Skip to content

GitLab

Unverified Commit 4987f13d authored by Gabe Goodhart's avatar Gabe Goodhart Committed by GitHub
Browse files

Llama cpp bump (df1b612): granite docling / mamba2 optimizations / multimodal... 


Llama cpp bump (df1b612): granite docling / mamba2 optimizations / multimodal encoding fixes (#12552)

* feat: Bump llama.cpp to df1b612

Branch: LlamaCPPBump-GraniteDocling
Signed-off-by: default avatarGabe Goodhart <ghart@us.ibm.com>

* fix(mtmd): Correctly encode text chunks during mtmd tokenization

There can be text chunks that appear interspersed with the image embeddings
that contain template delimiter tokens for some models. These need to be
correctly translated to text tokens.

Branch: LlamaCPPBump-GraniteDocling
Signed-off-by: default avatarGabe Goodhart <ghart@us.ibm.com>

* tests: Use MtmdChunk in image_test

Branch: LlamaCPPBump-GraniteDocling
Signed-off-by: default avatarGabe Goodhart <ghart@us.ibm.com>

* style: Fix unnecessary conversion linting

Branch: LlamaCPPBump-GraniteDocling
Signed-off-by: default avatarGabe Goodhart <ghart@us.ibm.com>

* fix(ggml): Revert changes to ggml_hip.cpp

These changes were done largely by our code assistant and are likely wrong

Branch: LlamaCPPBump-GraniteDocling
Signed-off-by: default avatarGabe Goodhart <ghart@us.ibm.com>

* fix: Revert changes in mem_nvml.cpp

Branch: LlamaCPPBump-GraniteDocling
Signed-off-by: default avatarGabe Goodhart <ghart@us.ibm.com>

* feat: Update sync point to 1deee0

This brings in several more optimization commits and model support for
EmbeddingGemma

Branch: LlamaCPPBump-GraniteDocling
Signed-off-by: default avatarGabe Goodhart <ghart@us.ibm.com>

* feat: Update patches for 1deee0

Branch: LlamaCPPBump-GraniteDocling
Signed-off-by: default avatarGabe Goodhart <ghart@us.ibm.com>

* feat: sync for bump to 1deee0

Branch: LlamaCPPBump-GraniteDocling
Signed-off-by: default avatarGabe Goodhart <ghart@us.ibm.com>

* fix: Bad patch updates with errant `+`

Branch: LlamaCPPBump-GraniteDocling
Signed-off-by: default avatarGabe Goodhart <ghart@us.ibm.com>

* feat: Bump llama.cpp/ggml to 7049736

Branch: LlamaCPPBump-GraniteDocling
Signed-off-by: default avatarGabe Goodhart <ghart@us.ibm.com>

* fix: format-patches after latest bump

Branch: LlamaCPPBump-GraniteDocling
Signed-off-by: default avatarGabe Goodhart <ghart@us.ibm.com>

