[bolt][aarch64] simplify rodata/literal load for X86 & AArch64 #165723
+201
−13
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
yavtuk
requested review from
aaupov,
ayermolo,
maksfb,
paschalis-mpeis,
rafaelauler,
yota9 and
yozhu
as code owners
Member
|
@llvm/pr-subscribers-bolt Author: Alexey Moksyakov (yavtuk) Changesldr reg, literal instruction is limited +/- 1MB range, emitCI put the constants by the end of function and the one is out of available range. Full diff: https://github.com/llvm/llvm-project/pull/165723.diff 1 Files Affected:
diff --git a/bolt/test/AArch64/materialize-constant.s b/bolt/test/AArch64/materialize-constant.s
new file mode 100644
index 0000000000000..1c15626b09594
--- /dev/null
+++ b/bolt/test/AArch64/materialize-constant.s
@@ -0,0 +1,74 @@
+// this test checks a load literal instructions changed to movk
+
+// REQUIRES: system-linux
+
+# RUN: llvm-mc -filetype=obj -triple aarch64-unknown-unknown %s -o %t.o
+
+# RUN: link_fdata %s %t.o %t.fdata
+# RUN: %clang %cflags -pie %t.o -o %t.exe -Wl,-q -Wl,-z,relro -Wl,-z,now
+# RUN: llvm-bolt %t.exe -o %t.bolt -data %t.fdata \
+# RUN: --keep-nops --eliminate-unreachable=false
+# RUN: llvm-objdump --disassemble-symbols=foo %t.bolt | FileCheck %s
+
+# CHECK: mov{{.*}} w19, #0
+# CHECK-NEXT: mov{{.*}} w22, #0
+# CHECK-NEXT: movk{{.*}} w23, #0, lsl #16
+# CHECK-NEXT: movk{{.*}} w23, #100
+# CHECK-NEXT: movk{{.*}} w24, #0, lsl #16
+# CHECK-NEXT: movk{{.*}} w24, #3
+
+ .text
+ .align 4
+ .local foo
+ .type foo, %function
+foo:
+# FDATA: 1 main 0 1 foo 0 0 10
+ stp x29, x30, [sp, #-32]!
+ stp x19, x20, [sp, #16]
+ mov x29, sp
+
+ mov w19, #0 // counter = 0
+ mov w22, #0 // result = 0
+
+ ldr w23, .Llimit
+ ldr w24, .LStep
+ b .LStub
+
+.LConstants:
+ .Llimit: .word 100
+ .LStep: .word 3
+
+.LStub:
+.rep 0x100000
+ nop
+.endr
+ b .Lmain_loop
+
+.Lmain_loop:
+ madd w22, w19, w24, w22 // result += counter * increment
+
+ add w19, w19, #1
+ cmp w19, w23
+ b.lt .Lmain_loop
+
+ mov w0, w22
+
+ b .Lreturn_point
+
+.Lreturn_point:
+ ldp x19, x20, [sp, #16]
+ ldp x29, x30, [sp], #32
+ ret
+.size foo, .-foo
+
+
+ .global main
+ .type main, %function
+main:
+ mov x0, #0
+ bl foo
+ mov x0, 0
+ mov w8, #93
+ svc #0
+
+ .size main, .-main
|
Contributor
Author
|
hi folks, I've faced with the issue related to ldr reg, literal instruction inside function with constant island. after emitCI the constants are placed by the end of the function and the distance to target for ldr is out of range. I think such cases are possibly for ADR instruction either and also after long jump pass. |
Contributor
|
Thanks for the patch! Wondering if it would be useful to combine this "ldr->movk" transformation with some sort of ldr relaxation (as in #165787), particularly for cases where the constant island content has (dynamic) relocation and need to be fixed up? |
Contributor
|
@yavtuk, great idea! If this is integrated into Extra step on top will be elimination of the constant island that is no longer referenced. Running |
Member
|
It would be good to handle this in the suggested pass, also reporting how many times it was applied. From discussion, this doesn't appear to be a common case. |
ldr reg, literal instruction is limited +/- 1MB range, emitCI put the constants by the end of function and the one is out of available range.
