<div dir="ltr">When I read the LLVM IR manual, the description of the Variable Argument Handling Intrinsics and the va_arg instruction lead me to expect IR that uses va_arg instead of loads and register references.<div>I was try to get hold of the IR before it becomes too machine dependent, i.e., before code generation, but it seems as though va_arg is replaced/expanded very early, before most optimization.</div><div>Is there a way I can suppress that expansion? I'd really like to get hold of the IR with the va_args intact.</div><div><br></div><div>My other question regarding -arch and -march I think involves a misleading (imo) or out-of-date man page.</div><div>Using -target is an effective way to get code for other machines.</div><div><br></div><div>Preston</div><div><br></div></div><div class="gmail_extra"><br><div class="gmail_quote">On Tue, Sep 29, 2015 at 1:06 AM, mats petersson <span dir="ltr"><<a href="mailto:mats@planetcatfish.com" target="_blank">mats@planetcatfish.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr"><div>What is it you are actually trying to do? <br><br>Varargs are architecture dependent, so will need to take into account the calling convention of the architecture it is on.<br><br></div><div>Your errors about 'mips' is probably based on the build of clang on your machine?<br></div><div><br>--<br></div>Mats<br></div><div class="gmail_extra"><br><div class="gmail_quote"><div><div class="h5">On 29 September 2015 at 00:07, Preston Briggs via llvm-dev <span dir="ltr"><<a href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a>></span> wrote:<br></div></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div><div class="h5"><div dir="ltr">When I use clang on an x86-64 to spit out the LLVM, like this<div><br></div><blockquote style="margin:0px 0px 0px 40px;border:none;padding:0px"><div><font face="monospace, monospace">clang -O -S -emit-llvm varargstest.c</font></div></blockquote><div><br></div><div>where varargstest.c looks like this</div><div><br></div><blockquote style="margin:0px 0px 0px 40px;border:none;padding:0px"><div><div><font face="monospace, monospace">int add_em_up(int count, ...) {</font></div></div><div><div><font face="monospace, monospace"> va_list ap;</font></div></div><div><div><font face="monospace, monospace"> int i, sum;</font></div></div><div><div><font face="monospace, monospace"> va_start(ap, count);</font></div></div><div><div><font face="monospace, monospace"> sum = 0;</font></div></div><div><div><font face="monospace, monospace"> for (i = 0; i < count; i++)</font></div></div><div><div><font face="monospace, monospace"> sum += va_arg(ap, int);</font></div></div><div><div><font face="monospace, monospace"> va_end(ap);</font></div></div><div><div><font face="monospace, monospace"> return sum;</font></div></div><div><div><font face="monospace, monospace">}</font></div></div></blockquote><div><br></div><div>I see LLVM that looks like it's been customized for the x86-64,</div><div>versus the varargs stuff I was led to expect from the LLVM IR documentation.</div><div><br></div><blockquote style="margin:0 0 0 40px;border:none;padding:0px"><div><div><font face="monospace, monospace">define i32 @add_em_up(i32 %count, ...) #0 {</font></div></div><div><div><font face="monospace, monospace">entry:</font></div></div><div><div><font face="monospace, monospace"> %ap = alloca [1 x %struct.__va_list_tag], align 16</font></div></div><div><div><font face="monospace, monospace"> %arraydecay1 = bitcast [1 x %struct.__va_list_tag]* %ap to i8*</font></div></div><div><div><font face="monospace, monospace"> call void @llvm.va_start(i8* %arraydecay1)</font></div></div><div><div><font face="monospace, monospace"> %cmp7 = icmp sgt i32 %count, 0</font></div></div><div><div><font face="monospace, monospace"> br i1 %cmp7, label %<a href="http://for.body.lr.ph" target="_blank">for.body.lr.ph</a>, label %for.end</font></div></div><div><div><font face="monospace, monospace"><br></font></div></div><div><div><font face="monospace, monospace"><a href="http://for.body.lr.ph" target="_blank">for.body.lr.ph</a>: ; preds = %entry</font></div></div><div><div><font face="monospace, monospace"> %gp_offset_p = getelementptr inbounds [1 x %struct.__va_list_tag]* %ap, i64 0, i64 0, i32 0</font></div></div><div><div><font face="monospace, monospace"> %0 = getelementptr inbounds [1 x %struct.__va_list_tag]* %ap, i64 0, i64 0, i32 3</font></div></div><div><div><font face="monospace, monospace"> %overflow_arg_area_p = getelementptr inbounds [1 x %struct.