---------
Signed-off-by: default avatarGabe Goodhart <ghart@us.ibm.com>
parent e638f2ac
Hide whitespace changes
Inline Side-by-side
Showing with 303 additions and 111 deletions
llama/patches/0016-add-C-API-for-mtmd_input_text.patch
View file @ 4987f13d
......@@ -10,11 +10,11 @@ Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
2 files changed, 13 insertions(+)
diff --git a/tools/mtmd/mtmd.cpp b/tools/mtmd/mtmd.cpp
index cd022c5e..3d680945 100644
index 4d487581..35a0d25e 100644
--- a/tools/mtmd/mtmd.cpp
+++ b/tools/mtmd/mtmd.cpp
@@ -79,6 +79,16 @@ enum mtmd_slice_tmpl {
// TODO @ngxson : add support for idefics (SmolVLM)
MTMD_SLICE_TMPL_IDEFICS3,
};
+mtmd_input_text* mtmd_input_text_init(const char * text, bool add_special, bool parse_special) {
......
llama/patches/0017-no-power-throttling-win32-with-gnuc.patch
View file @ 4987f13d
......@@ -8,10 +8,10 @@ Subject: [PATCH] no power throttling win32 with gnuc
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/ggml/src/ggml-cpu/ggml-cpu.c b/ggml/src/ggml-cpu/ggml-cpu.c
index f8574d01..530efce0 100644
index 99509b0c..b13a491d 100644
--- a/ggml/src/ggml-cpu/ggml-cpu.c
+++ b/ggml/src/ggml-cpu/ggml-cpu.c
@@ -2431,7 +2431,7 @@ static bool ggml_thread_apply_priority(int32_t prio) {
@@ -2437,7 +2437,7 @@ static bool ggml_thread_apply_priority(int32_t prio) {
// Newer Windows 11 versions aggresively park (offline) CPU cores and often place
// all our threads onto the first 4 cores which results in terrible performance with
// n_threads > 4
......
llama/patches/0019-Enable-CUDA-Graphs-for-gemma3n.patch
View file @ 4987f13d
......@@ -13,10 +13,10 @@ checks.
1 file changed, 18 insertions(+)
diff --git a/ggml/src/ggml-cuda/ggml-cuda.cu b/ggml/src/ggml-cuda/ggml-cuda.cu
index ad389ece..e51c5035 100644
index 5b852f69..827e3205 100644
--- a/ggml/src/ggml-cuda/ggml-cuda.cu
+++ b/ggml/src/ggml-cuda/ggml-cuda.cu
@@ -2686,14 +2686,26 @@ static bool check_node_graph_compatibility_and_refresh_copy_ops(ggml_backend_cud
@@ -2689,14 +2689,26 @@ static bool check_node_graph_compatibility_and_refresh_copy_ops(ggml_backend_cud
// Loop over nodes in GGML graph to obtain info needed for CUDA graph
cuda_ctx->cuda_graph->cpy_dest_ptrs.clear();
......@@ -43,7 +43,7 @@ index ad389ece..e51c5035 100644
for (int i = 0; i < cgraph->n_nodes; i++) {
ggml_tensor * node = cgraph->nodes[i];
@@ -2717,6 +2729,12 @@ static bool check_node_graph_compatibility_and_refresh_copy_ops(ggml_backend_cud
@@ -2720,6 +2732,12 @@ static bool check_node_graph_compatibility_and_refresh_copy_ops(ggml_backend_cud
if (node->op == GGML_OP_ADD &&
node->src[1] && node->src[1]->ne[1] > 1 &&
......
llama/patches/0022-ggml-No-alloc-mode.patch
View file @ 4987f13d
......@@ -16,10 +16,10 @@ must be recreated with no-alloc set to false before loading data.
5 files changed, 310 insertions(+), 44 deletions(-)
diff --git a/ggml/include/ggml-backend.h b/ggml/include/ggml-backend.h
index 48777212..d4352663 100644
index 229bf387..1ff53ed0 100644
--- a/ggml/include/ggml-backend.h
+++ b/ggml/include/ggml-backend.h
@@ -303,6 +303,7 @@ extern "C" {
@@ -305,6 +305,7 @@ extern "C" {
// Initialize a backend scheduler, backends with low index are given priority over backends with high index
GGML_API ggml_backend_sched_t ggml_backend_sched_new(ggml_backend_t * backends, ggml_backend_buffer_type_t * bufts, int n_backends, size_t graph_size, bool parallel, bool op_offload);
......@@ -28,7 +28,7 @@ index 48777212..d4352663 100644
// Initialize backend buffers from a measure graph
diff --git a/ggml/src/ggml-backend-impl.h b/ggml/src/ggml-backend-impl.h
index 07784d6f..869dc07d 100644
index 6792ba98..3c3f22fc 100644
--- a/ggml/src/ggml-backend-impl.h
+++ b/ggml/src/ggml-backend-impl.h
@@ -26,12 +26,17 @@ extern "C" {
......@@ -218,7 +218,7 @@ index cb2b9956..6ef5eeaf 100644
void ggml_backend_sched_set_tensor_backend(ggml_backend_sched_t sched, struct ggml_tensor * node, ggml_backend_t backend) {
diff --git a/ggml/src/ggml-cuda/common.cuh b/ggml/src/ggml-cuda/common.cuh
index c4246b65..448badf0 100644
index e0abde54..28d6bcd7 100644
--- a/ggml/src/ggml-cuda/common.cuh
+++ b/ggml/src/ggml-cuda/common.cuh
@@ -35,6 +35,31 @@
......@@ -253,7 +253,7 @@ index c4246b65..448badf0 100644
#define STRINGIZE_IMPL(...) #__VA_ARGS__
#define STRINGIZE(...) STRINGIZE_IMPL(__VA_ARGS__)
@@ -880,6 +905,9 @@ struct ggml_cuda_pool {
@@ -856,6 +881,9 @@ struct ggml_cuda_pool {
virtual void * alloc(size_t size, size_t * actual_size) = 0;
virtual void free(void * ptr, size_t size) = 0;
......@@ -263,7 +263,7 @@ index c4246b65..448badf0 100644
};
template<typename T>
@@ -1023,11 +1051,11 @@ struct ggml_backend_cuda_context {
@@ -999,11 +1027,11 @@ struct ggml_backend_cuda_context {
// pool
std::unique_ptr<ggml_cuda_pool> pools[GGML_CUDA_MAX_DEVICES];
......@@ -277,7 +277,7 @@ index c4246b65..448badf0 100644
}
return *pools[device];
}
@@ -1035,4 +1063,20 @@ struct ggml_backend_cuda_context {
@@ -1011,4 +1039,20 @@ struct ggml_backend_cuda_context {
ggml_cuda_pool & pool() {
return pool(device);
}
......@@ -299,7 +299,7 @@ index c4246b65..448badf0 100644
+ }
};
diff --git a/ggml/src/ggml-cuda/ggml-cuda.cu b/ggml/src/ggml-cuda/ggml-cuda.cu
index e51c5035..d324bc68 100644
index 827e3205..811462c7 100644
--- a/ggml/src/ggml-cuda/ggml-cuda.cu
+++ b/ggml/src/ggml-cuda/ggml-cuda.cu
@@ -350,6 +350,8 @@ const ggml_cuda_device_info & ggml_cuda_info() {
......@@ -540,7 +540,7 @@ index e51c5035..d324bc68 100644
};
ggml_backend_buffer_type_t ggml_backend_cuda_buffer_type(int device) {
