[llvm-dev] Missing code depending on a #ifdef within the .ll file

Wed Dec 18 15:02:56 PST 2019

               if (width == 0)
became
               %1 = icmp eq i32 %width, 0
which is straightforward; and
               else if (width > PNG_UINT_31_MAX)
became (assuming the constant is 0x7fffffff)
               %3 icmp slt i32 %width, 0
(note this is a signed comparison with 0, which is equivalent to the comparison in the source)
and finally
               if (width > png_ptr->user_width_max)
became
               %8 = icmp ult i32 %7, %width
after loading the png_ptr field into %7 in the previous instruction.

Does that help?
--paulr

From: llvm-dev <llvm-dev-bounces at lists.llvm.org> On Behalf Of David Blaikie via llvm-dev
Sent: Wednesday, December 18, 2019 5:30 PM
To: Alberto Barbaro <barbaro.alberto at gmail.com>
Cc: llvm-dev <llvm-dev at lists.llvm.org>
Subject: Re: [llvm-dev] Missing code depending on a #ifdef within the .ll file

I'm sorry, I don't understand the question - the source code shows 3 calls to png_warning (one of them from the if block you're interested in), and the IR shows 3 calls to png_warning - so it looks like the body of the if block is included in the IR you've shown?

On Wed, Dec 18, 2019 at 2:24 PM Alberto Barbaro <barbaro.alberto at gmail.com<mailto:barbaro.alberto at gmail.com>> wrote:
Hi David,
My question is: why both #ifdef and #else branches are missing? I think at least one of the two should be present... In fact there is a case where the width could be greater then
PNG_USER_WIDTH_MAX but not greater then PNG_UINT_31_MAX. That's why I was expecting at least one of the two...

Thanks
Alberto

On Wed, Dec 18, 2019, 22:12 David Blaikie <dblaikie at gmail.com<mailto:dblaikie at gmail.com>> wrote:
Not sure I understand the question - but usually a compiler would, on one compilation, only compile one versoin of the code with one set of defines.

There could be invalid code (it could literally be junk - or it could be code that's only valid when the define is true/false, etc) in an unreachable ifdef, so it doesn't make sense to compile it.

On Wed, Dec 18, 2019 at 2:09 PM Alberto Barbaro via llvm-dev <llvm-dev at lists.llvm.org<mailto:llvm-dev at lists.llvm.org>> wrote:
Hi all,
I have managed to compile libpng using wllvm and obtain the IR of pngpixel ( small tool which is part of libpng ).

libpng has a function called png_check_IHDR:

void /* PRIVATE */
png_check_IHDR(png_const_structrp png_ptr,
   png_uint_32 width, png_uint_32 height, int bit_depth,
   int color_type, int interlace_type, int compression_type,
   int filter_type)
{
   int error = 0;

   /* Check for width and height valid values */
   if (width == 0)
   {
      png_warning(png_ptr, "Image width is zero in IHDR");
      error = 1;
   }
   else if (width > PNG_UINT_31_MAX)
   {
      png_warning(png_ptr, "Invalid image width in IHDR");
      error = 1;
   }
   else
   {
#     ifdef PNG_SET_USER_LIMITS_SUPPORTED
      if (width > png_ptr->user_width_max)
#     else
      if (width > PNG_USER_WIDTH_MAX)
#     endif
      {
         png_warning(png_ptr, "Image width exceeds user limit in IHDR");
         error = 1;
      }
   }
...
}

Unfortunately the relative IR does not seem to have the branch highlighted in bold:

; Function Attrs: nounwind uwtable
define void @png_check_IHDR(%struct.png_struct_def* noalias %png_ptr, i32 %width, i32 %height, i32 %bit_depth, i32 %color_type, i32 %interlace_type, i32 %compression_type, i32 %filter_type) #5 {
  %1 = icmp eq i32 %width, 0
  br i1 %1, label %.thread, label %2

.thread:                                          ; preds = %0
  tail call void @png_warning(%struct.png_struct_def* %png_ptr, i8* getelementptr inbounds ([28 x i8], [28 x i8]* @.str50, i64 0, i64 0)) #15
  br label %5

; <label>:2:                                      ; preds = %0
  %3 = icmp slt i32 %width, 0
  br i1 %3, label %4, label %5

; <label>:4:                                      ; preds = %2
  tail call void @png_warning(%struct.png_struct_def* %png_ptr, i8* getelementptr inbounds ([28 x i8], [28 x i8]* @.str51, i64 0, i64 0)) #15
  br label %5

; <label>:5:                                      ; preds = %4, %2, %.thread
  %error.1 = phi i32 [ 1, %4 ], [ 0, %2 ], [ 1, %.thread ]
  %6 = getelementptr inbounds %struct.png_struct_def, %struct.png_struct_def* %png_ptr, i64 0, i32 132
  %7 = load i32, i32* %6, align 4, !tbaa !22
  %8 = icmp ult i32 %7, %width ; if (width > PNG_UINT_31_MAX)
  br i1 %8, label %9, label %10

; <label>:9:                                      ; preds = %5
  tail call void @png_warning(%struct.png_struct_def* %png_ptr, i8* getelementptr inbounds ([39 x i8], [39 x i8]* @.str53, i64 0, i64 0)) #15
  br label %10

; <label>:10:                                     ; preds = %9, %5
  %error.3 = phi i32 [ 1, %9 ], [ %error.1, %5 ]
  %11 = icmp eq i32 %height, 0 ; first check on the height
  br i1 %11, label %.thread27, label %12

Does anybody know why the else branch is completely missing?

Thanks,
Alberto

_______________________________________________
LLVM Developers mailing list
llvm-dev at lists.llvm.org<mailto:llvm-dev at lists.llvm.org>
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20191218/07ebbd69/attachment-0001.html>