<div dir="ltr"><div style>Umesh,</div><div style><br></div>Makes some sort of sense to me, OTOH:<div>If instead of choosing r11 as a "dummy" to align the stack we had chosen some other register in the range r0-r7 then we could have emitted the PUSH encoding T1 (2 bytes opcode) as opposed to the encoding T2 (which is a 4 bytes opcode).<div>
<br></div><div style>A</div></div></div><div class="gmail_extra"><br><br><div class="gmail_quote">On Tue, Oct 15, 2013 at 2:59 AM, Umesh Kalappa <span dir="ltr"><<a href="mailto:umesh.kalappa0@gmail.com" target="_blank">umesh.kalappa0@gmail.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Hi Andrea,<br>
<br>
That is because the LR is the fixed register as per the<br>
<a href="http://infocenter.arm.com/help/topic/com.arm.doc.ihi0042e/IHI0042E_aapcs.pdf" target="_blank">http://infocenter.arm.com/help/topic/com.arm.doc.ihi0042e/IHI0042E_aapcs.pdf</a><br>
and out_char() function is not the leaf function ,Hence compiler<br>
tends to save and restore the LR and the save and restore of<br>
register r11 is to align stack for 8 bytes as per ARM EABI.<br>
<br>
<br>
<br>
Thanks<br>
<br>
~Umesh<br>
<br>
<br>
On Tuesday, October 15, 2013, Umesh Kalappa <<a href="mailto:umesh.kalappa0@gmail.com">umesh.kalappa0@gmail.com</a>> wrote:<br>
<div><div class="h5">> ---------- Forwarded message ----------<br>
> From: Andrea Mucignat <<a href="mailto:andrea@nestlabs.com">andrea@nestlabs.com</a>><br>
> Date: Tue, Oct 15, 2013 at 7:05 AM<br>
> Subject: [LLVMdev] Unwanted push/pop on Cortex-M.<br>
> To: <a href="mailto:llvmdev@cs.uiuc.edu">llvmdev@cs.uiuc.edu</a><br>
><br>
><br>
> Hi,<br>
><br>
> I have this code:<br>
><br>
> void platform_putchar(int, char);<br>
> void out_char( char ch );<br>
><br>
> void out_char( char ch )<br>
> {<br>
> platform_putchar (0, ch);<br>
> }<br>
><br>
> I'm compiling with the following clang invocation:<br>
><br>
> $ /usr/local/vendor/toolchains/llvm/3.3/armv7m/bin/armv7m-none-eabi-clang<br>
> -mcpu=cortex-m4 -mfloat-abi=soft -mthumb -nostdinc -ffreestanding<br>
> -ffunction-sections -fdata-sections -fno-exceptions -fno-rtti<br>
> -fomit-frame-pointer -momit-leaf-frame-pointer -nostdinc -v -Os -S -o<br>
> a.s a.c<br>
> vendor-clang version 3.3 (tags/RELEASE_33/final) (based on LLVM 3.3)<br>
> Target: armv7m-none--eabi<br>
> Thread model: posix<br>
> "/usr/local/vendor-20130805-b8d59d2/toolchains/llvm/3.3/armv7m/bin/armv7m-none-eabi-clang"<br>
> -cc1 -triple thumbv7em-none--eabi -S -disable-free -main-file-name a.c<br>
> -mrelocation-model static -fmath-errno -mconstructor-aliases<br>
> -target-abi aapcs -target-cpu cortex-m4 -msoft-float -mfloat-abi soft<br>
> -target-feature +soft-float -target-feature +soft-float-abi<br>
> -target-feature -neon -target-linker-version 2.22<br>
> -momit-leaf-frame-pointer -v -ffunction-sections -fdata-sections<br>
> -coverage-file /tmp/a.s -nostdsysteminc -nobuiltininc -resource-dir<br>
> /usr/local/vendor-20130805-b8d59d2/toolchains/llvm/3.3/armv7m/bin/../lib/clang/3.3<br>
> -Os -fno-dwarf-directory-asm -fdebug-compilation-dir /tmp<br>
> -ferror-limit 19 -fmessage-length 207 -ffreestanding -mstackrealign<br>
> -fno-rtti -fno-signed-char -fobjc-runtime=gcc<br>
> -fobjc-default-synthesize-properties -fdiagnostics-show-option<br>
> -fcolor-diagnostics -backend-option -vectorize-loops -o a.s -x c a.c<br>
> clang -cc1 version 3.3 based upon LLVM 3.3 default target armv7m-none-eabi<br>
> #include "..." search starts here:<br>
> End of search list.<br>
><br>
> $ cat a.s<br>
> .syntax unified<br>
> .eabi_attribute 6, 10<br>
> .eabi_attribute 9, 2<br>
> .eabi_attribute 10, 5<br>
> .fpu vfpv4<br>
> .eabi_attribute 20, 1<br>
> .eabi_attribute 21, 1<br>
> .eabi_attribute 23, 3<br>
> .eabi_attribute 24, 1<br>
> .eabi_attribute 25, 1<br>
> .eabi_attribute 44, 1<br>
> .file "a.c"<br>
> .section .text.out_char,"ax",%progbits<br>
> .globl out_char<br>
> .align 2<br>
> .type out_char,%function<br>
> .code 16<br>
> .thumb_func<br>
> out_char:<br>
> push.w {r11, lr}<br>
> mov r1, r0<br>
> movs r0, #0<br>
> bl platform_putchar<br>
> pop.w {r11, pc}<br>
> .Ltmp0:<br>
> .size out_char, .Ltmp0-out_char<br>
><br>
> The one question I have is:<br>
> why can't this out_char function be compiled into:<br>
><br>
> out_char:<br>
> mov r1, r0<br>
> mov r0, #0<br>
> bl platform_putchar<br>
> bx lr<br>
><br>
> What's the clang/llvm module responsible for generating the prologue<br>
> and epilogue for this function?<br>
> I looked into CodeGen/PrologEpilogInserter, ARMFrameLowering.cpp<br>
> Thumb1FrameLowering.cpp but it's not 100% clear how the code gen<br>
> decides what the stack frame look like.<br>
><br>
> Here's the -emit-llvm output<br>
><br>
> target datalayout =<br>
> "e-p:32:32:32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:64:128-a0:0:32-n32-S64"<br>
> target triple = "thumbv7em-none--eabi"<br>
><br>
> ; Function Attrs: nounwind optsize<br>
> define void @out_char(char)(i8 zeroext %ch) #0 {<br>
> entry:<br>
> tail call void @llvm.dbg.value(metadata !{i8 %ch}, i64 0, metadata<br>
> !10), !dbg !11<br>
> tail call void @platform_putchar(int, char)(i32 0, i8 zeroext %ch)<br>
> #3, !dbg !12<br>
> ret void, !dbg !13<br>
> }<br>
><br>
> ; Function Attrs: optsize<br>
> declare void @platform_putchar(int, char)(i32, i8 zeroext) #1<br>
><br>
> ; Function Attrs: nounwind readnone<br>
> declare void @llvm.dbg.value(metadata, i64, metadata) #2<br>
><br>
> attributes #0 = { nounwind optsize "less-precise-fpmad"="false"<br>
> "no-frame-pointer-elim"="false"<br>
</div></div>> "n_______________________________________________<br>
> LLVM Developers mailing list<br>
> <a href="mailto:LLVMdev@cs.uiuc.edu">LLVMdev@cs.uiuc.edu</a> <a href="http://llvm.cs.uiuc.edu" target="_blank">http://llvm.cs.uiuc.edu</a><br>
> <a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev" target="_blank">http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev</a><br>
><br>
</blockquote></div><br></div>