Merge branch 'layla-build' of https://github.com/l3utterfly/llama.cpp into layla-build

Author: l3utterfly

.gitignore

@@ -161,3 +161,5 @@ prebuilts/QNN_SDK/v2.36.0.250627.zip
 prebuilts/Hexagon_SDK/minimal-hexagon-sdk-6.2.0.1.xz
 prebuilts/OpenCL_SDK/
 prebuilts/Vulkan_SDK/
+
+pkg-adb/

src/llama-context.cpp

@@ -12,10 +12,6 @@
 #include <limits>
 #include <stdexcept>
 
-#ifdef GGML_USE_HEXAGON
-#include "ggml-hexagon.h"
-#endif
-
 //
 // llama_context
 //
@@ -142,15 +138,7 @@ llama_context::llama_context(
     if (!hparams.vocab_only) {
         // GPU backends
         for (auto * dev : model.devices) {
-#ifdef GGML_USE_HEXAGON
-            if (model.params.main_gpu == HEXAGON_BACKEND_GGML)
-                break;
-#endif
-#ifndef GGML_USE_HEXAGON
             ggml_backend_t backend = ggml_backend_dev_init(dev, nullptr);
-#else
-            ggml_backend_t backend = ggml_backend_dev_init(dev,reinterpret_cast<const char *>(model.params.main_gpu));
-#endif
             if (backend == nullptr) {
                 throw std::runtime_error(format("failed to initialize %s backend", ggml_backend_dev_name(dev)));
             }
@@ -159,18 +147,9 @@ llama_context::llama_context(
 
         // add ACCEL backends (such as BLAS)
         for (size_t i = 0; i < ggml_backend_dev_count(); ++i) {
-#ifdef GGML_USE_HEXAGON
-            if (model.params.main_gpu == HEXAGON_BACKEND_GGML)
-                break;
-#endif
             ggml_backend_dev_t dev = ggml_backend_dev_get(i);
-
             if (ggml_backend_dev_type(dev) == GGML_BACKEND_DEVICE_TYPE_ACCEL) {
-#ifndef GGML_USE_HEXAGON
                 ggml_backend_t backend = ggml_backend_dev_init(dev, nullptr);
-#else
-                ggml_backend_t backend = ggml_backend_dev_init(dev,reinterpret_cast<const char *>(model.params.main_gpu));
-#endif
                 if (backend == nullptr) {
                     throw std::runtime_error(format("failed to initialize %s backend", ggml_backend_dev_name(dev)));
                 }

src/llama.cpp

@@ -9,9 +9,6 @@
 
 #include "ggml.h"
 #include "ggml-backend.h"
-#ifdef GGML_USE_HEXAGON
-#include "ggml-hexagon.h"
-#endif
 
 #include <algorithm>
 #include <cstddef>
@@ -207,12 +204,6 @@ static struct llama_model * llama_model_load_from_file_impl(
                     break;
 
                 case GGML_BACKEND_DEVICE_TYPE_GPU: {
-#if GGML_USE_HEXAGON
-                    if (params.main_gpu == HEXAGON_BACKEND_GGML) {
-                        break;
-                    }
-#endif
-                  
                     ggml_backend_reg_t reg = ggml_backend_dev_backend_reg(dev);
                     if (ggml_backend_reg_name(reg) == std::string("RPC")) {
                         rpc_servers.push_back(dev);

tests/CMakeLists.txt

@@ -203,10 +203,6 @@ endif()
 llama_build_and_test(test-gguf.cpp)
 llama_build_and_test(test-backend-ops.cpp)
 
-#dedicated for ggml-hexagon
-llama_build_and_test(ggmlhexagon-benchmark.cpp)
-llama_build_and_test(ggmlhexagon-testops.cpp)
-
 llama_build_and_test(test-model-load-cancel.cpp  LABEL "model")
 llama_build_and_test(test-autorelease.cpp        LABEL "model")