mirrors/ZLUDA
0
0
Fork 0
mirror of https://github.com/vosen/ZLUDA.git synced 2025-09-27 11:49:04 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

Improve constant folding for sign extension

Browse source
This commit is contained in:
Violet 2025-09-23 18:16:03 +00:00
commit da9c0e77d9
2 changed files with 14 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

BIN
ptx/lib/zluda_ptx_impl.bc
View file

Binary file not shown.

15
ptx/lib/zluda_ptx_impl.cpp
View file

@ -812,6 +812,14 @@ typedef uint32_t ShflSyncResult __attribute__((ext_vector_type(2)));
{
uint8_t sel = static_cast<uint8_t>(s >> (i * 4));
uint8_t addr = sel & 0x7;
if (sel & 0x8)
{
if (addr % 2 == 1)
{
v_perm_selector.u8x4[i] = 0x8 + addr / 2;
continue;
}
}
v_perm_selector.u8x4[i] = addr;
}
@ -821,9 +829,14 @@ typedef uint32_t ShflSyncResult __attribute__((ext_vector_type(2)));
for (size_t i = 0; i < 4; i++)
{
uint8_t sel = static_cast<uint8_t>(s >> (i * 4));
uint8_t addr = sel & 0x7;
if (sel & 0x8)
{
output.u8x4[i] = (output.u8x4[i] & 0x80) * 0xff;
if (addr % 2 != 1)
{
output.u8x4[i] = (output.u8x4[i] & 0x80) * 0xff;
continue;
}
}
}