@@ -3008,6 +3070,7 @@ static bool ggml_cuda_can_fuse(const struct ggml_cgraph * cgraph, int node_idx,
@@ -3011,6 +3073,7 @@ static bool ggml_cuda_can_fuse(const struct ggml_cgraph * cgraph, int node_idx,
static void evaluate_and_capture_cuda_graph(ggml_backend_cuda_context * cuda_ctx, ggml_cgraph * cgraph,
bool & graph_evaluated_or_captured, bool & use_cuda_graph, bool & cuda_graph_update_required) {
......@@ -548,7 +548,7 @@ index e51c5035..d324bc68 100644
// flag used to determine whether it is an integrated_gpu
const bool integrated = ggml_cuda_info().devices[cuda_ctx->device].integrated;
@@ -3023,6 +3086,11 @@ static void evaluate_and_capture_cuda_graph(ggml_backend_cuda_context * cuda_ctx
@@ -3026,6 +3089,11 @@ static void evaluate_and_capture_cuda_graph(ggml_backend_cuda_context * cuda_ctx
continue;
}
......@@ -560,7 +560,7 @@ index e51c5035..d324bc68 100644
static bool disable_fusion = (getenv("GGML_CUDA_DISABLE_FUSION") != nullptr);
if (!disable_fusion) {
@@ -3149,6 +3217,7 @@ static void evaluate_and_capture_cuda_graph(ggml_backend_cuda_context * cuda_ctx
@@ -3152,6 +3220,7 @@ static void evaluate_and_capture_cuda_graph(ggml_backend_cuda_context * cuda_ctx
static enum ggml_status ggml_backend_cuda_graph_compute(ggml_backend_t backend, ggml_cgraph * cgraph) {
ggml_backend_cuda_context * cuda_ctx = (ggml_backend_cuda_context *)backend->context;
......@@ -568,7 +568,7 @@ index e51c5035..d324bc68 100644
ggml_cuda_set_device(cuda_ctx->device);
@@ -3228,6 +3297,71 @@ static enum ggml_status ggml_backend_cuda_graph_compute(ggml_backend_t backend,
@@ -3231,6 +3300,71 @@ static enum ggml_status ggml_backend_cuda_graph_compute(ggml_backend_t backend,
return GGML_STATUS_SUCCESS;
}
......@@ -640,7 +640,7 @@ index e51c5035..d324bc68 100644
static void ggml_backend_cuda_event_record(ggml_backend_t backend, ggml_backend_event_t event) {
ggml_backend_cuda_context * cuda_ctx = (ggml_backend_cuda_context *)backend->context;
@@ -3268,6 +3402,9 @@ static const ggml_backend_i ggml_backend_cuda_interface = {
@@ -3271,6 +3405,9 @@ static const ggml_backend_i ggml_backend_cuda_interface = {
/* .event_record = */ ggml_backend_cuda_event_record,
/* .event_wait = */ ggml_backend_cuda_event_wait,
/* .graph_optimize = */ NULL,
......
llama/patches/0023-decode-disable-output_all.patch
View file @ 4987f13d
......@@ -8,7 +8,7 @@ Subject: [PATCH] decode: disable output_all
1 file changed, 1 insertion(+), 2 deletions(-)
diff --git a/src/llama-context.cpp b/src/llama-context.cpp
index d8a8b5e6..09247cef 100644
index e7526e7d..53a5e3a9 100644
--- a/src/llama-context.cpp
+++ b/src/llama-context.cpp
@@ -974,8 +974,7 @@ int llama_context::decode(const llama_batch & batch_inp) {
......
llama/patches/0024-ggml-Enable-resetting-backend-devices.patch
View file @ 4987f13d
......@@ -10,12 +10,12 @@ unused then it can be reset to free these data structures.
ggml/include/ggml-backend.h | 1 +
ggml/src/ggml-backend-impl.h | 4 ++++
ggml/src/ggml-backend.cpp | 8 ++++++++
ggml/src/ggml-cuda/ggml-cuda.cu | 17 +++++++++++++++--
ggml/src/ggml-cuda/ggml-cuda.cu | 16 +++++++++++++++-
ggml/src/ggml-cuda/vendors/hip.h | 1 +
5 files changed, 29 insertions(+), 2 deletions(-)
5 files changed, 29 insertions(+), 1 deletion(-)
diff --git a/ggml/include/ggml-backend.h b/ggml/include/ggml-backend.h
index d4352663..0a2dae26 100644
index 1ff53ed0..ba181d09 100644
--- a/ggml/include/ggml-backend.h
+++ b/ggml/include/ggml-backend.h
@@ -178,6 +178,7 @@ extern "C" {
......@@ -27,7 +27,7 @@ index d4352663..0a2dae26 100644
GGML_API ggml_backend_buffer_type_t ggml_backend_dev_host_buffer_type(ggml_backend_dev_t device);
GGML_API ggml_backend_buffer_t ggml_backend_dev_buffer_from_host_ptr(ggml_backend_dev_t device, void * ptr, size_t size, size_t max_tensor_size);
diff --git a/ggml/src/ggml-backend-impl.h b/ggml/src/ggml-backend-impl.h
index 869dc07d..4889df79 100644
index 3c3f22fc..43c91d9f 100644
--- a/ggml/src/ggml-backend-impl.h
+++ b/ggml/src/ggml-backend-impl.h
@@ -195,6 +195,10 @@ extern "C" {
......@@ -61,7 +61,7 @@ index 6ef5eeaf..0b757af5 100644
GGML_ASSERT(device);
return device->iface.get_buffer_type(device);
diff --git a/ggml/src/ggml-cuda/ggml-cuda.cu b/ggml/src/ggml-cuda/ggml-cuda.cu
index d324bc68..531d6e27 100644
index 811462c7..87c6c34a 100644
--- a/ggml/src/ggml-cuda/ggml-cuda.cu
+++ b/ggml/src/ggml-cuda/ggml-cuda.cu
@@ -107,6 +107,11 @@ int ggml_cuda_get_device() {
......@@ -76,7 +76,7 @@ index d324bc68..531d6e27 100644
static cudaError_t ggml_cuda_device_malloc(void ** ptr, size_t size, int device) {
ggml_cuda_set_device(device);
cudaError_t err;
@@ -3512,7 +3517,10 @@ static void ggml_backend_cuda_device_get_props(ggml_backend_dev_t dev, ggml_back
@@ -3515,7 +3520,10 @@ static void ggml_backend_cuda_device_get_props(ggml_backend_dev_t dev, ggml_back
props->id = ggml_backend_cuda_device_get_id(dev);
props->type = ggml_backend_cuda_device_get_type(dev);
props->device_id = ctx->pci_bus_id.empty() ? nullptr : ctx->pci_bus_id.c_str();
......@@ -88,7 +88,7 @@ index d324bc68..531d6e27 100644
bool host_buffer = getenv("GGML_CUDA_NO_PINNED") == nullptr;
#ifdef GGML_CUDA_NO_PEER_COPY
@@ -3945,6 +3953,11 @@ static void ggml_backend_cuda_device_event_synchronize(ggml_backend_dev_t dev, g
@@ -3948,6 +3956,11 @@ static void ggml_backend_cuda_device_event_synchronize(ggml_backend_dev_t dev, g
CUDA_CHECK(cudaEventSynchronize((cudaEvent_t)event->context));
}
......@@ -100,7 +100,7 @@ index d324bc68..531d6e27 100644
