90ba33099c
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.
66 lines
2.7 KiB
LLVM
66 lines
2.7 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
|
|
; RUN: opt -S -passes=instcombine < %s | FileCheck %s
|
|
|
|
; This test is used to verify we are not crashing at Assertion `CastInst::castIsValid(opc, C, Ty) && "Invalid constantexpr cast!".
|
|
define <vscale x 2 x ptr> @gep_index_type_is_scalable(ptr %p) {
|
|
; CHECK-LABEL: @gep_index_type_is_scalable(
|
|
; CHECK-NEXT: [[GEP:%.*]] = getelementptr i8, ptr [[P:%.*]], <vscale x 2 x i64> undef
|
|
; CHECK-NEXT: ret <vscale x 2 x ptr> [[GEP]]
|
|
;
|
|
%gep = getelementptr i8, ptr %p, <vscale x 2 x i64> undef
|
|
ret <vscale x 2 x ptr> %gep
|
|
}
|
|
|
|
; This test serves to verify code changes for "GEP.getNumIndices() == 1".
|
|
define ptr @gep_num_of_indices_1(ptr %p) {
|
|
; CHECK-LABEL: @gep_num_of_indices_1(
|
|
; CHECK-NEXT: [[GEP:%.*]] = getelementptr <vscale x 4 x i32>, ptr [[P:%.*]], i64 1
|
|
; CHECK-NEXT: ret ptr [[GEP]]
|
|
;
|
|
%gep = getelementptr <vscale x 4 x i32>, ptr %p, i64 1
|
|
ret ptr %gep
|
|
}
|
|
|
|
; This test serves to verify code changes for "GEP.getNumOperands() == 2".
|
|
define void @gep_bitcast(ptr %p) {
|
|
; CHECK-LABEL: @gep_bitcast(
|
|
; CHECK-NEXT: store <vscale x 16 x i8> zeroinitializer, ptr [[P:%.*]], align 16
|
|
; CHECK-NEXT: [[GEP2:%.*]] = getelementptr <vscale x 16 x i8>, ptr [[P]], i64 1
|
|
; CHECK-NEXT: store <vscale x 16 x i8> zeroinitializer, ptr [[GEP2]], align 16
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
store <vscale x 16 x i8> zeroinitializer, ptr %p
|
|
%gep2 = getelementptr <vscale x 16 x i8>, ptr %p, i64 1
|
|
store <vscale x 16 x i8> zeroinitializer, ptr %gep2
|
|
ret void
|
|
}
|
|
|
|
; These tests serve to verify code changes when underlying gep ptr is alloca.
|
|
; This test is to verify 'inbounds' is added when it's valid to accumulate constant offset.
|
|
define i32 @gep_alloca_inbounds_vscale_zero() {
|
|
; CHECK-LABEL: @gep_alloca_inbounds_vscale_zero(
|
|
; CHECK-NEXT: [[A:%.*]] = alloca <vscale x 4 x i32>, align 16
|
|
; CHECK-NEXT: [[TMP:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 8
|
|
; CHECK-NEXT: [[LOAD:%.*]] = load i32, ptr [[TMP]], align 4
|
|
; CHECK-NEXT: ret i32 [[LOAD]]
|
|
;
|
|
%a = alloca <vscale x 4 x i32>
|
|
%tmp = getelementptr <vscale x 4 x i32>, ptr %a, i32 0, i32 2
|
|
%load = load i32, ptr %tmp
|
|
ret i32 %load
|
|
}
|
|
|
|
; This test is to verify 'inbounds' is not added when a constant offset can not be determined at compile-time.
|
|
define i32 @gep_alloca_inbounds_vscale_nonzero() {
|
|
; CHECK-LABEL: @gep_alloca_inbounds_vscale_nonzero(
|
|
; CHECK-NEXT: [[A:%.*]] = alloca <vscale x 4 x i32>, align 16
|
|
; CHECK-NEXT: [[TMP:%.*]] = getelementptr <vscale x 4 x i32>, ptr [[A]], i64 1, i64 2
|
|
; CHECK-NEXT: [[LOAD:%.*]] = load i32, ptr [[TMP]], align 4
|
|
; CHECK-NEXT: ret i32 [[LOAD]]
|
|
;
|
|
%a = alloca <vscale x 4 x i32>
|
|
%tmp = getelementptr <vscale x 4 x i32>, ptr %a, i32 1, i32 2
|
|
%load = load i32, ptr %tmp
|
|
ret i32 %load
|
|
}
|