|
|
; BE-NEXT: store i32 0, ptr [[PSTOR3]], align 4
|
|
; BE-NEXT: [[PSTOR4:%.*]] = getelementptr i8, ptr [[PCMP]], i64 16
|
|
; BE-NEXT: store i32 0, ptr [[PSTOR4]], align 4
|
|
; BE-NEXT: [[PSTOR5:%.*]] = getelementptr i8, ptr [[PCMP]], i64 20
|
|
; BE-NEXT: store i32 0, ptr [[PSTOR5]], align 4
|
|
; BE-NEXT: [[PSTOR6:%.*]] = getelementptr i8, ptr [[PCMP]], i64 24
|
|
; BE-NEXT: store i32 0, ptr [[PSTOR6]], align 4
|
|
; BE-NEXT: [[PSTOR7:%.*]] = getelementptr i8, ptr [[PCMP]], i64 28
|
|
; BE-NEXT: store i32 0, ptr [[PSTOR7]], align 4
|
|
; BE-NEXT: [[PSTOR8:%.*]] = getelementptr i8, ptr [[PCMP]], i64 32
|
|
; BE-NEXT: store i32 1, ptr [[PSTOR8]], align 4
|
|
; BE-NEXT: ret void
|
|
;
|
|
; LE-LABEL: @fold_memcmp_ia16a_i8a(
|
|
; LE-NEXT: store i32 0, ptr [[PCMP:%.*]], align 4
|
|
; LE-NEXT: [[PSTOR1:%.*]] = getelementptr i8, ptr [[PCMP]], i64 4
|
|
; LE-NEXT: store i32 1, ptr [[PSTOR1]], align 4
|
|
; LE-NEXT: [[PSTOR2:%.*]] = getelementptr i8, ptr [[PCMP]], i64 8
|
|
; LE-NEXT: store i32 1, ptr [[PSTOR2]], align 4
|
|
; LE-NEXT: [[PSTOR3:%.*]] = getelementptr i8, ptr [[PCMP]], i64 12
|
|
; LE-NEXT: store i32 1, ptr [[PSTOR3]], align 4
|
|
; LE-NEXT: [[PSTOR4:%.*]] = getelementptr i8, ptr [[PCMP]], i64 16
|
|
; LE-NEXT: store i32 1, ptr [[PSTOR4]], align 4
|
|
; LE-NEXT: [[PSTOR5:%.*]] = getelementptr i8, ptr [[PCMP]], i64 20
|
|
; LE-NEXT: store i32 1, ptr [[PSTOR5]], align 4
|
|
; LE-NEXT: [[PSTOR6:%.*]] = getelementptr i8, ptr [[PCMP]], i64 24
|
|
; LE-NEXT: store i32 1, ptr [[PSTOR6]], align 4
|
|
; LE-NEXT: [[PSTOR7:%.*]] = getelementptr i8, ptr [[PCMP]], i64 28
|
|
; LE-NEXT: store i32 1, ptr [[PSTOR7]], align 4
|
|
; LE-NEXT: [[PSTOR8:%.*]] = getelementptr i8, ptr [[PCMP]], i64 32
|
|
; LE-NEXT: store i32 1, ptr [[PSTOR8]], align 4
|
|
; LE-NEXT: ret void
|
|
;
|
|
|
|
%cmp0 = call i32 @memcmp(ptr @ia16a, ptr @i8a, i64 0)
|
|
store i32 %cmp0, ptr %pcmp
|
|
|
|
%cmp1 = call i32 @memcmp(ptr @ia16a, ptr @i8a, i64 1)
|
|
%pstor1 = getelementptr i32, ptr %pcmp, i64 1
|
|
store i32 %cmp1, ptr %pstor1
|
|
|
|
%cmp2 = call i32 @memcmp(ptr @ia16a, ptr @i8a, i64 2)
|
|
%pstor2 = getelementptr i32, ptr %pcmp, i64 2
|
|
store i32 %cmp2, ptr %pstor2
|
|
|
|
%cmp3 = call i32 @memcmp(ptr @ia16a, ptr @i8a, i64 3)
|
|
%pstor3 = getelementptr i32, ptr %pcmp, i64 3
|
|
store i32 %cmp3, ptr %pstor3
|
|
|
|
%cmp4 = call i32 @memcmp(ptr @ia16a, ptr @i8a, i64 4)
|
|
%pstor4 = getelementptr i32, ptr %pcmp, i64 4
|
|
store i32 %cmp4, ptr %pstor4
|
|
|
|
%cmp5 = call i32 @memcmp(ptr @ia16a, ptr @i8a, i64 5)
|
|
%pstor5 = getelementptr i32, ptr %pcmp, i64 5
|
|
store i32 %cmp5, ptr %pstor5
|
|
|
|
%cmp6 = call i32 @memcmp(ptr @ia16a, ptr @i8a, i64 6)
|
|
%pstor6 = getelementptr i32, ptr %pcmp, i64 6
|
|
store i32 %cmp6, ptr %pstor6
|
|
|
|
%cmp7 = call i32 @memcmp(ptr @ia16a, ptr @i8a, i64 7)
|
|
%pstor7 = getelementptr i32, ptr %pcmp, i64 7
|
|
store i32 %cmp7, ptr %pstor7
|
|
|
|
%cmp8 = call i32 @memcmp(ptr @ia16a, ptr @i8a, i64 8)
|
|
%pstor8 = getelementptr i32, ptr %pcmp, i64 8
|
|
store i32 %cmp8, ptr %pstor8
|
|
|
|
ret void
|
|
}
|
|
|
|
|
|
; Fold memcmp(ia16a + 1, i8a + 2, N) for N in [0, 6].
