happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

refactor: replace ggml_hexagon_mul_mat_id with template-based binary operation for improved flexibility

This commit is contained in:
chraac 2025-11-24 12:57:42 +08:00
parent d12247a60d
commit 020f6bf3f2
1 changed files with 7 additions and 83 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

90
ggml/src/ggml-hexagon/ggml-hexagon.cpp
View File

@ -2322,85 +2322,6 @@ static void hex_dump_dspbuf(const struct ggml_tensor * t, const dspqueue_buffer
(unsigned int) d->size);
}
static void ggml_hexagon_mul_mat_id(const struct ggml_tensor * op, uint32_t flags) {
const struct ggml_tensor * src0 = op->src[0];
const struct ggml_tensor * src1 = op->src[1];
const struct ggml_tensor * src2 = op->src[2];
const struct ggml_tensor * dst = op;
uint64_t t1, t2;
t1 = ggml_time_us();
// Construct HTP message
htp_general_req req;
req.op = HTP_OP_MUL_MAT_ID;
req.flags = flags;
init_htp_tensor(&req.src0, src0);
init_htp_tensor(&req.src1, src1);
init_htp_tensor(&req.src2, src2);
init_htp_tensor(&req.dst, dst);
// Use opmask to override flags
if (!(opt_opmask & HTP_OPMASK_QUANTIZE)) {
req.flags |= HTP_OPFLAGS_SKIP_QUANTIZE;
}
if (!(opt_opmask & HTP_OPMASK_COMPUTE)) {
req.flags |= HTP_OPFLAGS_SKIP_COMPUTE;
}
dspqueue_buffer bufs[4];
// First buffer Weights.
// The content is static, there is no need to do any cache management
dspqueue_buffers_init(bufs, src0, false, false);
// Second buffer Input Activations. This is a buffer that the CPU
// writes and the DSP reads, so we'll need to flush CPU caches and
// invalidate DSP ones. On platforms with I/O coherency support the
// framework will automatically skip cache operations where possible.
dspqueue_buffers_init(&bufs[1], src1, true, true);
// Third buffer expert IDs. This is a buffer that the CPU
// writes and the DSP reads, so we'll need to flush CPU caches and
// invalidate DSP ones. On platforms with I/O coherency support the
// framework will automatically skip cache operations where possible.
dspqueue_buffers_init(&bufs[2], src2, true, true);
// Forth buffer Output Activations. We'll handle DSP
// cache maintenance in the response message but need to flush
// CPU caches to ensure any previously written dirty lines are
// written out before writes from the DSP start.
dspqueue_buffers_init(&bufs[3], dst, true, false);
auto * sess = get_session_from_tensor(src0);
if (opt_verbose) {
hex_print_op_info(op, sess, req.flags);
if (opt_verbose > 1) {
hex_dump_dspbuf(src0, &bufs[0]);
hex_dump_dspbuf(src1, &bufs[1]);
hex_dump_dspbuf(src2, &bufs[2]);
hex_dump_dspbuf(dst, &bufs[3]);
}
}
if ((opt_opmask & HTP_OPMASK_QUEUE)) {
sess->enqueue(req, bufs, 4, opt_opsync);
}
t2 = ggml_time_us();
HEX_PROFILE(
"ggml-hex: %s matmul-id %s %u:%u:%u:%u x %s %u:%u:%u:%u (%s %u:%u:%u:%u) -> %s %u:%u:%u:%u : op-usec %u "
"op-cycles %u op-pkts %u (%f) call-usec %llu\n",
sess->name.c_str(), src0->name, (uint32_t) src0->ne[0], (uint32_t) src0->ne[1], (uint32_t) src0->ne[2],
(uint32_t) src0->ne[3], src1->name, (uint32_t) src1->ne[0], (uint32_t) src1->ne[1], (uint32_t) src1->ne[2],
(uint32_t) src1->ne[3], src2->name, (uint32_t) src2->ne[0], (uint32_t) src2->ne[1], (uint32_t) src2->ne[2],
(uint32_t) src2->ne[3], dst->name, (uint32_t) dst->ne[0], (uint32_t) dst->ne[1], (uint32_t) dst->ne[2],
(uint32_t) dst->ne[3], sess->prof_usecs, sess->prof_cycles, sess->prof_pkts,
(float) sess->prof_cycles / sess->prof_pkts, (unsigned long long) t2 - t1);
}
template <bool _IsSrc0Constant> static void ggml_hexagon_binary(const struct ggml_tensor * op, uint32_t flags) {
const struct ggml_tensor * node = op;
const struct ggml_tensor * src0 = node->src[0];
@ -2493,7 +2414,7 @@ template <bool _IsSrc0Constant> static void ggml_hexagon_binary(const struct ggm
(float) sess->prof_cycles / sess->prof_pkts, (unsigned long long) t2 - t1);
}
static void ggml_hexagon_add_id(const struct ggml_tensor * op, uint32_t flags) {
template <bool _IsSrc0Constant> static void ggml_hexagon_binary_id(const struct ggml_tensor * op, uint32_t flags) {
const struct ggml_tensor * node = op;
const struct ggml_tensor * src0 = node->src[0];
const struct ggml_tensor * src1 = node->src[1];
@ -2518,6 +2439,9 @@ static void ggml_hexagon_add_id(const struct ggml_tensor * op, uint32_t flags) {
}
switch (node->op) {
case GGML_OP_MUL_MAT_ID:
req.op = HTP_OP_MUL_MAT_ID;
break;
case GGML_OP_ADD_ID:
req.op = HTP_OP_ADD_ID;
break;
@ -2532,7 +2456,7 @@ static void ggml_hexagon_add_id(const struct ggml_tensor * op, uint32_t flags) {
dspqueue_buffer bufs[4];
// First buffer = input activations
dspqueue_buffers_init(bufs, src0, true, true);
dspqueue_buffers_init(bufs, src0, !_IsSrc0Constant, !_IsSrc0Constant);
// Second buffer = experts bias
dspqueue_buffers_init(&bufs[1], src1, true, true);
// Third buffer = activated experts
@ -2875,7 +2799,7 @@ static ggml_status ggml_backend_hexagon_graph_compute(ggml_backend_t backend, gg
prev_quant_op = node;
break;
case GGML_OP_MUL_MAT_ID:
ggml_hexagon_mul_mat_id(node, flags);
ggml_hexagon_binary_id<true>(node, flags);
prev_quant_op = node;
break;
case GGML_OP_MUL:
@ -2884,7 +2808,7 @@ static ggml_status ggml_backend_hexagon_graph_compute(ggml_backend_t backend, gg
ggml_hexagon_binary<false>(node, flags);
break;
case GGML_OP_ADD_ID:
ggml_hexagon_add_id(node, flags);
ggml_hexagon_binary_id<false>(node, flags);
break;
case GGML_OP_RMS_NORM:
ggml_hexagon_unary(node, flags);