Skip to content

Commit

Permalink
fix: cann buffer alloc zero
Browse files Browse the repository at this point in the history 
Signed-off-by: thxCode <thxcode0824@gmail.com>
  • Loading branch information
@thxCode
thxCode committed Oct 25, 2024
1 parent 5877391 commit 39631bb
Showing 1 changed file with 31 additions and 14 deletions.
45 changes: 31 additions & 14 deletions llama-box/patches/cann.patch
View file
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
diff --git a/ggml/src/ggml-cann.cpp b/ggml/src/ggml-cann.cpp
index af0fb603..d7f73859 100644
index af0fb603..d20407ce 100644
--- a/ggml/src/ggml-cann.cpp
+++ b/ggml/src/ggml-cann.cpp
@@ -122,6 +122,10 @@ static ggml_cann_device_info ggml_cann_init() {
Expand Down Expand Up @@ -82,7 +82,7 @@ index af0fb603..d7f73859 100644
- size = alignment * ((size + alignment - 1) / alignment);
+ size = GGML_PAD(size, alignment);
+ if (size == 0) {
+ size = alignment;
+ size = alignment;
+ }

size_t avail = pool_size - pool_used;
Expand Down Expand Up @@ -192,11 +192,11 @@ index af0fb603..d7f73859 100644
- (ggml_backend_cann_buffer_type_context*)buft->context;
-
- ggml_cann_set_device(buft_ctx->device);
-
- size = std::max(size, (size_t)1);
+ ggml_backend_cann_context* cann_ctx =
+ (ggml_backend_cann_context*)buft->device->context;

- size = std::max(size, (size_t)1);
-
- void* dev_ptr;
- aclError err = aclrtMalloc(&dev_ptr, size, ACL_MEM_MALLOC_HUGE_FIRST);
- if (err != ACL_SUCCESS) {
Expand Down Expand Up @@ -227,7 +227,24 @@ index af0fb603..d7f73859 100644
/* .context = */
new ggml_backend_cann_buffer_type_context{
i, "CANN" + std::to_string(i)},
@@ -1706,11 +1710,15 @@ static bool ggml_backend_cann_supports_op(ggml_backend_dev_t dev,
@@ -1215,10 +1219,15 @@ static void * ggml_cann_host_malloc(size_t size) {
return nullptr;
}

+ const size_t alignment = 128;
+ size = GGML_PAD(size, alignment);
+ if (size == 0) {
+ size = alignment;
+ }
+
void * hostPtr = nullptr;
aclError err = aclrtMallocHost((void **) &hostPtr, size);
if (err != ACL_SUCCESS) {
-
GGML_LOG_WARN("%s: failed to allocate %.2f MiB of pinned memory: %s\n", __func__,
size / 1024.0 / 1024.0, aclGetRecentErrMsg());
return nullptr;
@@ -1706,11 +1715,15 @@ static bool ggml_backend_cann_supports_op(ggml_backend_dev_t dev,
switch (op->src[0]->type) {
case GGML_TYPE_F16:
case GGML_TYPE_F32:
Expand All @@ -243,7 +260,7 @@ index af0fb603..d7f73859 100644
return true;
default:
return false;
@@ -1723,8 +1731,12 @@ static bool ggml_backend_cann_supports_op(ggml_backend_dev_t dev,
@@ -1723,8 +1736,12 @@ static bool ggml_backend_cann_supports_op(ggml_backend_dev_t dev,
switch (op->src[0]->type) {
case GGML_TYPE_F32:
case GGML_TYPE_F16:
Expand All @@ -256,7 +273,7 @@ index af0fb603..d7f73859 100644
return true;
default:
return false;
@@ -1734,8 +1746,12 @@ static bool ggml_backend_cann_supports_op(ggml_backend_dev_t dev,
@@ -1734,8 +1751,12 @@ static bool ggml_backend_cann_supports_op(ggml_backend_dev_t dev,
switch (op->type) {
case GGML_TYPE_F32:
case GGML_TYPE_F16:
Expand All @@ -269,7 +286,7 @@ index af0fb603..d7f73859 100644
return true;
default:
return false;
@@ -1902,17 +1918,17 @@ struct ggml_backend_cann_device_context {
@@ -1902,17 +1923,17 @@ struct ggml_backend_cann_device_context {
};

