[libc] Align src buffer instead of dst buffer

[gchatelet]

Summary

We used to align destination buffer instead of source buffer for the loop of block copy.
This is a mistake.

[avieira]

Hi Guillaume,

Sorry for the late reply, I was looking at some other things and then holidays happened. I started looking at making some changes to CopyAlignedBlocks just like we discussed earlier. One of them is this one and I noticed this makes the MemcpyUtilsTest.CopyAlignedBlocks fail in libc_string_unittests. Specifically the ones that test destination misalignment. It seems the testing infrastructure keeps track of what bytes were read and written. I checked and making the source the misaligned input and switching the read and written expected traces makes it passs. Which makes sense I guess.

Would be good to make the change at the same time though.

[gchatelet]

• Address comments and rebase

[gchatelet]

In D93457#2481596, @avieira wrote:

Hi Guillaume,

Sorry for the late reply, I was looking at some other things and then holidays happened. I started looking at making some changes to CopyAlignedBlocks just like we discussed earlier. One of them is this one and I noticed this makes the MemcpyUtilsTest.CopyAlignedBlocks fail in libc_string_unittests. Specifically the ones that test destination misalignment. It seems the testing infrastructure keeps track of what bytes were read and written. I checked and making the source the misaligned input and switching the read and written expected traces makes it passs. Which makes sense I guess.

Would be good to make the change at the same time though.

Thx for noticing, it should be good now. I'll move you back to reviewers if you don't mind.

[avieira]

Sure, LGTM. What is the usual convention for Libc reviewing? Should I 'accept revision' if I agree with it even though I'm not a maintainer?

[gchatelet]

In D93457#2481610, @avieira wrote:

Sure, LGTM. What is the usual convention for Libc reviewing? Should I 'accept revision' if I agree with it even though I'm not a maintainer?

I don't know TBH :-)
I think it's fine if you accept revision even if you're not a maintainer.

libc/src/string/memory_utils/memcpy_utils.h

 // Precondition: `count > 2 * kBlockSize` for efficiency.
 //               `count >= kBlockSize` for correctness.
 template <size_t kBlockSize>
 static void CopyAlignedBlocks(char *__restrict dst, const char *__restrict src,
                               size_t count) {
   CopyBlock<kBlockSize>(dst, src); // Copy first block
 
   // Copy aligned blocks
-  const size_t ofla = offset_from_last_aligned<kBlockSize>(dst);
+  const size_t ofla = offset_from_last_aligned<kBlockSize>(src);
   const size_t limit = count + ofla - kBlockSize;
   for (size_t offset = kBlockSize; offset < limit; offset += kBlockSize)
     CopyBlock<kBlockSize>(dst - ofla + offset, src - ofla + offset);
 
   CopyLastBlock<kBlockSize>(dst, src, count); // Copy last block
 }
 
 } // namespace __llvm_libc
 
 #endif //  LIBC_SRC_STRING_MEMORY_UTILS_MEMCPY_UTILS_H