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
6 lines
273 B
Plaintext
6 lines
273 B
Plaintext
// Exact single-buffer enc1: reload the original input for the residual, preserve WGP mode.
|
|
// d4r-build-flags: -DPOS_VSHARE1 -DPWIN_VEC_SCALAR -DPWIN_PK_ELEM
|
|
#define PWIN_F32ACC
|
|
#include "pwin_pos_recompute.h"
|
|
POS_RECOMPUTE_ENCODER(dltss_pwin_enc1_layer, 2, 64, 64, 0, 1)
|