|
|
|
|
define void @fold_memcmp_ia16a_p1_i8a_p1(ptr %pcmp) {
|
|
; BE-LABEL: @fold_memcmp_ia16a_p1_i8a_p1(
|
|
; BE-NEXT: store i32 0, ptr [[PCMP:%.*]], align 4
|
|
; BE-NEXT: [[PSTOR1:%.*]] = getelementptr i8, ptr [[PCMP]], i64 4
|
|
; BE-NEXT: store i32 1, ptr [[PSTOR1]], align 4
|
|
; BE-NEXT: [[PSTOR2:%.*]] = getelementptr i8, ptr [[PCMP]], i64 8
|
|
; BE-NEXT: store i32 1, ptr [[PSTOR2]], align 4
|
|
; BE-NEXT: [[PSTOR3:%.*]] = getelementptr i8, ptr [[PCMP]], i64 12
|
|
; BE-NEXT: store i32 1, ptr [[PSTOR3]], align 4
|
|
; BE-NEXT: [[PSTOR4:%.*]] = getelementptr i8, ptr [[PCMP]], i64 16
|
|
; BE-NEXT: store i32 1, ptr [[PSTOR4]], align 4
|
|
; BE-NEXT: [[PSTOR5:%.*]] = getelementptr i8, ptr [[PCMP]], i64 20
|
|
; BE-NEXT: store i32 1, ptr [[PSTOR5]], align 4
|
|
; BE-NEXT: [[PSTOR6:%.*]] = getelementptr i8, ptr [[PCMP]], i64 24
|
|
; BE-NEXT: store i32 1, ptr [[PSTOR6]], align 4
|
|
; BE-NEXT: ret void
|
|
;
|
|
; LE-LABEL: @fold_memcmp_ia16a_p1_i8a_p1(
|
|
; LE-NEXT: store i32 0, ptr [[PCMP:%.*]], align 4
|
|
; LE-NEXT: [[PSTOR1:%.*]] = getelementptr i8, ptr [[PCMP]], i64 4
|
|
; LE-NEXT: store i32 1, ptr [[PSTOR1]], align 4
|
|
; LE-NEXT: [[PSTOR2:%.*]] = getelementptr i8, ptr [[PCMP]], i64 8
|
|
; LE-NEXT: store i32 1, ptr [[PSTOR2]], align 4
|
|
; LE-NEXT: [[PSTOR3:%.*]] = getelementptr i8, ptr [[PCMP]], i64 12
|
|
; LE-NEXT: store i32 1, ptr [[PSTOR3]], align 4
|
|
; LE-NEXT: [[PSTOR4:%.*]] = getelementptr i8, ptr [[PCMP]], i64 16
|
|
; LE-NEXT: store i32 1, ptr [[PSTOR4]], align 4
|
|
; LE-NEXT: [[PSTOR5:%.*]] = getelementptr i8, ptr [[PCMP]], i64 20
|
|
; LE-NEXT: store i32 1, ptr [[PSTOR5]], align 4
|
|
; LE-NEXT: [[PSTOR6:%.*]] = getelementptr i8, ptr [[PCMP]], i64 24
|
|
; LE-NEXT: store i32 1, ptr [[PSTOR6]], align 4
|
|
; LE-NEXT: ret void
|
|
;
|
|
%p0 = getelementptr [4 x i16], ptr @ia16a, i64 0, i64 1
|
|
%q = getelementptr [8 x i8], ptr @i8a, i64 0, i64 1
|
|
|
|
%cmp0 = call i32 @memcmp(ptr %p0, ptr %q, i64 0)
|
|
store i32 %cmp0, ptr %pcmp
|
|
|
|
%cmp1 = call i32 @memcmp(ptr %p0, ptr %q, i64 1)
|
|
%pstor1 = getelementptr i32, ptr %pcmp, i64 1
|
|
store i32 %cmp1, ptr %pstor1
|
|
|
|
%cmp2 = call i32 @memcmp(ptr %p0, ptr %q, i64 2)
|
|
%pstor2 = getelementptr i32, ptr %pcmp, i64 2
|
|
store i32 %cmp2, ptr %pstor2
|
|
|
|
%cmp3 = call i32 @memcmp(ptr %p0, ptr %q, i64 3)
|
|
%pstor3 = getelementptr i32, ptr %pcmp, i64 3
|
|
store i32 %cmp3, ptr %pstor3
|
|
|
|
%cmp4 = call i32 @memcmp(ptr %p0, ptr %q, i64 4)
|
|
%pstor4 = getelementptr i32, ptr %pcmp, i64 4
|
|
store i32 %cmp4, ptr %pstor4
|
|
|
|
%cmp5 = call i32 @memcmp(ptr %p0, ptr %q, i64 5)
|
|
%pstor5 = getelementptr i32, ptr %pcmp, i64 5
|
|
store i32 %cmp5, ptr %pstor5
|
|
|
|
%cmp6 = call i32 @memcmp(ptr %p0, ptr %q, i64 6)
|
|
%pstor6 = getelementptr i32, ptr %pcmp, i64 6
|
|
store i32 %cmp6, ptr %pstor6
|
|
|
|
ret void
|
|
}
|