ayo
22026ede12
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
4 lines
157 B
Plaintext
4 lines
157 B
Plaintext
// NG = 8 (32 waves per window): with the padded LDS rows faster than NG = 4 here
|
|
#include "pwin_wide.h"
|
|
WIDE_DECODER(dltss_pwin_dec4_layer, 4, 128, 160, 8)
|