static const ggml_backend_device_i ggml_backend_cuda_device_interface = {
/* .get_name = */ ggml_backend_cuda_device_get_name,
/* .get_description = */ ggml_backend_cuda_device_get_description,
@@ -3961,6 +3974,7 @@ static const ggml_backend_device_i ggml_backend_cuda_device_interface = {
@@ -3964,6 +3977,7 @@ static const ggml_backend_device_i ggml_backend_cuda_device_interface = {
/* .event_new = */ ggml_backend_cuda_device_event_new,
/* .event_free = */ ggml_backend_cuda_device_event_free,
/* .event_synchronize = */ ggml_backend_cuda_device_event_synchronize,
......@@ -108,19 +108,11 @@ index d324bc68..531d6e27 100644
};
// backend reg
@@ -4076,7 +4090,6 @@ ggml_backend_reg_t ggml_backend_cuda_reg() {
dev_ctx->device = i;
dev_ctx->name = GGML_CUDA_NAME + std::to_string(i);
- ggml_cuda_set_device(i);
cudaDeviceProp prop;
CUDA_CHECK(cudaGetDeviceProperties(&prop, i));
dev_ctx->description = prop.name;
diff --git a/ggml/src/ggml-cuda/vendors/hip.h b/ggml/src/ggml-cuda/vendors/hip.h
index 37386afc..06f9e7c1 100644
index 890c1036..1f06be80 100644
--- a/ggml/src/ggml-cuda/vendors/hip.h
+++ b/ggml/src/ggml-cuda/vendors/hip.h
@@ -41,6 +41,7 @@
@@ -45,6 +45,7 @@
#define cudaDeviceDisablePeerAccess hipDeviceDisablePeerAccess
#define cudaDeviceEnablePeerAccess hipDeviceEnablePeerAccess
#define cudaDeviceProp hipDeviceProp_t
......
llama/patches/0026-GPU-discovery-enhancements.patch
View file @ 4987f13d
......@@ -9,17 +9,17 @@ management libraries for more accurate VRAM usage reporting if available.
ggml/include/ggml-backend.h | 9 +
ggml/src/CMakeLists.txt | 2 +
ggml/src/ggml-cuda/ggml-cuda.cu | 72 +++++
ggml/src/ggml-cuda/vendors/hip.h | 4 +
ggml/src/ggml-cuda/vendors/hip.h | 3 +
ggml/src/ggml-impl.h | 8 +
ggml/src/ggml-metal/ggml-metal.cpp | 3 +-
ggml/src/ggml-metal/ggml-metal.cpp | 2 +
ggml/src/mem_hip.cpp | 449 +++++++++++++++++++++++++++++
ggml/src/mem_nvml.cpp | 209 ++++++++++++++
8 files changed, 755 insertions(+), 1 deletion(-)
8 files changed, 754 insertions(+)
create mode 100644 ggml/src/mem_hip.cpp
create mode 100644 ggml/src/mem_nvml.cpp
diff --git a/ggml/include/ggml-backend.h b/ggml/include/ggml-backend.h
index 0a2dae26a..a6bf33785 100644
index ba181d09..09ff75f9 100644
--- a/ggml/include/ggml-backend.h
+++ b/ggml/include/ggml-backend.h
@@ -169,6 +169,15 @@ extern "C" {
......@@ -39,10 +39,10 @@ index 0a2dae26a..a6bf33785 100644
GGML_API const char * ggml_backend_dev_name(ggml_backend_dev_t device);
diff --git a/ggml/src/CMakeLists.txt b/ggml/src/CMakeLists.txt
index 33b3a15f0..86191ef2c 100644
index 0609c650..aefe43bd 100644
--- a/ggml/src/CMakeLists.txt
+++ b/ggml/src/CMakeLists.txt
@@ -206,6 +206,8 @@ add_library(ggml-base
@@ -209,6 +209,8 @@ add_library(ggml-base
ggml-threading.h
ggml-quants.c
ggml-quants.h
......@@ -52,7 +52,7 @@ index 33b3a15f0..86191ef2c 100644
target_include_directories(ggml-base PRIVATE .)
diff --git a/ggml/src/ggml-cuda/ggml-cuda.cu b/ggml/src/ggml-cuda/ggml-cuda.cu
index 531d6e272..3fa3a0575 100644
index 87c6c34a..6a278b5e 100644
--- a/ggml/src/ggml-cuda/ggml-cuda.cu
+++ b/ggml/src/ggml-cuda/ggml-cuda.cu
@@ -261,6 +261,16 @@ static ggml_cuda_device_info ggml_cuda_init() {
......@@ -84,7 +84,7 @@ index 531d6e272..3fa3a0575 100644
GGML_LOG_INFO(" Device %d: %s, compute capability %d.%d, VMM: %s, ID: %s\n",
id, prop.name, prop.major, prop.minor, device_vmm ? "yes" : "no",
ggml_cuda_parse_uuid(prop, id).c_str());
@@ -3481,6 +3496,14 @@ struct ggml_backend_cuda_device_context {
@@ -3484,6 +3499,14 @@ struct ggml_backend_cuda_device_context {
std::string description;
std::string pci_bus_id;
std::string id;
......@@ -99,7 +99,7 @@ index 531d6e272..3fa3a0575 100644
};
static const char * ggml_backend_cuda_device_get_name(ggml_backend_dev_t dev) {
@@ -3501,6 +3524,28 @@ static const char * ggml_backend_cuda_device_get_id(ggml_backend_dev_t dev) {
@@ -3504,6 +3527,28 @@ static const char * ggml_backend_cuda_device_get_id(ggml_backend_dev_t dev) {
static void ggml_backend_cuda_device_get_memory(ggml_backend_dev_t dev, size_t * free, size_t * total) {
ggml_backend_cuda_device_context * ctx = (ggml_backend_cuda_device_context *)dev->context;
ggml_cuda_set_device(ctx->device);
......@@ -128,7 +128,7 @@ index 531d6e272..3fa3a0575 100644
CUDA_CHECK(cudaMemGetInfo(free, total));
}
@@ -3509,6 +3554,7 @@ static enum ggml_backend_dev_type ggml_backend_cuda_device_get_type(ggml_backend
@@ -3512,6 +3557,7 @@ static enum ggml_backend_dev_type ggml_backend_cuda_device_get_type(ggml_backend
return GGML_BACKEND_DEVICE_TYPE_GPU;
}
......@@ -136,7 +136,7 @@ index 531d6e272..3fa3a0575 100644
static void ggml_backend_cuda_device_get_props(ggml_backend_dev_t dev, ggml_backend_dev_props * props) {
ggml_backend_cuda_device_context * ctx = (ggml_backend_cuda_device_context *)dev->context;
@@ -3522,6 +3568,22 @@ static void ggml_backend_cuda_device_get_props(ggml_backend_dev_t dev, ggml_back
@@ -3525,6 +3571,22 @@ static void ggml_backend_cuda_device_get_props(ggml_backend_dev_t dev, ggml_back
// If you need the memory data, call ggml_backend_dev_memory() explicitly.
props->memory_total = props->memory_free = 0;
......@@ -159,7 +159,7 @@ index 531d6e272..3fa3a0575 100644
bool host_buffer = getenv("GGML_CUDA_NO_PINNED") == nullptr;
#ifdef GGML_CUDA_NO_PEER_COPY
bool events = false;
@@ -4084,6 +4146,8 @@ ggml_backend_reg_t ggml_backend_cuda_reg() {
@@ -4087,6 +4149,8 @@ ggml_backend_reg_t ggml_backend_cuda_reg() {
std::lock_guard<std::mutex> lock(mutex);
if (!initialized) {
ggml_backend_cuda_reg_context * ctx = new ggml_backend_cuda_reg_context;
