Port K kernel tuning and M gfx1101 flags
Preset K, byte-identical to the previous output in all captured buffers and the harness image, measured on an RX 7700 XT: - LDS row padding so WMMA operand loads hit distinct banks - WGP mode for enc1/enc2/dec1/dec2, ds_bpermute half-wave exchange for enc0/dec0/dec3, more waves per window in enc4/dec4/dec5 - cheaper operand plumbing in common/wmma_layout.h (packed f16 conversions, no dynamic vector indexing), raw buffer weight loads, accumulator-init vectors prepared once - enc1 as a single-buffer recompute layer, dec1 Q/K/V chains split - dec5 as four dependent phase kernels
This commit is contained in:
1 parent
8ce8618181
commit
22026ede12
45 files changed
+4514
-354
No files matched your search
@@ -1,2 +1,4 @@
|
||||
#include "pwin_wide.h"
|
||||
WIDE_PLAIN(dltss_pwin_dec5_layer, 8, 160, 4)
|
||||
// dec5 (bottleneck, H 8, C 160) as four phase kernels (pwin_phase.h). Needs the patched
|
||||
// ZLUDA hook (post kernels). MAXW 64 windows (1440p: 9).
|
||||
#include "pwin_phase.h"
|
||||
PHASE_PLAIN(dltss_pwin_dec5_layer, 8, 160, 64, 128, 96, 384, 96)
|
||||
Reference in new issue
Block a user