static const char * ggml_backend_cann_device_get_name(ggml_backend_dev_t dev) {
Expand All @@ -290,7 +307,7 @@ index af0fb603..d7f73859 100644
ggml_backend_cann_get_device_memory(ctx->device, free, total);
}

@@ -1939,7 +1955,7 @@ static void ggml_backend_cann_device_get_props(ggml_backend_dev_t dev, ggml_back
@@ -1939,7 +1960,7 @@ static void ggml_backend_cann_device_get_props(ggml_backend_dev_t dev, ggml_back

static ggml_backend_t ggml_backend_cann_device_init(ggml_backend_dev_t dev, const char * params) {
GGML_UNUSED(params);
Expand All @@ -299,7 +316,7 @@ index af0fb603..d7f73859 100644
return ggml_backend_cann_init(ctx->device);
}

@@ -1959,7 +1975,7 @@ static ggml_backend_t ggml_backend_cann_device_init(ggml_backend_dev_t dev, cons
@@ -1959,7 +1980,7 @@ static ggml_backend_t ggml_backend_cann_device_init(ggml_backend_dev_t dev, cons
static bool ggml_backend_cann_supports_buft(
ggml_backend_dev_t dev, ggml_backend_buffer_type_t buft) {
if (ggml_backend_buft_is_cann(buft)) {
Expand All @@ -308,7 +325,7 @@ index af0fb603..d7f73859 100644
ggml_backend_cann_buffer_type_context * buft_ctx =
(ggml_backend_cann_buffer_type_context *)buft->context;
return buft_ctx->device == dev_ctx->device;
@@ -1968,7 +1984,7 @@ static bool ggml_backend_cann_supports_buft(
@@ -1968,7 +1989,7 @@ static bool ggml_backend_cann_supports_buft(
}

static ggml_backend_buffer_type_t ggml_backend_cann_device_get_buffer_type(ggml_backend_dev_t dev) {
Expand All @@ -317,7 +334,7 @@ index af0fb603..d7f73859 100644
return ggml_backend_cann_buffer_type(ctx->device);
}

@@ -1989,7 +2005,7 @@ static ggml_backend_buffer_type_t ggml_backend_cann_device_get_host_buffer_type(
@@ -1989,7 +2010,7 @@ static ggml_backend_buffer_type_t ggml_backend_cann_device_get_host_buffer_type(
*/
static ggml_backend_event_t ggml_backend_cann_device_event_new(
ggml_backend_dev_t dev) {
Expand All @@ -326,7 +343,7 @@ index af0fb603..d7f73859 100644

ggml_cann_set_device(dev_ctx->device);

@@ -2097,11 +2113,7 @@ ggml_backend_reg_t ggml_backend_cann_reg() {
@@ -2097,11 +2118,7 @@ ggml_backend_reg_t ggml_backend_cann_reg() {
ggml_backend_cann_reg_context * ctx = new ggml_backend_cann_reg_context;

for (int i = 0; i < ggml_cann_info().device_count; i++) {
Expand All @@ -339,7 +356,7 @@ index af0fb603..d7f73859 100644
ggml_backend_dev_t dev = new ggml_backend_device {
/* .interface = */ ggml_backend_cann_device_interface,
/* .reg = */ &reg,
@@ -2129,17 +2141,12 @@ ggml_backend_t ggml_backend_cann_init(int32_t device) {
@@ -2129,17 +2146,12 @@ ggml_backend_t ggml_backend_cann_init(int32_t device) {
return nullptr;
}

Expand Down

0 comments on commit 39631bb

Please sign in to comment.