......@@ -168,7 +168,7 @@ index 531d6e272..3fa3a0575 100644
for (int i = 0; i < ggml_cuda_info().device_count; i++) {
ggml_backend_cuda_device_context * dev_ctx = new ggml_backend_cuda_device_context;
@@ -4099,6 +4163,14 @@ ggml_backend_reg_t ggml_backend_cuda_reg() {
@@ -4102,6 +4166,14 @@ ggml_backend_reg_t ggml_backend_cuda_reg() {
snprintf(pci_bus_id, sizeof(pci_bus_id), "%04x:%02x:%02x.0", prop.pciDomainID, prop.pciBusID, prop.pciDeviceID);
dev_ctx->pci_bus_id = pci_bus_id;
......@@ -184,20 +184,19 @@ index 531d6e272..3fa3a0575 100644
/* .iface = */ ggml_backend_cuda_device_interface,
/* .reg = */ &reg,
diff --git a/ggml/src/ggml-cuda/vendors/hip.h b/ggml/src/ggml-cuda/vendors/hip.h
index 06f9e7c1e..eb8f66cb0 100644
index 1f06be80..2f9ef2dc 100644
--- a/ggml/src/ggml-cuda/vendors/hip.h
+++ b/ggml/src/ggml-cuda/vendors/hip.h
@@ -5,6 +5,9 @@
@@ -5,6 +5,8 @@
#include <hipblas/hipblas.h>
#include <hip/hip_fp16.h>
#include <hip/hip_bf16.h>
+// for rocblas_initialize()
+#include "rocblas/rocblas.h"
+
#define CUBLAS_GEMM_DEFAULT HIPBLAS_GEMM_DEFAULT
#define CUBLAS_GEMM_DEFAULT_TENSOR_OP HIPBLAS_GEMM_DEFAULT
@@ -43,6 +46,7 @@
#if defined(GGML_HIP_ROCWMMA_FATTN)
#include <rocwmma/rocwmma-version.hpp>
@@ -47,6 +49,7 @@
#define cudaDeviceProp hipDeviceProp_t
#define cudaDeviceReset hipDeviceReset
#define cudaDeviceSynchronize hipDeviceSynchronize
......@@ -206,10 +205,10 @@ index 06f9e7c1e..eb8f66cb0 100644
#define cudaErrorPeerAccessAlreadyEnabled hipErrorPeerAccessAlreadyEnabled
#define cudaErrorPeerAccessNotEnabled hipErrorPeerAccessNotEnabled
diff --git a/ggml/src/ggml-impl.h b/ggml/src/ggml-impl.h
index 86a1ebf62..9fc9fbfcf 100644
index d0fb3bcc..80597b6e 100644
--- a/ggml/src/ggml-impl.h
+++ b/ggml/src/ggml-impl.h
@@ -635,6 +635,14 @@ static inline bool ggml_can_fuse(const struct ggml_cgraph * cgraph, int node_idx
@@ -638,6 +638,14 @@ static inline bool ggml_can_fuse(const struct ggml_cgraph * cgraph, int node_idx
return ggml_can_fuse_ext(cgraph, idxs, ops, num_ops);
}
......@@ -225,7 +224,7 @@ index 86a1ebf62..9fc9fbfcf 100644
}
#endif
diff --git a/ggml/src/ggml-metal/ggml-metal.cpp b/ggml/src/ggml-metal/ggml-metal.cpp
index 08ab4fc91..17999a616 100644
index f2ff9f32..f356e4a0 100644
--- a/ggml/src/ggml-metal/ggml-metal.cpp
+++ b/ggml/src/ggml-metal/ggml-metal.cpp
@@ -535,6 +535,7 @@ static enum ggml_backend_dev_type ggml_backend_metal_device_get_type(ggml_backen
......@@ -236,18 +235,17 @@ index 08ab4fc91..17999a616 100644
static void ggml_backend_metal_device_get_props(ggml_backend_dev_t dev, ggml_backend_dev_props * props) {
props->name = ggml_backend_metal_device_get_name(dev);
props->description = ggml_backend_metal_device_get_description(dev);
@@ -542,7 +543,7 @@ static void ggml_backend_metal_device_get_props(ggml_backend_dev_t dev, ggml_bac
props->type = ggml_backend_metal_device_get_type(dev);
@@ -543,6 +544,7 @@ static void ggml_backend_metal_device_get_props(ggml_backend_dev_t dev, ggml_bac
ggml_backend_metal_device_get_memory(dev, &props->memory_free, &props->memory_total);
-
+ props->library = GGML_METAL_NAME;
props->caps = {
/* .async = */ true,
/* .host_buffer = */ false,
diff --git a/ggml/src/mem_hip.cpp b/ggml/src/mem_hip.cpp
new file mode 100644
index 000000000..8ef19b8cf
index 00000000..8ef19b8c
--- /dev/null
+++ b/ggml/src/mem_hip.cpp
@@ -0,0 +1,449 @@
......@@ -703,7 +701,7 @@ index 000000000..8ef19b8cf
\ No newline at end of file
diff --git a/ggml/src/mem_nvml.cpp b/ggml/src/mem_nvml.cpp
new file mode 100644
index 000000000..c9073cef0
index 00000000..c9073cef
--- /dev/null
+++ b/ggml/src/mem_nvml.cpp
@@ -0,0 +1,209 @@
......
ml/backend/ggml/ggml/include/ggml-backend.h
View file @ 4987f13d
......@@ -226,6 +226,8 @@ extern "C" {
// Backend registry
//
GGML_API void ggml_backend_register(ggml_backend_reg_t reg);
GGML_API void ggml_backend_device_register(ggml_backend_dev_t device);
// Backend (reg) enumeration
......
ml/backend/ggml/ggml/include/ggml-rpc.h
View file @ 4987f13d
......@@ -7,26 +7,25 @@
extern "C" {
#endif
#define RPC_PROTO_MAJOR_VERSION 2
#define RPC_PROTO_MAJOR_VERSION 3
#define RPC_PROTO_MINOR_VERSION 0
#define RPC_PROTO_PATCH_VERSION 0
#define GGML_RPC_MAX_SERVERS 16
// backend API
GGML_BACKEND_API ggml_backend_t ggml_backend_rpc_init(const char * endpoint);
GGML_BACKEND_API ggml_backend_t ggml_backend_rpc_init(const char * endpoint, uint32_t device);
GGML_BACKEND_API bool ggml_backend_is_rpc(ggml_backend_t backend);
GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_rpc_buffer_type(const char * endpoint);
GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_rpc_buffer_type(const char * endpoint, uint32_t device);
GGML_BACKEND_API void ggml_backend_rpc_get_device_memory(const char * endpoint, size_t * free, size_t * total);
GGML_BACKEND_API void ggml_backend_rpc_get_device_memory(const char * endpoint, uint32_t device, size_t * free, size_t * total);
GGML_BACKEND_API void ggml_backend_rpc_start_server(ggml_backend_t backend, const char * endpoint,
const char * cache_dir,
size_t free_mem, size_t total_mem);
GGML_BACKEND_API void ggml_backend_rpc_start_server(const char * endpoint, const char * cache_dir,
size_t n_threads, size_t n_devices,
ggml_backend_dev_t * devices, size_t * free_mem, size_t * total_mem);
GGML_BACKEND_API ggml_backend_reg_t ggml_backend_rpc_reg(void);
GGML_BACKEND_API ggml_backend_dev_t ggml_backend_rpc_add_device(const char * endpoint);
GGML_BACKEND_API ggml_backend_reg_t ggml_backend_rpc_add_server(const char * endpoint);
