http://llvm.org/bugs/show_bug.cgi?id=19483
Bug ID: 19483
Summary: Clang should include more information about the target
in the IR to help reproducing issues
Product: clang
Version: trunk
Hardware: PC
OS: All
Status: NEW
Severity: normal
Priority: P
Component: LLVM Codegen
Assignee: [email protected]
Reporter: [email protected]
CC: [email protected]
Classification: Unclassified
Created attachment 12409
--> http://llvm.org/bugs/attachment.cgi?id=12409&action=edit
An example
It would be nice if we could have the exact same codegen from ‘clang -S’ and
'clang -S -emit-llvm | llc’.
Currently, clang sets the triple information and some function attributes in
the IR file, but this is not enough. In particular, the target features as well
as the cpu are important pieces of information that are missing.
One can get this information with 'clang -v’, but this does not fit nicely in
the work flow.
I’ve attached an example that demonstrates that on MacOS X:
clang -O3 -S -o orig.s reproduce.c -arch x86_64h
clang -O3 -S -o - reproduce.c -arch x86_64h -emit-llvm | llc -o new.s
diff -U 10 orig.s new.s
--- orig.s 2014-04-18 16:53:31.000000000 -0700
+++ new.s 2014-04-18 16:52:56.000000000 -0700
@@ -6,20 +6,20 @@
.cfi_startproc
## BB#0: ## %entry
pushq %rbp
Ltmp0:
.cfi_def_cfa_offset 16
Ltmp1:
.cfi_offset %rbp, -16
movq %rsp, %rbp
Ltmp2:
.cfi_def_cfa_register %rbp
- vxorps %xmm0, %xmm0, %xmm0
- vxorps %xmm1, %xmm1, %xmm1
- vxorps %xmm2, %xmm2, %xmm2
- vxorps %xmm3, %xmm3, %xmm3
+ xorps %xmm0, %xmm0
+ xorps %xmm1, %xmm1
+ xorps %xmm2, %xmm2
+ xorps %xmm3, %xmm3
popq %rbp
jmp _bar ## TAILCALL
.cfi_endproc
clang -O3 -S -o orig.s reproduce.c -arch arm64
clang -O3 -S -o - reproduce.c -arch arm64 -emit-llvm | llc -o new.s
diff -U 10 orig.s new.s
--- orig.s 2014-04-18 17:11:18.000000000 -0700
+++ new.s 2014-04-18 17:11:53.000000000 -0700
@@ -1,14 +1,14 @@
.section __TEXT,__text,regular,pure_instructions
.ios_version_min 5, 0
.globl _foo
.align 2
_foo: ; @foo
; BB#0: ; %entry
- movi.2d v0, #0000000000000000
- movi.2d v1, #0000000000000000
- movi.2d v2, #0000000000000000
- movi.2d v3, #0000000000000000
+ fmov d0, xzr
+ orr.16b v1, v0, v0
+ orr.16b v2, v0, v0
+ orr.16b v3, v0, v0
b _bar
.subsections_via_symbols
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
LLVMbugs mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/llvmbugs
