f573d489b6
The concept of G_GLOBAL_VALUE is nice and simple, but always using it as the representation for global var addressing until selection time creates some problems in optimizing accesses in certain code/relocation models. The problem comes from trying to optimize adrp -> add -> load/store sequences in the most common "small" code model. These accesses can be optimized into an adrp -> load with the add offset being folded into the load's immediate field. If we try to keep all global var references as a single generic instruction then by the time we get to the complex operand trying to match these, we end up generating an adrp at the point of use. The real issue here is that we don't have any form of CSE during selection, so the code size will bloat from many redundant adrp's. This patch custom legalizes small code mode non-GOT G_GLOBALs into target ADRP and a new "target specific generic opcode" G_ADD_LOW. We also teach the localizer to localize these instructions via the custom hook that was added recently. Finally, the complex pattern for indexed loads/stores is extended to try to fold these G_ADD_LOW instructions into the load immediate. On -O0 CTMark, we see a 0.8% geomean code size improvement. We should also see some minor performance improvements too. Differential Revision: https://reviews.llvm.org/D78465
81 lines
3.5 KiB
YAML
81 lines
3.5 KiB
YAML
# RUN: llc -O0 -verify-machineinstrs -mtriple aarch64-- -run-pass=legalizer %s -o - | FileCheck %s
|
|
|
|
# Check that when we combine ZEXT/ANYEXT we assign the correct location.
|
|
# CHECK: !8 = !DILocation(line: 23, column: 5, scope: !4)
|
|
# CHECK: G_AND %16, %15, debug-location !8
|
|
|
|
--- |
|
|
target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128"
|
|
target triple = "arm64-apple-ios13.0.0"
|
|
|
|
@.str = external dso_local unnamed_addr constant [4 x i8], align 1
|
|
|
|
define void @main() personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) !dbg !4 {
|
|
entry:
|
|
%tobool = trunc i8 undef to i1
|
|
%conv = zext i1 %tobool to i32
|
|
call void (i8*, ...) @printf(i8* getelementptr inbounds ([4 x i8], [4 x i8]* @.str, i64 0, i64 0), i32 %conv), !dbg !8
|
|
%0 = load i32, i32* undef, align 4, !dbg !9
|
|
call void (i8*, ...) @printf(i8* getelementptr inbounds ([4 x i8], [4 x i8]* @.str, i64 0, i64 0), i32 %0)
|
|
ret void
|
|
}
|
|
|
|
declare void @printf(i8*, ...)
|
|
declare i32 @__gxx_personality_v0(...)
|
|
|
|
!llvm.dbg.cu = !{!0}
|
|
!llvm.module.flags = !{!2, !3}
|
|
|
|
!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !1, producer: "clang version 10.0.0", isOptimized: false, runtimeVersion: 0, emissionKind: LineTablesOnly, nameTableKind: None, sysroot: "/pata/tino", sdk: "iPhoneOS13.0.sdk")
|
|
!1 = !DIFile(filename: "/pata/tino/main.cpp", directory: "/pata/tino")
|
|
!2 = !{i32 2, !"Debug Info Version", i32 3}
|
|
!3 = !{i32 7, !"PIC Level", i32 2}
|
|
!4 = distinct !DISubprogram(name: "main", scope: !5, file: !5, line: 19, type: !6, scopeLine: 20, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0)
|
|
!5 = !DIFile(filename: "main.cpp", directory: "/pata/tino")
|
|
!6 = !DISubroutineType(types: !7)
|
|
!7 = !{}
|
|
!8 = !DILocation(line: 23, column: 5, scope: !4)
|
|
!9 = !DILocation(line: 36, column: 21, scope: !4)
|
|
|
|
...
|
|
---
|
|
name: main
|
|
registers:
|
|
- { id: 0, class: _, preferred-register: '' }
|
|
- { id: 1, class: _, preferred-register: '' }
|
|
- { id: 2, class: _, preferred-register: '' }
|
|
- { id: 3, class: _, preferred-register: '' }
|
|
- { id: 4, class: _, preferred-register: '' }
|
|
- { id: 5, class: _, preferred-register: '' }
|
|
- { id: 6, class: _, preferred-register: '' }
|
|
- { id: 7, class: _, preferred-register: '' }
|
|
- { id: 8, class: _, preferred-register: '' }
|
|
- { id: 9, class: _, preferred-register: '' }
|
|
body: |
|
|
bb.1.entry:
|
|
%0:_(s8) = G_IMPLICIT_DEF
|
|
%4:_(p0) = COPY $x0
|
|
%10:_(p0) = G_IMPLICIT_DEF debug-location !DILocation(line: 0, scope: !4)
|
|
%1:_(s1) = G_TRUNC %0(s8)
|
|
%2:_(s32) = G_ZEXT %1(s1)
|
|
ADJCALLSTACKDOWN 8, 0, implicit-def $sp, implicit $sp, debug-location !8
|
|
$x0 = COPY %4(p0), debug-location !8
|
|
%5:_(p0) = COPY $sp, debug-location !8
|
|
%6:_(s64) = G_CONSTANT i64 0
|
|
%7:_(p0) = G_PTR_ADD %5, %6(s64), debug-location !8
|
|
%8:_(s64) = G_ANYEXT %2(s32), debug-location !8
|
|
G_STORE %8(s64), %7(p0), debug-location !8 :: (store 8 into stack, align 1)
|
|
BL @printf, csr_aarch64_aapcs, implicit-def $lr, implicit $sp, implicit $x0, debug-location !8
|
|
ADJCALLSTACKUP 8, 0, implicit-def $sp, implicit $sp, debug-location !8
|
|
%13:_(s64) = G_LOAD %10(p0), debug-location !9 :: (load 4 from `i32* undef`)
|
|
ADJCALLSTACKDOWN 8, 0, implicit-def $sp, implicit $sp
|
|
$x0 = COPY %4(p0)
|
|
%11:_(p0) = COPY $sp
|
|
%12:_(p0) = G_PTR_ADD %11, %6(s64)
|
|
G_STORE %13(s64), %12(p0) :: (store 8 into stack, align 1)
|
|
BL @printf, csr_aarch64_aapcs, implicit-def $lr, implicit $sp, implicit $x0
|
|
ADJCALLSTACKUP 8, 0, implicit-def $sp, implicit $sp
|
|
RET_ReallyLR
|
|
|
|
...
|