Assembly listing problems with version 6.6?

[hbell]

The notes to the CVF version 6.6 upgrade indicate there is a problem with the assembly language compile listing and that it may be fixed in a future release. Does anyone know the nature of this problem? It seems that this could signify a serious fundamental problem with the compiler.

[Steven_L_Intel1]

This is an obscure option that affects only those who ask for a separate assembly listing and then try to run it through the MASM assembler. It has no impact on normal users of the compiler. Keep in mind that CVF generates object code directly - it does not feed assembly code through a separate assembler.

Offhand, I don't know what this particular issue is - if you think it's important to you (you compile CVF-generated .asm files with MASM), let me know and I'll try to find out.

Steve

[Intel_C_Intel]

The main problem is that the source listing looks more appropriate for ASAXP than for MASM. Just looking at some of the error messages MASM 6.14 tosses out, it looks like the processor type is specified to be .486 for a listing of Athlon code. I could fix this with the directives .686 .MMX .K3D and that got rid of a couple of hundred error messages. MASM doesn't like some of symbol names like .bss$, .data$, .literal$, .drectve$ and I can't tell you how to fix that because the linker is evidently intended to do different things with these data areas. There is also the annoying thing about MASM that it doesn't like syntax like cmp dword ptr -4[ebp], 0 and all of these would have to be changed to cmp dword ptr [ebp-4], 0 to assemble. If you ever did get a source listing to assemble the first thing you would want to do with it is to rewrite it from scratch anyway. Would you ever have sequences like shl eax, 3 / add eax, 7 / and eax, -8 in your code or would you really try to use the x87 FP stack like it were a register file, or would you never use scaled indices in addressing memory?

[hbell]

Steve,
Thanks for the reply and offer to investigate the details. This issue is important to me. I sometimes re-compile CVF-generated .asm files with MASM. Among other reasons for doing this I like to get an instruction-level clock count for a better handle on code efficiency (low level profiling). Would you please find out the details if somebody else doesn't provide the answer. I have a number of earlier postings to this Forum on MASM topics.

Harry Bell

[TimP]

Testing modified CVF code with MASM has been one of my methods for testing ideas on improvements in code generation. MASM is by no means my favorite assembler, but I need something which links with CVF.

[Steven_L_Intel1]

Ok - Stan is back and I bugged him about this. He says that this text has been in the release notes for a long time (at least back to 6.0, by my checking), and refers to a bug that we fixed a while back, but we never took out the release note. So... Never Mind! If you discover a problem with the assembly code output, let us know at vf-support@compaq.com

Steve