#ifdef __cplusplus
}
......
ml/backend/ggml/ggml/include/ggml.h
View file @ 4987f13d
......@@ -237,6 +237,8 @@
#define GGML_EXIT_SUCCESS 0
#define GGML_EXIT_ABORTED 1
// TODO: convert to enum https://github.com/ggml-org/llama.cpp/pull/16187#discussion_r2388538726
#define GGML_ROPE_TYPE_NORMAL 0
#define GGML_ROPE_TYPE_NEOX 2
#define GGML_ROPE_TYPE_MROPE 8
#define GGML_ROPE_TYPE_VISION 24
......@@ -574,6 +576,7 @@ extern "C" {
GGML_UNARY_OP_HARDSIGMOID,
GGML_UNARY_OP_EXP,
GGML_UNARY_OP_GELU_ERF,
GGML_UNARY_OP_XIELU,
GGML_UNARY_OP_COUNT,
};
......@@ -1148,6 +1151,18 @@ extern "C" {
struct ggml_context * ctx,
struct ggml_tensor * a);
// xIELU activation function
// x = x * (c_a(alpha_n) + c_b(alpha_p, beta) * sigmoid(beta * x)) + eps * (x > 0)
// where c_a = softplus and c_b(a, b) = softplus(a) + b are constraining functions
// that constrain the positive and negative source alpha values respectively
GGML_API struct ggml_tensor * ggml_xielu(
struct ggml_context * ctx,
struct ggml_tensor * a,
float alpha_n,
float alpha_p,
float beta,
float eps);
// gated linear unit ops
// A: n columns, r rows,
// result is n / 2 columns, r rows,
......@@ -1615,6 +1630,13 @@ extern "C" {
float scale,
float max_bias);
GGML_API struct ggml_tensor * ggml_soft_max_ext_inplace(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * mask,
float scale,
float max_bias);
GGML_API void ggml_soft_max_add_sinks(
struct ggml_tensor * a,
struct ggml_tensor * sinks);
......
ml/backend/ggml/ggml/src/CMakeLists.txt
View file @ 4987f13d
......@@ -145,6 +145,9 @@ endif()
# which was introduced in POSIX.1-2008, forcing us to go higher
if (CMAKE_SYSTEM_NAME MATCHES "OpenBSD")
add_compile_definitions(_XOPEN_SOURCE=700)
elseif (CMAKE_SYSTEM_NAME MATCHES "AIX")
# Don't define _XOPEN_SOURCE. We need _ALL_SOURCE, which is the default,
# in order to define _SC_PHYS_PAGES.
else()
add_compile_definitions(_XOPEN_SOURCE=600)
endif()
......
ml/backend/ggml/ggml/src/ggml-alloc.c
View file @ 4987f13d
......@@ -392,12 +392,8 @@ static void ggml_dyn_tallocr_free(struct ggml_dyn_tallocr * alloc) {
free(alloc);
}
static size_t ggml_dyn_tallocr_max_size(struct ggml_dyn_tallocr * alloc) {
size_t max_size = 0;
for (int i = 0; i < alloc->n_chunks; i++) {
max_size += alloc->chunks[i]->max_size;
}
return max_size;
static size_t ggml_dyn_tallocr_max_size(struct ggml_dyn_tallocr * alloc, int chunk) {
return chunk < alloc->n_chunks ? alloc->chunks[chunk]->max_size : 0;
}
......@@ -417,10 +413,8 @@ static void ggml_vbuffer_free(struct vbuffer * buf) {
free(buf);
}
static int ggml_vbuffer_n_chunks(struct vbuffer * buf) {
int n = 0;
while (n < GGML_VBUFFER_MAX_CHUNKS && buf->chunks[n]) n++;
return n;
static size_t ggml_vbuffer_chunk_size(struct vbuffer * buf, int chunk) {
return buf->chunks[chunk] ? ggml_backend_buffer_get_size(buf->chunks[chunk]) : 0;
}
static size_t ggml_vbuffer_size(struct vbuffer * buf) {
......@@ -892,12 +886,20 @@ bool ggml_gallocr_reserve_n(ggml_gallocr_t galloc, struct ggml_cgraph * graph, c
}
}
size_t cur_size = galloc->buffers[i] ? ggml_vbuffer_size(galloc->buffers[i]) : 0;
size_t new_size = ggml_dyn_tallocr_max_size(galloc->buf_tallocs[i]);
// even if there are no tensors allocated in this buffer, we still need to allocate it to initialize views
if (new_size > cur_size || galloc->buffers[i] == NULL) {
bool realloc = galloc->buffers[i] == NULL;
size_t new_size = 0;
for (int c = 0; c < galloc->buf_tallocs[i]->n_chunks; c++) {
size_t cur_chunk_size = galloc->buffers[i] ? ggml_vbuffer_chunk_size(galloc->buffers[i], c) : 0;
size_t new_chunk_size = ggml_dyn_tallocr_max_size(galloc->buf_tallocs[i], c);
new_size += new_chunk_size;
if (new_chunk_size > cur_chunk_size) {
realloc = true;
}
}
if (realloc) {
#ifndef NDEBUG
size_t cur_size = galloc->buffers[i] ? ggml_vbuffer_size(galloc->buffers[i]) : 0;
GGML_LOG_DEBUG("%s: reallocating %s buffer from size %.02f MiB to %.02f MiB\n", __func__, ggml_backend_buft_name(galloc->bufts[i]), cur_size / 1024.0 / 1024.0, new_size / 1024.0 / 1024.0);
#endif
......
ml/backend/ggml/ggml/src/ggml-backend-impl.h
View file @ 4987f13d
......@@ -229,9 +229,6 @@ extern "C" {
void * context;
};
// Internal backend registry API
GGML_API void ggml_backend_register(ggml_backend_reg_t reg);
// Add backend dynamic loading support to the backend
// Initialize the backend
......
ml/backend/ggml/ggml/src/ggml-cpu/amx/amx.cpp
View file @ 4987f13d
......@@ -149,6 +149,7 @@ class extra_buffer_type : ggml::cpu::extra_buffer_type {
if (op->op == GGML_OP_MUL_MAT && is_contiguous_2d(op->src[0]) && // src0 must be contiguous
is_contiguous_2d(op->src[1]) && // src1 must be contiguous
op->src[0]->buffer && op->src[0]->buffer->buft == ggml_backend_amx_buffer_type() &&
op->src[0]->ne[0] % (TILE_K * 2 * 32) == 0 && // TODO: not sure if correct (https://github.com/ggml-org/llama.cpp/pull/16315)
op->ne[0] % (TILE_N * 2) == 0 && // out_features is 32x
(qtype_has_amx_kernels(op->src[0]->type) || (op->src[0]->type == GGML_TYPE_F16))) {
// src1 must be host buffer
......
ml/backend/ggml/ggml/src/ggml-cpu/ggml-cpu-impl.h
View file @ 4987f13d
......@@ -68,7 +68,7 @@ struct ggml_compute_params {
#endif // __VXE2__
#endif // __s390x__ && __VEC__
#if defined(__ARM_FEATURE_SVE)
#if defined(__ARM_FEATURE_SVE) && defined(__linux__)
#include <sys/prctl.h>
#endif
......
ml/backend/ggml/ggml/src/ggml-cpu/ggml-cpu.c
View file @ 4987f13d
......@@ -691,8 +691,13 @@ bool ggml_is_numa(void) {
#endif
static void ggml_init_arm_arch_features(void) {
