derek/gem5
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

arch-vega: Fix SDWA dst select (#1120)

Browse Source 
The destination select should take a value of the selection size (dword,
word, or byte) starting at bit 0, move that to the selected destination,
and then apply the unused constraint (DST_U) to the remaining word or
bytes. Currently the code is selecting the word/byte currently being
iterated over, rather than the least significant word/byte. As a result,
any selection that is not word 0 or byte 0 will be replaced with the
original destination value at those bits. This results in the wrong
value.

This commit changes the orig bits to be the original dest value at the
lowest word / byte location. Tested with the mfma_i32_16x16x16i8 example
which uses an SDWA V_OR_B32 to pack i8 values into VGPRs for the MFMA.

Change-Id: I54ed819479a25fa9276d29a8f14f0fea7fd71afe
This commit is contained in:
Matthew Poremba
2024-05-10 08:49:13 -07:00
committed by GitHub
parent 8c4d5f8e27
commit e3c2a322a1
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
src/arch/amdgpu/vega/insts/inst_util.hh
View File

@@ -701,7 +701,7 @@ namespace VegaISA
if (sel < SDWA_WORD_0) { // we are selecting 1 byte
// if we sign extended depends on upper-most bit of byte 0
signExt = (signExt &&
(bits(currDstVal, VegaISA::MSB_PER_WORD, 0) & 0x80));
(bits(currDstVal, VegaISA::MSB_PER_BYTE, 0) & 0x80));
for (int byte = 0; byte < 4; ++byte) {
low_bit = byte * VegaISA::BITS_PER_BYTE;
@@ -714,7 +714,7 @@ namespace VegaISA
3. byte > sel && signExt: we're sign extending and
this byte is one of the bytes we need to sign extend
*/
origBits_thisByte = bits(origDstVal, high_bit, low_bit);
origBits_thisByte = bits(origDstVal, VegaISA::MSB_PER_BYTE, 0);
currBits_thisByte = bits(currDstVal, high_bit, low_bit);
newBits = ((byte == sel) ? origBits_thisByte :
((preserve) ? currBits_thisByte :
@@ -739,7 +739,7 @@ namespace VegaISA
3. word > (sel & 1) && signExt: we're sign extending and
this word is one of the words we need to sign extend
*/
origBits_thisWord = bits(origDstVal, high_bit, low_bit);
origBits_thisWord = bits(origDstVal, VegaISA::MSB_PER_WORD, 0);
currBits_thisWord = bits(currDstVal, high_bit, low_bit);
newBits = ((word == (sel & 0x1)) ? origBits_thisWord :
((preserve) ? currBits_thisWord :