Nikita Popov 90ba33099c
[InstCombine] Canonicalize constant GEPs to i8 source element type (#68882)
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.
2024-01-24 15:25:29 +01:00

133 lines
4.8 KiB
LLVM

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -passes=instcombine -S | FileCheck %s
;
; Exercise folding of memcmp calls with addition expressions involving
; pointers into constant arrays of types larger than char and fractional
; offsets.
declare i32 @memcmp(ptr, ptr, i64)
@i32a = constant [2 x i16] [i16 4386, i16 13124]
@i32b = constant [2 x i16] [i16 4386, i16 13124]
define void @fold_memcmp_i32a_i32b_pIb(i32 %I, ptr %pcmp)
; CHECK-LABEL: @fold_memcmp_i32a_i32b_pIb(
; CHECK-NEXT: store i32 0, ptr [[PCMP:%.*]], align 4
; CHECK-NEXT: [[PST_1_1_2:%.*]] = getelementptr i8, ptr [[PCMP]], i64 4
; CHECK-NEXT: store i32 0, ptr [[PST_1_1_2]], align 4
; CHECK-NEXT: [[PST_1_1_3:%.*]] = getelementptr i8, ptr [[PCMP]], i64 8
; CHECK-NEXT: store i32 0, ptr [[PST_1_1_3]], align 4
; CHECK-NEXT: ret void
;
{
%pi8ap1 = getelementptr i8, ptr @i32a, i32 1
%pi8bp1 = getelementptr i8, ptr @i32b, i32 1
%cmp_1_1_1 = call i32 @memcmp(ptr %pi8ap1, ptr %pi8ap1, i64 1)
store i32 %cmp_1_1_1, ptr %pcmp
%pst_1_1_2 = getelementptr i32, ptr %pcmp, i32 1
%cmp_1_1_2 = call i32 @memcmp(ptr %pi8ap1, ptr %pi8ap1, i64 2)
store i32 %cmp_1_1_2, ptr %pst_1_1_2
%pst_1_1_3 = getelementptr i32, ptr %pcmp, i32 2
%cmp_1_1_3 = call i32 @memcmp(ptr %pi8ap1, ptr %pi8ap1, i64 3)
store i32 %cmp_1_1_3, ptr %pst_1_1_3
ret void
}
%struct.A = type { [4 x i8] }
%struct.B = type { [2 x i8], [2 x i8] }
@a = constant [1 x %struct.A] [%struct.A { [4 x i8] [i8 1, i8 2, i8 3, i8 4] }]
@b = constant [1 x %struct.B] [%struct.B { [2 x i8] [i8 1, i8 2], [2 x i8] [i8 3, i8 4]}]
define void @fold_memcmp_A_B_pIb(i32 %I, ptr %pcmp) {
; CHECK-LABEL: @fold_memcmp_A_B_pIb(
; CHECK-NEXT: store i32 0, ptr [[PCMP:%.*]], align 4
; CHECK-NEXT: [[PST_0_0_2:%.*]] = getelementptr i8, ptr [[PCMP]], i64 4
; CHECK-NEXT: store i32 0, ptr [[PST_0_0_2]], align 4
; CHECK-NEXT: [[PST_0_0_3:%.*]] = getelementptr i8, ptr [[PCMP]], i64 8
; CHECK-NEXT: store i32 0, ptr [[PST_0_0_3]], align 4
; CHECK-NEXT: [[PST_0_0_4:%.*]] = getelementptr i8, ptr [[PCMP]], i64 12
; CHECK-NEXT: store i32 0, ptr [[PST_0_0_4]], align 4
; CHECK-NEXT: [[PST_0_1_1:%.*]] = getelementptr i8, ptr [[PCMP]], i64 16
; CHECK-NEXT: store i32 -1, ptr [[PST_0_1_1]], align 4
; CHECK-NEXT: [[PST_0_1_2:%.*]] = getelementptr i8, ptr [[PCMP]], i64 20