__va_list_tag]* %ap, i64 0, i64 0, i32 2</font></div></div><div><div><font face="monospace, monospace"> %gp_offset.pre = load i32* %gp_offset_p, align 16</font></div></div><div><div><font face="monospace, monospace"> br label %for.body</font></div></div><div><div><font face="monospace, monospace"><br></font></div></div><div><div><font face="monospace, monospace">for.body: ; preds = %vaarg.end, %<a href="http://for.body.lr.ph" target="_blank">for.body.lr.ph</a></font></div></div><div><div><font face="monospace, monospace"> %gp_offset = phi i32 [ %gp_offset.pre, %<a href="http://for.body.lr.ph" target="_blank">for.body.lr.ph</a> ], [ %gp_offset10, %vaarg.end ]</font></div></div><div><div><font face="monospace, monospace"> %sum.09 = phi i32 [ 0, %<a href="http://for.body.lr.ph" target="_blank">for.body.lr.ph</a> ], [ %add, %vaarg.end ]</font></div></div><div><div><font face="monospace, monospace"> %i.08 = phi i32 [ 0, %<a href="http://for.body.lr.ph" target="_blank">for.body.lr.ph</a> ], [ %inc, %vaarg.end ]</font></div></div><div><div><font face="monospace, monospace"> %fits_in_gp = icmp ult i32 %gp_offset, 41</font></div></div><div><div><font face="monospace, monospace"> br i1 %fits_in_gp, label %vaarg.in_reg, label %vaarg.in_mem</font></div></div><div><div><font face="monospace, monospace"><br></font></div></div><div><div><font face="monospace, monospace">vaarg.in_reg: ; preds = %for.body</font></div></div><div><div><font face="monospace, monospace"> %reg_save_area = load i8** %0, align 16</font></div></div><div><div><font face="monospace, monospace"> %1 = sext i32 %gp_offset to i64</font></div></div><div><div><font face="monospace, monospace"> %2 = getelementptr i8* %reg_save_area, i64 %1</font></div></div><div><div><font face="monospace, monospace"> %3 = add i32 %gp_offset, 8</font></div></div><div><div><font face="monospace, monospace"> store i32 %3, i32* %gp_offset_p, align 16</font></div></div><div><div><font face="monospace, monospace"> br label %vaarg.end</font></div></div><div><div><font face="monospace, monospace"><br></font></div></div><div><div><font face="monospace, monospace">vaarg.in_mem: ; preds = %for.body</font></div></div><div><div><font face="monospace, monospace"> %overflow_arg_area = load i8** %overflow_arg_area_p, align 8</font></div></div><div><div><font face="monospace, monospace"> %overflow_arg_area.next = getelementptr i8* %overflow_arg_area, i64 8</font></div></div><div><div><font face="monospace, monospace"> store i8* %overflow_arg_area.next, i8** %overflow_arg_area_p, align 8</font></div></div><div><div><font face="monospace, monospace"> br label %vaarg.end</font></div></div><div><div><font face="monospace, monospace"><br></font></div></div><div><div><font face="monospace, monospace">vaarg.end: ; preds = %vaarg.in_mem, %vaarg.in_reg</font></div></div><div><div><font face="monospace, monospace"> %gp_offset10 = phi i32 [ %3, %vaarg.in_reg ], [ %gp_offset, %vaarg.in_mem ]</font></div></div><div><div><font face="monospace, monospace"> %<a href="http://vaarg.addr.in" target="_blank">vaarg.addr.in</a> = phi i8* [ %2, %vaarg.in_reg ], [ %overflow_arg_area, %vaarg.in_mem ]</font></div></div><div><div><font face="monospace, monospace"> %vaarg.addr = bitcast i8* %<a href="http://vaarg.addr.in" target="_blank">vaarg.addr.in</a> to i32*</font></div></div><div><div><font face="monospace, monospace"> %4 = load i32* %vaarg.addr, align 4</font></div></div><div><div><font face="monospace, monospace"> %add = add nsw i32 %4, %sum.09</font></div></div><div><div><font face="monospace, monospace"> %inc = add nsw i32 %i.08, 1</font></div></div><div><div><font face="monospace, monospace"> %exitcond = icmp eq i32 %inc, %count</font></div></div><div><div><font face="monospace, monospace"> br i1 %exitcond, label %for.end, label %for.body</font></div></div><div><div><font face="monospace, monospace"><br></font></div></div><div><div><font face="monospace, monospace">for.end: ; preds = %vaarg.end, %entry</font></div></div><div><div><font face="monospace, monospace"> %sum.0.lcssa = phi i32 [ 0, %entry ], [ %add, %vaarg.end ]</font></div></div><div><div><font face="monospace, monospace"> call void @llvm.va_end(i8* %arraydecay1)</font></div></div><div><div><font face="monospace, monospace"> ret i32 %sum.0.lcssa</font></div></div><div><div><font face="monospace, monospace">}</font></div></div></blockquote><div><br></div><div>Notice at the bottom of the block labeled "for.body" that there's a test that determines</div><div>whether to look for an argument on the stack or in a register. I see something similar w/ or w/o the -O flag.</div><div><br></div><div>This isn't what I was led to expect by the LLVM IR documentation.</div><div>Is there a way to avoid this "premature" optimization?</div><div>I tried things like -arch mips and -march=mips but get complaints</div><div>about unrecognized flags (-arch) or unknown target architecture CPU 'mips'.</div><div>Why's that? The man page suggests both should work.</div><div><br></div><div>Thanks,</div><div>Preston</div><div><br></div><div><br></div><div><br></div></div>
<br></div></div>_______________________________________________<br>
LLVM Developers mailing list<br>
<a href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a><br>
<a href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev" rel="noreferrer" target="_blank">http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev</a><br>
<br></blockquote></div><br></div>
</blockquote></div><br></div>