#if defined(__linux__) && defined(__aarch64__) && defined(__ARM_FEATURE_SVE)
#if defined(__aarch64__) && defined(__ARM_FEATURE_SVE)
#if defined(__linux__)
ggml_arm_arch_features.sve_cnt = PR_SVE_VL_LEN_MASK & prctl(PR_SVE_GET_VL);
#else
// TODO: add support of SVE for non-linux systems
#error "TODO: SVE is not supported on this platform. To use SVE, sve_cnt needs to be initialized here."
#endif
#endif
}
......@@ -2189,6 +2194,7 @@ static int ggml_get_n_tasks(struct ggml_tensor * node, int n_threads) {
case GGML_UNARY_OP_GELU_ERF:
case GGML_UNARY_OP_GELU_QUICK:
case GGML_UNARY_OP_SILU:
case GGML_UNARY_OP_XIELU:
{
n_tasks = n_threads;
} break;
......
ml/backend/ggml/ggml/src/ggml-cpu/ops.cpp
View file @ 4987f13d
......@@ -3467,31 +3467,27 @@ static void ggml_compute_forward_norm_f32(
GGML_ASSERT(eps >= 0.0f);
// TODO: optimize
for (int64_t i03 = 0; i03 < ne03; i03++) {
for (int64_t i02 = 0; i02 < ne02; i02++) {
for (int64_t i01 = ith; i01 < ne01; i01 += nth) {
const float * x = (float *) ((char *) src0->data + i01*nb01 + i02*nb02 + i03*nb03);
ggml_float sum = 0.0;
for (int64_t i00 = 0; i00 < ne00; i00++) {
sum += (ggml_float)x[i00];
}
float sum = 0.0;
ggml_vec_sum_f32(ne00, &sum, x);
float mean = sum/ne00;
float * y = (float *) ((char *) dst->data + i01*nb1 + i02*nb2 + i03*nb3);
float variance = 0;
ggml_float sum2 = 0.0;
for (int64_t i00 = 0; i00 < ne00; i00++) {
float v = x[i00] - mean;
y[i00] = v;
sum2 += (ggml_float)(v*v);
}
#ifdef GGML_USE_ACCELERATE
mean = -mean;
vDSP_vsadd(x, 1, &mean, y, 1, ne00);
vDSP_measqv(y, 1, &variance, ne00);
#else
variance = ggml_vec_cvar_f32(ne00, y, x, mean);
#endif //GGML_USE_ACCELERATE
float variance = sum2/ne00;
const float scale = 1.0f/sqrtf(variance + eps);
ggml_vec_scale_f32(ne00, y, scale);
}
}
......@@ -8178,7 +8174,7 @@ static void ggml_compute_forward_flash_attn_ext_f16(
}
// V /= S
const float S_inv = 1.0f/S;
const float S_inv = S == 0.0f ? 0.0f : 1.0f/S;
ggml_vec_scale_f32(DV, VKQ32, S_inv);
// dst indices
......@@ -8680,7 +8676,7 @@ static void ggml_compute_forward_ssm_scan_f32(
// n_head
for (int h = ih0; h < ih1; ++h) {
// ref: https://github.com/state-spaces/mamba/blob/62db608da60f6fc790b8ed9f4b3225e95ca15fde/mamba_ssm/ops/triton/softplus.py#L16
const float dt_soft_plus = dt[h] <= 20.0f ? log1pf(expf(dt[h])) : dt[h];
const float dt_soft_plus = ggml_softplus(dt[h]);
const float dA = expf(dt_soft_plus * A[h]);
const int g = h / (nh / ng); // repeat_interleave
......@@ -8777,7 +8773,7 @@ static void ggml_compute_forward_ssm_scan_f32(
// n_head
for (int h = ih0; h < ih1; ++h) {
// ref: https://github.com/state-spaces/mamba/blob/62db608da60f6fc790b8ed9f4b3225e95ca15fde/mamba_ssm/ops/triton/softplus.py#L16
const float dt_soft_plus = dt[h] <= 20.0f ? log1pf(expf(dt[h])) : dt[h];
const float dt_soft_plus = ggml_softplus(dt[h]);
const int g = h / (nh / ng); // repeat_interleave
// dim
......@@ -9040,6 +9036,10 @@ void ggml_compute_forward_unary(
{
ggml_compute_forward_exp(params, dst);
} break;
case GGML_UNARY_OP_XIELU:
{
ggml_compute_forward_xielu(params, dst);
} break;
default:
{
GGML_ABORT("fatal error");
......
ml/backend/ggml/ggml/src/ggml-cpu/unary-ops.cpp
View file @ 4987f13d
......@@ -52,6 +52,15 @@ static inline float op_sqrt(float x) {
return sqrtf(x);
}
static inline float op_xielu(float x, float alpha_n, float alpha_p, float beta, float eps) {
if (x > 0.0f) {
return alpha_p * x * x + beta * x;
} else {
const float min_x_eps = fminf(x, eps);
return (expm1f(min_x_eps) - x) * alpha_n + beta * x;
}
}
static inline float op_sin(float x) {
return sinf(x);
}
......@@ -121,6 +130,86 @@ static void unary_op(const ggml_compute_params * params, ggml_tensor * dst) {
}
}
template <float (*op)(float, ggml_tensor *)>
static void unary_op_params(const ggml_compute_params * params, ggml_tensor * dst) {
const ggml_tensor * src0 = dst->src[0];
/* */ if (src0->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32) { // all f32
apply_unary_op<op, float, float>(params, dst);
} else if (src0->type == GGML_TYPE_F16 && dst->type == GGML_TYPE_F16) { // all f16
apply_unary_op<op, ggml_fp16_t, ggml_fp16_t>(params, dst);
} else if (src0->type == GGML_TYPE_BF16 && dst->type == GGML_TYPE_BF16) { // all bf16
apply_unary_op<op, ggml_bf16_t, ggml_bf16_t>(params, dst);
} else if (src0->type == GGML_TYPE_BF16 && dst->type == GGML_TYPE_F32) {
apply_unary_op<op, ggml_bf16_t, float>(params, dst);
} else if (src0->type == GGML_TYPE_F16 && dst->type == GGML_TYPE_F32) {
apply_unary_op<op, ggml_fp16_t, float>(params, dst);
} else {
fprintf(stderr, "%s: unsupported types: dst: %s, src0: %s\n", __func__,
ggml_type_name(dst->type), ggml_type_name(src0->type));
GGML_ABORT("fatal error");
}
}
// Extend vec_unary_op to support functors
template <typename Op, typename src0_t, typename dst_t>
static inline void vec_unary_op_functor(int64_t n, dst_t * y, const src0_t * x, Op op) {
constexpr auto src0_to_f32 = type_conversion_table<src0_t>::to_f32;
constexpr auto f32_to_dst = type_conversion_table<dst_t >::from_f32;
for (int i = 0; i < n; i++) {
y[i] = f32_to_dst(op(src0_to_f32(x[i])));
}
}
// Extend apply_unary_op to support functors
template <typename Op, typename src0_t, typename dst_t>
static void apply_unary_op_functor(const ggml_compute_params * params, ggml_tensor * dst, Op op) {
const ggml_tensor * src0 = dst->src[0];
GGML_ASSERT(ggml_is_contiguous_1(src0) && ggml_is_contiguous_1(dst) && ggml_are_same_shape(src0, dst));
GGML_TENSOR_UNARY_OP_LOCALS
GGML_ASSERT( nb0 == sizeof(dst_t));
GGML_ASSERT(nb00 == sizeof(src0_t));
