nikola/llama_cpp
1
0
Fork 0
Code Issues Pull Requests Actions 30 Packages Projects Releases Wiki Activity

ggml-hexagon: gelu operation (#17921)

Browse Source 
* feat: inital support for gelu using sigmoid approximation

* snapshot: faster gelu using polynomial approximation

* test: disable l2-block prefetch in polynomail approximation

* Revert "test: disable l2-block prefetch in polynomail approximation"

This reverts commit 72339994d45b2bed887e79994403c378d90b62b5.

* Revert "snapshot: faster gelu using polynomial approximation"

This reverts commit 2a787a61d11f9e63e5943a2e6d134b2f0c402ace.

* debug: temporarily disable unnecessary log message for debug purpose

* Feat: optiized unaligned sigmoid_f32

* Feat: larger l2prefetch block

* feat: apply unaligned-load optimization on mul and mul_scalar

* Revert "debug: temporarily disable unnecessary log message for debug purpose"

This reverts commit 84f2f23aa9f17e2fa826db969cd825d0ab192995.

* refactor: cleanup commented unused code

* chore: reformat code with clang-formatter to pass cli test

* Revert "chore: reformat code with clang-formatter to pass cli test"

This reverts commit 952877ec24732b12010c7fa7ed3fc8de4b74e718.

* fix: fix loop overflow

* chore: fix formating ci error
This commit is contained in:
Shouyu
2025-12-17 10:39:32 -08:00
committed by GitHub
parent 4301e27319
commit 4470a0764a
6 changed files with 256 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-hexagon/ggml-hexagon.cpp
+18 -3
View File
@@ -2161,8 +2161,14 @@ static bool ggml_hexagon_supported_activations(const struct ggml_hexagon_session
}
// src0, src1 & dst must be mapped to the same session
if (!hex_supported_buffer(sess, src0, src1, dst)) {
return false;
if(src1){
if (!hex_supported_buffer(sess, src0, src1, dst)) {
return false;
}
}else{
if (!hex_supported_buffer(sess, src0, dst)) {
return false;
}
}
return true;
@@ -2662,6 +2668,10 @@ static void ggml_hexagon_unary(const struct ggml_tensor * op, uint32_t flags) {
req.op = HTP_OP_UNARY_SILU;
supported = true;
}
else if (ggml_get_unary_op(dst) == GGML_UNARY_OP_GELU){
req.op = HTP_OP_UNARY_GELU;
supported = true;
}
break;
case GGML_OP_GLU:
@@ -2677,6 +2687,7 @@ static void ggml_hexagon_unary(const struct ggml_tensor * op, uint32_t flags) {
case GGML_OP_SOFT_MAX:
req.op = HTP_OP_SOFTMAX;
supported = true;
break;
default:
break;
@@ -2956,6 +2967,8 @@ static ggml_status ggml_backend_hexagon_graph_compute(ggml_backend_t backend, gg
case GGML_OP_UNARY:
if (ggml_get_unary_op(node) == GGML_UNARY_OP_SILU) {
ggml_hexagon_unary(node, flags);
} else if (ggml_get_unary_op(node) == GGML_UNARY_OP_GELU) {
ggml_hexagon_unary(node, flags);
}
break;
case GGML_OP_GLU:
@@ -3254,7 +3267,6 @@ static bool ggml_backend_hexagon_device_supports_op(ggml_backend_dev_t dev, cons
auto sess = static_cast<ggml_hexagon_session *>(dev->context);
bool supp = false;
switch (op->op) {
case GGML_OP_NONE:
case GGML_OP_RESHAPE:
@@ -3294,6 +3306,9 @@ static bool ggml_backend_hexagon_device_supports_op(ggml_backend_dev_t dev, cons
if (ggml_get_unary_op(op) == GGML_UNARY_OP_SILU) {
supp = ggml_hexagon_supported_activations(sess, op);
}
else if (ggml_get_unary_op(op) == GGML_UNARY_OP_GELU){
supp = ggml_hexagon_supported_activations(sess, op);
}
break;
case GGML_OP_GLU: