<html>
  <head>
    <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
  </head>
  <body text="#000000" bgcolor="#FFFFFF">
    <p><br>
    </p>
    <div class="moz-cite-prefix">On 05/22/2018 10:42 AM, Dávid Bolvanský
      wrote:<br>
    </div>
    <blockquote type="cite"
cite="mid:CAOrgDVN2BNvmDmes4CWiVkX2sRQzEmnO0hUXjiE7Q0c2ifr2Og@mail.gmail.com">
      <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
      <div dir="ltr">Thanks.
        <div><br>
        </div>
        <div>Yes, to <span style="font-size:12.8px">substitute only some
            of the arguments. F</span><span style="font-size:12.8px">ormatting
            used by printf depends on the locale but only for double,
            float types I think - yes, I would not place </span><span
            style="font-size:12.8px">double/float </span><span
            style="font-size:12.8px">constants </span><span
            style="font-size:12.8px">into the format string.</span></div>
      </div>
    </blockquote>
    <br>
    Okay. I think it's true that integers will be the same regardless of
    locale (so long as the ' flag is not used, as that brings in a
    dependence on LC_NUMERIC).<br>
    <br>
    <blockquote type="cite"
cite="mid:CAOrgDVN2BNvmDmes4CWiVkX2sRQzEmnO0hUXjiE7Q0c2ifr2Og@mail.gmail.com">
      <div dir="ltr">
        <div>
          <div>
            <div><span style="font-size:12.8px"><br>
              </span></div>
            <div><span style="font-size:12.8px">Why? To reduce number of
                constants (some of them could be merged into the format
                string) and number of args when
                calling printf/fprintf/sprintf, etc..<br>
              </span></div>
          </div>
        </div>
      </div>
    </blockquote>
    <br>
    Sure, but it seems to me unlikely that this will affect performance.
    Is it a code-size optimization (this actually isn't obvious to me
    because the string representation might be longer than the binary
    form of the constant plus the extra instructions)?<br>
    <br>
     -Hal<br>
    <br>
    <blockquote type="cite"
cite="mid:CAOrgDVN2BNvmDmes4CWiVkX2sRQzEmnO0hUXjiE7Q0c2ifr2Og@mail.gmail.com">
      <div class="gmail_extra"><br>
        <div class="gmail_quote">2018-05-22 16:22 GMT+02:00 Hal Finkel <span
            dir="ltr"><<a href="mailto:hfinkel@anl.gov"
              target="_blank" moz-do-not-send="true">hfinkel@anl.gov</a>></span>:<br>
          <blockquote class="gmail_quote" style="margin:0 0 0
            .8ex;border-left:1px #ccc solid;padding-left:1ex">
            <div text="#000000" bgcolor="#FFFFFF"><span class="">
                <p><br>
                </p>
                <div class="m_7035160539680259771moz-cite-prefix">On
                  05/22/2018 04:32 AM, Dávid Bolvanský via llvm-dev
                  wrote:<br>
                </div>
                <blockquote type="cite">
                  <div dir="ltr">Hello,
                    <div><br>
                    </div>
                    <div>A new patch:</div>
                    <div><a href="https://reviews.llvm.org/D47159"
                        target="_blank" moz-do-not-send="true">https://reviews.llvm.org/<wbr>D47159</a></div>
                    <div><br>
                    </div>
                    <div>proposes transformations like:</div>
                    printf("Hello, %s %d", "world", 123) - >
                    printf("Hello world 123")<br>
                  </div>
                </blockquote>
                <br>
              </span> To clarify, the real question here comes up when
              you can only substitute some of the arguments? If you can
              substitute all of the arguments, then you can turn this
              into a call to puts.<br>
              <br>
              In any case , why do you want to do this? Also, doesn't
              the formatting used by printf depend on the process's
              current locale?<br>
              <br>
               -Hal<br>
              <br>
              <blockquote type="cite"><span class="">
                  <div dir="ltr"><br>
                    As Eli noted:<br>
                    <br>
                    "I'm not sure we can rewrite calls to varargs
                    functions safely in general given the current state
                    of the C ABI rules in LLVM.<br>
                    <br>
                    Sometimes clang does weird things to conform with
                    the ABI rules, because the LLVM type system isn't
                    the same as the C system. For most functions, it's
                    pretty easy to tell it happened: if the IR signature
                    of the function doesn't match the expected
                    signature, something weird happened, so we can just
                    bail out. But varargs functions don't specify a
                    complete signature, so we can't tell if the clang
                    ABI code was forced to do something weird, like
                    split an argument into multiple values, or insert a
                    padding value. For example, for the target
                    mips64-unknown-linux-gnu, a call like
                    printf("asdf%Lf", 1.0L); gets lowered to the
                    following:<br>
                    <br>
                    %call = call i32 (i8*, ...) @printf(i8*
                    getelementptr inbounds ([5 x i8], [5 x i8]* @.str,
                    i32 0, i32 0), i64 undef, fp128
                    0xL00000000000000003FFF0000000<wbr>00000) #2"
                    <div><br>
                    </div>
                    <div><br>
                    </div>
                    <div>I would to hear more suggestions whether it is
                      safe or not. Seems like for mips Clang produces
                      some weird IR, but e.g. x86 IR seems ok.</div>
                    <div><br>
                    </div>
                    <div>Any folks from Clang/LLVM to bring more
                      information about "varargs vs ABI vs LLVM vs
                      Clang"? </div>
                    <div>And whether we can rewrite calls
                      to varargs functions safely under some
                      conditions..</div>
                    <div><br>
                    </div>
                    <div>Thanks</div>
                    <div><br>
                    </div>
                    <div><br>
                    </div>
                  </div>
                  <br>
                  <fieldset
                    class="m_7035160539680259771mimeAttachmentHeader"></fieldset>
                  <br>
                </span>
                <pre>______________________________<wbr>_________________
LLVM Developers mailing list
<a class="m_7035160539680259771moz-txt-link-abbreviated" href="mailto:llvm-dev@lists.llvm.org" target="_blank" moz-do-not-send="true">llvm-dev@lists.llvm.org</a>
<a class="m_7035160539680259771moz-txt-link-freetext" href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev" target="_blank" moz-do-not-send="true">http://lists.llvm.org/cgi-bin/<wbr>mailman/listinfo/llvm-dev</a><span class="HOEnZb"><font color="#888888">
</font></span></pre>
                <span class="HOEnZb"><font color="#888888"> </font></span></blockquote>
              <span class="HOEnZb"><font color="#888888"> <br>
                  <pre class="m_7035160539680259771moz-signature" cols="72">-- 
Hal Finkel
Lead, Compiler Technology and Programming Languages
Leadership Computing Facility
Argonne National Laboratory</pre>
                </font></span></div>
          </blockquote>
        </div>
        <br>
      </div>
    </blockquote>
    <br>
    <pre class="moz-signature" cols="72">-- 
Hal Finkel
Lead, Compiler Technology and Programming Languages
Leadership Computing Facility
Argonne National Laboratory</pre>
  </body>
</html>