shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions 6 Packages Projects Releases Wiki Activity
llvm-project/llvm/test/Transforms/SROA/slice-width.ll
Björn Pettersson 1c305aebd6
[SROA] Use shufflevector instead of select for vector blend (#175756) 
A patch from May 2013, commit 1e211913b56f390, changed SROA into using a
select instruction to perform vector blend. Idea was that using a select
was the canonical form, and that we optimize select better than
shufflevector.

This patch is changing SROA back into using shufflevector instead of
select when doing the blend (inserting a smaller vector into a larger
vector).

Motivation:
Nowadays InstCombine is canonicalizing this kind of vector blends, using
vector select instructions, into a shufflevector instruction. So it is
assumed that shufflevector is the canonical form now. It is also assumed
that we are better at optimizing shufflevector today, compared to back
in 2013.

Commit f26710d97d9c272be8a55 includes links to a discussion from 2016
(https://discourse.llvm.org/t/ir-canonicalization-vector-select-or-shufflevector/42257/6)
about picking shufflevector as the canonical form.
2026-01-13 18:06:00 +00:00

176 lines
7.2 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals
; RUN: opt < %s -passes='sroa<preserve-cfg>' -S | FileCheck %s --check-prefixes=CHECK,CHECK-PRESERVE-CFG
; RUN: opt < %s -passes='sroa<modify-cfg>' -S | FileCheck %s --check-prefixes=CHECK,CHECK-MODIFY-CFG
target datalayout = "e-p:64:64:64-p1:16:16:16-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:64:64-f80:128-v64:64:64-v128:128:128-a0:0:64-n8:16:32:64"
declare void @llvm.memcpy.p0.p0.i32(ptr nocapture, ptr nocapture, i32, i1) nounwind
declare void @llvm.memset.p0.i32(ptr nocapture, i8, i32, i1) nounwind
declare void @llvm.memset.p0.i64(ptr nocapture, i8, i64, i1) nounwind
; This tests that allocas are not split into slices that are not byte width multiple
;.
; CHECK: @foo_copy_source = external constant %union.Foo
; CHECK: @i64_sink = global i64 0
;.
define void @no_split_on_non_byte_width(i32) {
; CHECK-LABEL: @no_split_on_non_byte_width(
; CHECK-NEXT: [[ARG_SROA_0:%.*]] = alloca i8, align 8
; CHECK-NEXT: [[ARG_SROA_0_0_EXTRACT_TRUNC:%.*]] = trunc i32 [[TMP0:%.*]] to i8
; CHECK-NEXT: store i8 [[ARG_SROA_0_0_EXTRACT_TRUNC]], ptr [[ARG_SROA_0]], align 8
; CHECK-NEXT: [[ARG_SROA_3_0_EXTRACT_SHIFT:%.*]] = lshr i32 [[TMP0]], 8
; CHECK-NEXT: [[ARG_SROA_3_0_EXTRACT_TRUNC:%.*]] = trunc i32 [[ARG_SROA_3_0_EXTRACT_SHIFT]] to i24
; CHECK-NEXT: br label [[LOAD_I32:%.*]]
; CHECK: load_i32:
; CHECK-NEXT: [[ARG_SROA_0_0_ARG_SROA_0_0_R01:%.*]] = load i8, ptr [[ARG_SROA_0]], align 8
; CHECK-NEXT: br label [[LOAD_I1:%.*]]
; CHECK: load_i1:
; CHECK-NEXT: [[ARG_SROA_0_0_ARG_SROA_0_0_T1:%.*]] = load i1, ptr [[ARG_SROA_0]], align 8
; CHECK-NEXT: ret void
;
%arg = alloca i32 , align 8
store i32 %0, ptr %arg
br label %load_i32
load_i32:
%r0 = load i32, ptr %arg
br label %load_i1
load_i1:
%t1 = load i1, ptr %arg
ret void
}
; PR18726: Check that we use memcpy and memset to fill out padding when we have
; a slice with a simple single type whose store size is smaller than the slice
; size.
%union.Foo = type { x86_fp80, i64, i64 }
@foo_copy_source = external constant %union.Foo
@i64_sink = global i64 0
define void @memcpy_fp80_padding() {
; CHECK-LABEL: @memcpy_fp80_padding(
; CHECK-NEXT: [[X_SROA_0:%.*]] = alloca x86_fp80, align 16
; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 16 [[X_SROA_0]], ptr align 16 @foo_copy_source, i32 16, i1 false)
; CHECK-NEXT: [[X_SROA_1_0_COPYLOAD:%.*]] = load i64, ptr getelementptr inbounds (i8, ptr @foo_copy_source, i64 16), align 16
; CHECK-NEXT: [[X_SROA_2_0_COPYLOAD:%.*]] = load i64, ptr getelementptr inbounds (i8, ptr @foo_copy_source, i64 24), align 8
; CHECK-NEXT: store i64 [[X_SROA_1_0_COPYLOAD]], ptr @i64_sink, align 4
; CHECK-NEXT: ret void
;
%x = alloca %union.Foo
; Copy from a global.
call void @llvm.memcpy.p0.p0.i32(ptr align 16 %x, ptr align 16 @foo_copy_source, i32 32, i1 false)
; Access a slice of the alloca to trigger SROA.
%mid_p = getelementptr %union.Foo, ptr %x, i32 0, i32 1
%elt = load i64, ptr %mid_p
