vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

metal : require contiguousness for float4 unary kernels (cont)

Browse source
This commit is contained in:
Georgi Gerganov 2024-04-15 22:51:42 +03:00
parent 7c1ab98183
commit 0010280042
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
1 changed files with 7 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

11
ggml-metal.m
View file

@ -1149,7 +1149,7 @@ static enum ggml_status ggml_metal_graph_compute(
id<MTLComputePipelineState> pipeline = nil; id<MTLComputePipelineState> pipeline = nil;
if (n % 4 == 0 && ggml_is_contiguous(src0)) { if (n % 4 == 0) {
n /= 4; n /= 4;
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_SCALE_4].pipeline; pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_SCALE_4].pipeline;
} else { } else {
@ -1184,6 +1184,9 @@ static enum ggml_status ggml_metal_graph_compute(
} break; } break;
case GGML_OP_UNARY: case GGML_OP_UNARY:
switch (ggml_get_unary_op(gf->nodes[i])) { switch (ggml_get_unary_op(gf->nodes[i])) {
// we are not taking into account the strides, so for now require contiguous tensors
GGML_ASSERT(ggml_is_contiguous(src0));
case GGML_UNARY_OP_TANH: case GGML_UNARY_OP_TANH:
{ {
id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_TANH].pipeline; id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_TANH].pipeline;
@ -1214,7 +1217,7 @@ static enum ggml_status ggml_metal_graph_compute(
id<MTLComputePipelineState> pipeline = nil; id<MTLComputePipelineState> pipeline = nil;
if (n % 4 == 0 && ggml_is_contiguous(src0)) { if (n % 4 == 0) {
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GELU_4].pipeline; pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GELU_4].pipeline;
n /= 4; n /= 4;
} else { } else {
@ -1233,7 +1236,7 @@ static enum ggml_status ggml_metal_graph_compute(
id<MTLComputePipelineState> pipeline = nil; id<MTLComputePipelineState> pipeline = nil;
if (n % 4 == 0 && ggml_is_contiguous(src0)) { if (n % 4 == 0) {
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GELU_QUICK_4].pipeline; pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GELU_QUICK_4].pipeline;
n /= 4; n /= 4;
} else { } else {
@ -1252,7 +1255,7 @@ static enum ggml_status ggml_metal_graph_compute(
id<MTLComputePipelineState> pipeline = nil; id<MTLComputePipelineState> pipeline = nil;
if (n % 4 == 0 && ggml_is_contiguous(src0)) { if (n % 4 == 0) {
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_SILU_4].pipeline; pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_SILU_4].pipeline;
n /= 4; n /= 4;
} else { } else {