Files
clang-p2996/llvm/test/Transforms/InstCombine/gep-addrspace.ll
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

78 lines
3.3 KiB
LLVM

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -passes=instcombine -S | FileCheck %s
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
target triple = "x86_64-pc-win32"
%myStruct = type { float, [3 x float], [4 x float], i32 }
; make sure that we are not crashing when creating an illegal type
define void @func(ptr addrspace(1) nocapture %p) nounwind {
; CHECK-LABEL: @func(
; CHECK-NEXT: ret void
;
%B = addrspacecast ptr addrspace(1) %p to ptr
%C = getelementptr inbounds %myStruct, ptr %B, i32 0, i32 1
%D = getelementptr inbounds [3 x float], ptr %C, i32 0, i32 2
%E = load float, ptr %D, align 4
%F = fsub float %E, undef
ret void
}
@array = internal addrspace(3) global [256 x float] zeroinitializer, align 4
@scalar = internal addrspace(3) global float 0.000000e+00, align 4
define void @keep_necessary_addrspacecast(i64 %i, ptr %out0, ptr %out1) {
; CHECK-LABEL: @keep_necessary_addrspacecast(
; CHECK-NEXT: [[T0:%.*]] = getelementptr [256 x float], ptr addrspacecast (ptr addrspace(3) @array to ptr), i64 0, i64 [[I:%.*]]
; CHECK-NEXT: [[T1:%.*]] = getelementptr [0 x float], ptr addrspacecast (ptr addrspace(3) @scalar to ptr), i64 0, i64 [[I]]
; CHECK-NEXT: store ptr [[T0]], ptr [[OUT0:%.*]], align 4
; CHECK-NEXT: store ptr [[T1]], ptr [[OUT1:%.*]], align 4
; CHECK-NEXT: ret void
;
%t0 = getelementptr [256 x float], ptr addrspacecast (ptr addrspace(3) @array to ptr), i64 0, i64 %i
%t1 = getelementptr [0 x float], ptr addrspacecast (ptr addrspace(3) @scalar to ptr), i64 0, i64 %i
store ptr %t0, ptr %out0, align 4
store ptr %t1, ptr %out1, align 4
ret void
}
declare void @escape_alloca(ptr)
; check that addrspacecast is stripped when trying to mark a GEP as inbounds
define { i8, i8 } @inbounds_after_addrspacecast() {
; CHECK-LABEL: @inbounds_after_addrspacecast(
; CHECK-NEXT: [[T0:%.*]] = alloca i16, align 2
; CHECK-NEXT: call void @escape_alloca(ptr nonnull [[T0]])
; CHECK-NEXT: [[T1:%.*]] = addrspacecast ptr [[T0]] to ptr addrspace(11)
; CHECK-NEXT: [[T2:%.*]] = getelementptr inbounds i8, ptr addrspace(11) [[T1]], i64 1
; CHECK-NEXT: [[T3:%.*]] = load i8, ptr addrspace(11) [[T2]], align 1
; CHECK-NEXT: [[INSERT:%.*]] = insertvalue { i8, i8 } zeroinitializer, i8 [[T3]], 1
; CHECK-NEXT: ret { i8, i8 } [[INSERT]]
;
%t0 = alloca i16, align 2
call void @escape_alloca(ptr %t0)
%t1 = addrspacecast ptr %t0 to ptr addrspace(11)
%t2 = getelementptr [2 x i8], ptr addrspace(11) %t1, i64 0, i64 1
%t3 = load i8, ptr addrspace(11) %t2, align 1
%insert = insertvalue { i8, i8 } zeroinitializer, i8 %t3, 1
ret { i8, i8 } %insert
}
declare spir_func <16 x i32> @my_extern_func()
; check that a bitcast is not generated when we need an addrspace cast
define void @bitcast_after_gep(ptr %t0) {
; CHECK-LABEL: @bitcast_after_gep(
; CHECK-NEXT: [[T2:%.*]] = addrspacecast ptr [[T0:%.*]] to ptr addrspace(3)
; CHECK-NEXT: [[CALL:%.*]] = call spir_func <16 x i32> @my_extern_func()
; CHECK-NEXT: store <16 x i32> [[CALL]], ptr addrspace(3) [[T2]], align 64
; CHECK-NEXT: ret void
;
%t2 = addrspacecast ptr %t0 to ptr addrspace(3)
%call = call spir_func <16 x i32> @my_extern_func()
store <16 x i32> %call, ptr addrspace(3) %t2
ret void
}