store i64 %elt, ptr @i64_sink
ret void
}
define void @memset_fp80_padding() {
; CHECK-LABEL: @memset_fp80_padding(
; CHECK-NEXT: [[X_SROA_0:%.*]] = alloca x86_fp80, align 16
; CHECK-NEXT: call void @llvm.memset.p0.i32(ptr align 16 [[X_SROA_0]], i8 -1, i32 16, i1 false)
; CHECK-NEXT: store i64 -1, ptr @i64_sink, align 4
; CHECK-NEXT: ret void
;
%x = alloca %union.Foo
; Set to all ones.
call void @llvm.memset.p0.i32(ptr align 16 %x, i8 -1, i32 32, i1 false)
; Access a slice of the alloca to trigger SROA.
%mid_p = getelementptr %union.Foo, ptr %x, i32 0, i32 1
%elt = load i64, ptr %mid_p
store i64 %elt, ptr @i64_sink
ret void
}
%S.vec3float = type { float, float, float }
%U.vec3float = type { <4 x float> }
declare i32 @memcpy_vec3float_helper(ptr)
; PR18726: Check that SROA does not rewrite a 12-byte memcpy into a 16-byte
; vector store, hence accidentally putting gibberish onto the stack.
define i32 @memcpy_vec3float_widening(ptr %x) !prof !0 {
; CHECK-LABEL: @memcpy_vec3float_widening(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP1_SROA_0_0_COPYLOAD:%.*]] = load <3 x float>, ptr [[X:%.*]], align 4
; CHECK-NEXT: [[TMP1_SROA_0_0_VEC_EXPAND:%.*]] = shufflevector <3 x float> [[TMP1_SROA_0_0_COPYLOAD]], <3 x float> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
; CHECK-NEXT: [[TMP1_SROA_0_0_VECBLEND:%.*]] = shufflevector <4 x float> [[TMP1_SROA_0_0_VEC_EXPAND]], <4 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 7>
; CHECK-NEXT: [[TMP2:%.*]] = alloca [[S_VEC3FLOAT:%.*]], align 4
; CHECK-NEXT: [[TMP1_SROA_0_0_VEC_EXTRACT:%.*]] = shufflevector <4 x float> [[TMP1_SROA_0_0_VECBLEND]], <4 x float> poison, <3 x i32> <i32 0, i32 1, i32 2>
; CHECK-NEXT: store <3 x float> [[TMP1_SROA_0_0_VEC_EXTRACT]], ptr [[TMP2]], align 4
; CHECK-NEXT: [[RESULT:%.*]] = call i32 @memcpy_vec3float_helper(ptr [[TMP2]])
; CHECK-NEXT: ret i32 [[RESULT]]
;
entry:
; Create a temporary variable %tmp1 and copy %x[0] into it
%tmp1 = alloca %S.vec3float, align 4
call void @llvm.memcpy.p0.p0.i32(ptr align 4 %tmp1, ptr align 4 %x, i32 12, i1 false)
; The following block does nothing; but appears to confuse SROA
%unused3 = load <4 x float>, ptr %tmp1, align 1
; Create a second temporary and copy %tmp1 into it
%tmp2 = alloca %S.vec3float, align 4
call void @llvm.memcpy.p0.p0.i32(ptr align 4 %tmp2, ptr align 4 %tmp1, i32 12, i1 false)
%result = call i32 @memcpy_vec3float_helper(ptr %tmp2)
ret i32 %result
}
; Don't crash on length that is constant expression.
define void @PR50888() {
; CHECK-LABEL: @PR50888(
; CHECK-NEXT: [[ARRAY:%.*]] = alloca i8, align 1
; CHECK-NEXT: call void @llvm.memset.p0.i64(ptr align 1 [[ARRAY]], i8 0, i64 ptrtoint (ptr @PR50888 to i64), i1 false)
; CHECK-NEXT: ret void
;
%array = alloca i8
call void @llvm.memset.p0.i64(ptr align 16 %array, i8 0, i64 ptrtoint (ptr @PR50888 to i64), i1 false)
ret void
}
; Don't crash on out-of-bounds length.
define void @PR50910() {
; CHECK-LABEL: @PR50910(
; CHECK-NEXT: [[T1:%.*]] = alloca i8, i64 1, align 8
; CHECK-NEXT: call void @llvm.memset.p0.i64(ptr align 8 [[T1]], i8 0, i64 1, i1 false)
; CHECK-NEXT: ret void
;
%t1 = alloca i8, i64 1, align 8
call void @llvm.memset.p0.i64(ptr align 8 %t1, i8 0, i64 4294967296, i1 false)
ret void
}
define i1 @presplit_overlarge_load() {
; CHECK-LABEL: @presplit_overlarge_load(
; CHECK-NEXT: [[A_SROA_0:%.*]] = alloca i8, align 2
; CHECK-NEXT: [[A_SROA_0_0_A_SROA_0_0_L11:%.*]] = load i8, ptr [[A_SROA_0]], align 2
; CHECK-NEXT: [[A_SROA_0_0_A_SROA_0_0_L2:%.*]] = load i1, ptr [[A_SROA_0]], align 2
; CHECK-NEXT: ret i1 [[A_SROA_0_0_A_SROA_0_0_L2]]
;
%A = alloca i16
%L1 = load i32, ptr %A
%L2 = load i1, ptr %A
ret i1 %L2
}
!0 = !{!"function_entry_count", i32 10}
;.
; CHECK: attributes #[[ATTR0:[0-9]+]] = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }
; CHECK: attributes #[[ATTR1:[0-9]+]] = { nocallback nofree nounwind willreturn memory(argmem: write) }
;.
; CHECK: [[META0:![0-9]+]] = !{!"function_entry_count", i32 10}
;.
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
; CHECK-MODIFY-CFG: {{.*}}
; CHECK-PRESERVE-CFG: {{.*}}
Reference in New Issue View Git Blame Copy Permalink