yavtuk
force-pushed
the
check-load-literal
branch
from
82564f4 to
95e84b2
Compare
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
🐧 Linux x64 Test Results
Failed Tests(click on a test name to see its output) BOLTBOLT.AArch64/materialize-constant.sBOLT.X86/rodata-simpl-loads.testBOLT.X86/srol-bug.testIf these failures are unrelated to your changes (for example tests are broken or flaky at HEAD), please open an issue at https://github.com/llvm/llvm-project/issues and add the |
yavtuk
force-pushed
the
check-load-literal
branch
2 times, most recently
from
ec9d1a5 to
54f11ca
Compare
yavtuk
changed the title
yavtuk
changed the title
yavtuk
force-pushed
the
check-load-literal
branch
from
54f11ca to
4ef587a
Compare
Contributor
Author
|
@maksfb @yozhu @paschalis-mpeis please take a look when you have a time |
yozhu
reviewed
Nov 26, 2025
yozhu
reviewed
Nov 26, 2025
yozhu
reviewed
Nov 26, 2025
| uint32_t Offset = TargetAddress - DataSection->getAddress(); | ||
| StringRef ConstantData = DataSection->getContents(); | ||
| const InstructionListType Instrs = | ||
| MIB->materializeConstant(Inst, ConstantData, Offset); |
Contributor
There was a problem hiding this comment.
For x86, could we keep the original code which directly update the Instr in place? The new code calling materializeConstant() will call replaceMemOperandWithImm(), create a new Instr and copy over the new instruction operand by operand, find the original instruction in basic block, and then replace it with the new instruction.
Contributor
Author
There was a problem hiding this comment.
I prefer to use extra wrapper to unify for all platform, replaceMemOperandWithImm is used in another optimisations passes related to X86
yozhu
reviewed
Nov 26, 2025
yozhu
reviewed
Nov 26, 2025
| I = {8, 4, AArch64::MOVKXi}; | ||
| break; | ||
| default: | ||
| llvm_unreachable("unexpected ldr instruction"); |
Contributor
There was a problem hiding this comment.
Could there be any other PC relative LDR instructions hit the default case, like byte or half-word load? Just in case, maybe replace llvm_unreachable() with a comment saying unsupported and then return Insts.
Contributor
Author
There was a problem hiding this comment.
I don't find instructions related to byte or half wold, now only 2 instructions are handled.
we can analyse the apps to figure out which and how much are used, I faced only 4 and 8 bytes
Contributor
Author
There was a problem hiding this comment.
we will be able to add extra logic as necessary
yozhu
reviewed
Nov 26, 2025
yozhu
reviewed
Nov 27, 2025
| b .LStub | ||
|
|
||
| // CI moved by emitCI function to the end of the function | ||
| // without materialization CI is outside available range (+/-1MB) |
yozhu
reviewed
Nov 27, 2025
yozhu
reviewed
Nov 27, 2025
|
|
||
| .size _start, .-_start | ||
|
|
||
| #--- materialize-ci-outside-func.s |
Contributor
There was a problem hiding this comment.
nit: this test case is almost the same as the previous one, except for the many nop's under .LStub. Could we keep the first copy and wrap the .LStub block with #if defined(...) and #endif?
yozhu
reviewed
Nov 27, 2025
Contributor
Thanks for the patch, and sorry for the delay on review. I left some comments in the code; and overall LGTM. |
yavtuk
force-pushed
the
check-load-literal
branch
from
4ef587a to
d80a5e6
Compare
Contributor
Author
|
@yozhu do we need one more verification pass to check that the distance is valid for target? |
yavtuk
force-pushed
the
check-load-literal
branch
2 times, most recently
from
b2d22b9 to
e96818d
Compare
This patch fixed the issue related to load literal for AArch64 (bolt/test/AArch64/materialize-constant.s), address range for literal is limited +/- 1MB, emitCI puts the constants by the end of function and the one is out of available range. SimplifyRODataLoads is enabled by default for X86 & AArch64 Signed-off-by: Moksyakov Alexey <[email protected]>
yavtuk
force-pushed
the
check-load-literal
branch
from
e96818d to
de57782
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
ldr reg, literal instruction is limited +/- 1MB range, emitCI put the constants by the end of function and the one is out of available range.