[llvm-commits] [llvm] r141365 - in /llvm/trunk: lib/Target/ARM/ARMISelDAGToDAG.cpp lib/Target/ARM/ARMISelLowering.cpp lib/Target/ARM/ARMInstrInfo.td lib/Target/ARM/ARMInstrThumb2.td test/CodeGen/ARM/iabs.ll test/CodeGen/Thumb/iabs.ll

[Jim Grosbach]

On Oct 7, 2011, at 5:08 PM, Ana Pazos wrote:

> Folks,
> 
> I tried to find the ARM Architecture Reference Manual for ARMv6 from ARM website. It is not available anymore according to this FAQ:
> 
> "The original ARM v6 Architecture Reference Manual for the ARM11 cores (ARM DDI 0100I) is superseded by the ARMv7-AR Architecture Reference Manual which is now the definitive document for Applications and Real-Time variants of the v6 and v7 ARM Architecture, including all ARM11-class cores."
> http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.faqs/ka6657.html
> 
> 
> But the Architecture Reference Manual for ARMv7A and ARMV7AR (ARM DDI 0406B) page A8-285 says this about RSB immediate instruction:
> 
> "The immediate value to be added to the value obtained from <Rn>. The only permitted value for encoding T1 is 0. See Modified immediate constants in Thumb instructions on page A6-17 or Modified immediate constants in ARM instructions on page A5-9 for the range of values for encoding T2 or A1".
> http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.ddi0406b/index.html


In Thumb1, there is no "rsb r0, r0, #0" instruction, only a "rsbs r0, r0, #0" which sets the flags.

> 
> 
> So it should be ok to encode rsb with immediate value 0 in Thumb1 mode... Not sure why the assembler complains:
> 
> arm-none-linux-gnueabi-gcc -march=armv6 floatsidf.s -o floatsidf.o
> floatsidf.s: Assembler messages:
> floatsidf.s:2: Error: character following name is not '#'
> floatsidf.s:3: Error: unknown pseudo-op: `.private_extern'
> floatsidf.s:7: Error: junk at end of line, first unrecognized character is`_'
> floatsidf.s:19: Error: cannot honor width suffix -- `rsb r0,r0,#0'
> floatsidf.s:100: Error: unknown pseudo-op: `.subsections_via_symbols'
> 
> Anyways, attached is the patch to disable the optimization in Thumb1 mode.
> 
> Thanks,
> Ana.
> 
> -----Original Message-----
> From: Anton Korobeynikov [mailto:anton at korobeynikov.info] 
> Sent: Friday, October 07, 2011 2:58 PM
> To: Ana Pazos
> Cc: Eric Christopher; llvm-commits at cs.uiuc.edu; rajav at codeaurora.org
> Subject: Re: [llvm-commits] [llvm] r141365 - in /llvm/trunk: lib/Target/ARM/ARMISelDAGToDAG.cpp lib/Target/ARM/ARMISelLowering.cpp lib/Target/ARM/ARMInstrInfo.td lib/Target/ARM/ARMInstrThumb2.td test/CodeGen/ARM/iabs.ll test/CodeGen/Thumb/iabs.ll
> 
> Hi Ana,
> 
>> What is your command line? Did you set any other flags?
> Attached is .ll and .s.
> For me it seems we should just disable this transform in Thumb1 mode.
> What do you think?
> 
> -- 
> With best regards, Anton Korobeynikov
> Faculty of Mathematics and Mechanics, Saint Petersburg State University
> <absthumbpatch.diff>_______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits