7 //      http://www.apache.org/licenses/LICENSE-2.0
17 // We are using (a modified form of) adler-32 checksum algorithm instead
18 // of CRC since adler-32 is faster than CRC.
19 // (Comparison: http://guru.multimedia.cx/crc32-vs-adler32/)
20 // This form of adler is bit modified, instead of treating the data in
21 // units of bytes, 32-bit data is taken as a unit and two 64-bit
25 // Adler-32 implementation:
26 //   Data is treated as 1-byte numbers and,
27 //   there are two 16-bit numbers a and b
35 // Adler-64+64 implementation:
37 //   Data is treated as 32-bit numbers and whole data is separated into two
88   // Use this data wrapper to access memory with 64bit read/write.  in CalculateAdlerChecksum()
119   checksum->Set(a1, a2, b1, b2);  in CalculateAdlerChecksum()
126   // Use this data wrapper to access memory with 64bit read/write.  in AdlerMemcpyC()
159   checksum->Set(a1, a2, b1, b2);  in AdlerMemcpyC()
167   // Use this data wrapper to access memory with 64bit read/write.  in AdlerMemcpyWarmC()
221   checksum->Set(a1, a2, b1, b2);  in AdlerMemcpyWarmC()
231 // Pull a bit of tricky preprocessing to make the inline asm both  in AdlerMemcpyAsm()
232 // 32 bit and 64 bit.  in AdlerMemcpyAsm()
269   // Number of 32-bit words which are not added to a1/a2 in the main loop.  in AdlerMemcpyAsm()
270   uint32 remaining_words = (size_in_bytes % 48) / 4;  in AdlerMemcpyAsm()
272   // Since we are moving 48 bytes at a time number of iterations = total size/48  in AdlerMemcpyAsm()
274   uint32 num_of_48_byte_units = size_in_bytes / 48;  in AdlerMemcpyAsm()
287       "movdqu   48(" rAX "), %%xmm6;"  in AdlerMemcpyAsm()
290       // Start of the loop which copies 48 bytes from source to dst each time.  in AdlerMemcpyAsm()
336       // Increment ESI and EDI by 48 bytes and decrement counter by 1.  in AdlerMemcpyAsm()
337       "add $48, " rSI ";"  in AdlerMemcpyAsm()
338       "add $48, " rDI ";"  in AdlerMemcpyAsm()
344       // Now only remaining_words 32-bit words are left.  in AdlerMemcpyAsm()
356       // Otherwise just go on and copy data in chunks of 4-words at a time till  in AdlerMemcpyAsm()
357       // whole data (<48 bytes) is copied.  in AdlerMemcpyAsm()
358       "movdqa  0(" rSI "), %%xmm0;"    // Copy next 4-words to XMM0 and to XMM1.  in AdlerMemcpyAsm()
363       "movntdq %%xmm0,  0(" rDI ");"   // Copy 4-words to destination.  in AdlerMemcpyAsm()
373       // Decrement %rDX by 4 since %rDX is number of 32-bit  in AdlerMemcpyAsm()
374       // words left after considering all 48-byte units.  in AdlerMemcpyAsm()
379       // 64 bit numbers and have to be converted to 64 bit numbers)  in AdlerMemcpyAsm()
394     checksum->Set(checksum_arr[0], checksum_arr[1],  in AdlerMemcpyAsm()
444       // Start of the loop which copies 48 bytes from source to dst each time.  in AdlerMemcpyAsm()
449       // word index 0 and word index 2. We'll swizzle them in a bit.  in AdlerMemcpyAsm()
467       // Repeat this for q8 - q13.  in AdlerMemcpyAsm()
497       // 64 bit numbers and have to be converted to 64 bit numbers)  in AdlerMemcpyAsm()
510     checksum->Set(checksum_arr[0], checksum_arr[1],  in AdlerMemcpyAsm()