fabi/Xenia-Canary
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[CPU] Fix f16 pack rounding and SHORT_2 test input

Browse Source 
Add round-to-nearest-even to the fast float16 pack path by folding a
0xFFF rounding bias into XMMF16PackLCPI0 and extracting bit 13 of the
source as the tie-breaker, matching the software fallback behavior.

Fix PACK_SHORT_2 test to use pre-biased float input (0x40400000 = 3.0)
as the hardware expects, rather than raw 0.0f which is out of range.
This commit is contained in:
Herman S.
2026-02-12 09:59:58 +09:00
parent bb70e5c651
commit 11815400cd
3 changed files with 20 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
src/xenia/cpu/testing/pack_test.cc
View File

@@ -89,11 +89,15 @@ TEST_CASE("PACK_SHORT_2", "[instr]") {
StoreVR(b, 3, b.Pack(LoadVR(b, 4), PACK_TYPE_SHORT_2));
b.Return();
});
test.Run([](PPCContext* ctx) { ctx->v[4] = vec128i(0); },
[](PPCContext* ctx) {
auto result = ctx->v[3];
REQUIRE(result == vec128i(0));
});
// SHORT_2 operates on pre-biased floats near 3.0 (0x40400000 = short 0)
test.Run(
[](PPCContext* ctx) {
ctx->v[4] = vec128i(0x40400000, 0x40400000, 0, 0);
},
[](PPCContext* ctx) {
auto result = ctx->v[3];
REQUIRE(result == vec128i(0));
});
test.Run(
[](PPCContext* ctx) {
ctx->v[4] = vec128i(0x43817E00, 0xC37CFC00, 0, 0);