; CHECK-NEXT: store i32 -1, ptr [[PST_0_1_2]], align 4
; CHECK-NEXT: [[PST_0_1_3:%.*]] = getelementptr i8, ptr [[PCMP]], i64 24
; CHECK-NEXT: store i32 -1, ptr [[PST_0_1_3]], align 4
; CHECK-NEXT: [[PST_1_0_1:%.*]] = getelementptr i8, ptr [[PCMP]], i64 16
; CHECK-NEXT: store i32 1, ptr [[PST_1_0_1]], align 4
; CHECK-NEXT: [[PST_1_0_2:%.*]] = getelementptr i8, ptr [[PCMP]], i64 20
; CHECK-NEXT: store i32 1, ptr [[PST_1_0_2]], align 4
; CHECK-NEXT: [[PST_1_0_3:%.*]] = getelementptr i8, ptr [[PCMP]], i64 24
; CHECK-NEXT: store i32 1, ptr [[PST_1_0_3]], align 4
; CHECK-NEXT: ret void
;
; Fold memcmp(&a, &b, 1) to 0;
%cmp_0_0_1 = call i32 @memcmp(ptr @a, ptr @b, i64 1)
store i32 %cmp_0_0_1, ptr %pcmp
; Fold memcmp(&a, &b, 2) to 0;
%pst_0_0_2 = getelementptr i32, ptr %pcmp, i32 1
%cmp_0_0_2 = call i32 @memcmp(ptr @a, ptr @b, i64 2)
store i32 %cmp_0_0_2, ptr %pst_0_0_2
; Fold memcmp(&a, &b, 3) to 0;
%pst_0_0_3 = getelementptr i32, ptr %pcmp, i32 2
%cmp_0_0_3 = call i32 @memcmp(ptr @a, ptr @b, i64 3)
store i32 %cmp_0_0_3, ptr %pst_0_0_3
; Fold memcmp(&a, &b, 4) to 0;
%pst_0_0_4 = getelementptr i32, ptr %pcmp, i32 3
%cmp_0_0_4 = call i32 @memcmp(ptr @a, ptr @b, i64 4)
store i32 %cmp_0_0_4, ptr %pst_0_0_4
%pi8bp1 = getelementptr i8, ptr @b, i32 1
; Fold memcmp(&a, (char*)&b + 1, 1) to -1;
%pst_0_1_1 = getelementptr i32, ptr %pcmp, i32 4
%cmp_0_1_1 = call i32 @memcmp(ptr @a, ptr %pi8bp1, i64 1)
store i32 %cmp_0_1_1, ptr %pst_0_1_1
; Fold memcmp(&a, (char*)&b + 1, 2) to -1;
%pst_0_1_2 = getelementptr i32, ptr %pcmp, i32 5
%cmp_0_1_2 = call i32 @memcmp(ptr @a, ptr %pi8bp1, i64 2)
store i32 %cmp_0_1_2, ptr %pst_0_1_2
; Fold memcmp(&a, (char*)&b + 1, 3) to -1;
%pst_0_1_3 = getelementptr i32, ptr %pcmp, i32 6
%cmp_0_1_3 = call i32 @memcmp(ptr @a, ptr %pi8bp1, i64 3)
store i32 %cmp_0_1_3, ptr %pst_0_1_3
%pi8ap1 = getelementptr i8, ptr @a, i32 1
; Fold memcmp((char*)&a + 1, &b, 1) to +1;
%pst_1_0_1 = getelementptr i32, ptr %pcmp, i32 4
%cmp_1_0_1 = call i32 @memcmp(ptr %pi8ap1, ptr @b, i64 1)
store i32 %cmp_1_0_1, ptr %pst_1_0_1
; Fold memcmp((char*)&a + 1, &b, 2) to +1;
%pst_1_0_2 = getelementptr i32, ptr %pcmp, i32 5
%cmp_1_0_2 = call i32 @memcmp(ptr %pi8ap1, ptr @b, i64 2)
store i32 %cmp_1_0_2, ptr %pst_1_0_2
; Fold memcmp((char*)&a + 1, &b, 3) to +1;
%pst_1_0_3 = getelementptr i32, ptr %pcmp, i32 6
%cmp_1_0_3 = call i32 @memcmp(ptr %pi8ap1, ptr @b, i64 3)
store i32 %cmp_1_0_3, ptr %pst_1_0_3
ret void
}