const auto [ir0, ir1] = get_thread_range(params, src0);
for (int64_t ir = ir0; ir < ir1; ++ir) {
const int64_t i03 = ir/(ne02*ne01);
const int64_t i02 = (ir - i03*ne02*ne01)/ne01;
const int64_t i01 = (ir - i03*ne02*ne01 - i02*ne01);
dst_t * dst_ptr = (dst_t *) ((char *) dst->data + i03*nb3 + i02*nb2 + i01*nb1 );
const src0_t * src0_ptr = (const src0_t *) ((const char *) src0->data + i03*nb03 + i02*nb02 + i01*nb01);
vec_unary_op_functor(ne0, dst_ptr, src0_ptr, op);
}
}
// Generic dispatcher for functors
template <typename Op>
static void unary_op_functor(const ggml_compute_params * params, ggml_tensor * dst, Op op) {
const ggml_tensor * src0 = dst->src[0];
/* */ if (src0->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32) { // all f32
apply_unary_op_functor<Op, float, float>(params, dst, op);
} else if (src0->type == GGML_TYPE_F16 && dst->type == GGML_TYPE_F16) { // all f16
apply_unary_op_functor<Op, ggml_fp16_t, ggml_fp16_t>(params, dst, op);
} else if (src0->type == GGML_TYPE_BF16 && dst->type == GGML_TYPE_BF16) { // all bf16
apply_unary_op_functor<Op, ggml_bf16_t, ggml_bf16_t>(params, dst, op);
} else if (src0->type == GGML_TYPE_BF16 && dst->type == GGML_TYPE_F32) {
apply_unary_op_functor<Op, ggml_bf16_t, float>(params, dst, op);
} else if (src0->type == GGML_TYPE_F16 && dst->type == GGML_TYPE_F32) {
apply_unary_op_functor<Op, ggml_fp16_t, float>(params, dst, op);
} else {
fprintf(stderr, "%s: unsupported types: dst: %s, src0: %s\n", __func__,
ggml_type_name(dst->type), ggml_type_name(src0->type));
GGML_ABORT("fatal error");
}
}
void ggml_compute_forward_abs(const ggml_compute_params * params, ggml_tensor * dst) {
unary_op<op_abs>(params, dst);
}
......@@ -184,3 +273,17 @@ void ggml_compute_forward_cos(const ggml_compute_params * params, ggml_tensor *
void ggml_compute_forward_log(const ggml_compute_params * params, ggml_tensor * dst) {
unary_op<op_log>(params, dst);
}
void ggml_compute_forward_xielu(const ggml_compute_params * params, ggml_tensor * dst) {
const float alpha_n = ggml_get_op_params_f32(dst, 1);
const float alpha_p = ggml_get_op_params_f32(dst, 2);
const float beta = ggml_get_op_params_f32(dst, 3);
const float eps = ggml_get_op_params_f32(dst, 4);
const auto xielu_op_params = [alpha_n, alpha_p, beta, eps](float f) {
return op_xielu(f, alpha_n, alpha_p, beta, eps);
};
unary_op_functor(params, dst, xielu_op_params);
}
ml/backend/ggml/ggml/src/ggml-cpu/unary-ops.h
View file @ 4987f13d
......@@ -22,6 +22,7 @@ void ggml_compute_forward_sqrt(const struct ggml_compute_params * params, struct
void ggml_compute_forward_sin(const struct ggml_compute_params * params, struct ggml_tensor * dst);
void ggml_compute_forward_cos(const struct ggml_compute_params * params, struct ggml_tensor * dst);
void ggml_compute_forward_log(const struct ggml_compute_params * params, struct ggml_tensor * dst);
void ggml_compute_forward_xielu(const struct ggml_compute_params * params, struct ggml_tensor * dst);
#ifdef __cplusplus
}
......
ml/backend/ggml/ggml/src/ggml-cpu/vec.cpp
View file @ 4987f13d
......@@ -404,6 +404,72 @@ void ggml_vec_swiglu_f32(const int n, float * y, const float * x, const float *
}
}
ggml_float ggml_vec_cvar_f32(const int n, float * y, const float * x, const float mean) {
int i = 0;
ggml_float sum = 0;
// TODO: optimize to process the remaining elements in groups using the smaller vector sizes from AVX2 and SSE
// ref: https://github.com/ggml-org/llama.cpp/pull/15953#pullrequestreview-3310928344
#if defined(__AVX512F__) && defined(__AVX512DQ__)
for (; i + 15 < n; i += 16) {
__m512 val = _mm512_sub_ps(_mm512_loadu_ps(x + i),
_mm512_set1_ps(mean));
_mm512_storeu_ps(y + i, val);
sum += (ggml_float)_mm512_reduce_add_ps(_mm512_mul_ps(val, val));
}
#elif defined(__AVX2__) && defined(__FMA__)
for (; i + 7 < n; i += 8) {
__m256 val = _mm256_sub_ps(_mm256_loadu_ps(x + i),
_mm256_set1_ps(mean));
_mm256_storeu_ps(y + i, val);
val = _mm256_mul_ps(val,val);
__m128 val2 = _mm_add_ps(_mm256_extractf128_ps(val, 1),
_mm256_castps256_ps128(val));
val2 = _mm_add_ps(val2, _mm_movehl_ps(val2, val2));
val2 = _mm_add_ss(val2, _mm_movehdup_ps(val2));
sum += (ggml_float)_mm_cvtss_f32(val2);
}
#elif defined(__SSE2__)
for (; i + 3 < n; i += 4) {
__m128 val = _mm_sub_ps(_mm_loadu_ps(x + i),
_mm_set1_ps(mean));
_mm_storeu_ps(y + i, val);
val = _mm_mul_ps(val, val);
#if defined(__AVX__) || defined(__AVX2__) || defined(__AVX512F__)
val = _mm_add_ps(val, _mm_movehl_ps(val, val));
val = _mm_add_ss(val, _mm_movehdup_ps(val));
#else
__m128 tmp = _mm_shuffle_ps(val, val, _MM_SHUFFLE(2, 3, 0, 1));
val = _mm_add_ps(val, tmp);
tmp = _mm_movehl_ps(tmp, val);
val = _mm_add_ss(val, tmp);
#endif // __AVX__ || __AVX2__ || __AVX512F__
sum += (ggml_float)_mm_cvtss_f32(val);
}
#elif defined(__ARM_NEON) && defined(__aarch64__)
for (; i + 3 < n; i += 4) {
float32x4_t val = vsubq_f32(vld1q_f32(x + i),
vdupq_n_f32(mean));
vst1q_f32(y + i, val);
val = vmulq_f32(val, val);
sum += (ggml_float)vaddvq_f32(val);
}
#elif defined(__VXE__) || defined(__VXE2__)
for (; i + 3 < n; i += 4) {
float32x4_t val = vec_sub(vec_xl(0, x + i), vec_splats(mean));
vec_xst(val, 0, y + i);
val = vec_mul(val, val);
sum += (ggml_float)vec_hsum_f32x4(val);
}
#endif
for (; i < n; ++i) {
float val = x[i] - mean;
y[i] = val;
val *= val;
sum += (ggml_float)val;
}
return sum/n;
}
ggml_float ggml_vec_soft_max_f32(const int n, float * y, const float * x, float max) {
int i = 0;
ggml_float sum = 0;
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment