Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

metal : add f16 and bf16 support for concat operator (#24724)

Browse Source 
* metal : add f16 and bf16 support for concat operator

Extend the Metal backend concat operator to support f16 and bf16 tensor
types in addition to the existing f32 and i32 support.

- Template kernel_concat on type T with specializations for float, half,
  bfloat, and int
- Add type-specific pipeline getter ggml_metal_library_get_pipeline_concat()
- Update device support check to allow f16 unconditionally and bf16 when
  device supports bfloat16
- Update dispatch to select the correct kernel specialization by type

Assisted-by: pi:llama.cpp/Qwen3.6-27B

* metal : extend concat operator to support f16, bf16, i8, i16 and i64

Assisted-by: pi:llama.cpp/Qwen3.6-27B
This commit is contained in:
Georgi Gerganov
2026-06-17 19:38:55 +03:00
committed by GitHub
parent 4b4d13ae72
commit 8d2e580632
5 changed files with 56 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-metal/ggml-metal.metal
+23 -12
View File
@@ -7513,14 +7513,15 @@ template [[host_name("kernel_cpy_q5_0_f16")]] kernel cpy_q_f_t kernel_cpy_q_f32<
template [[host_name("kernel_cpy_q5_1_f16")]] kernel cpy_q_f_t kernel_cpy_q_f32<half4x4, block_q5_1, 2, dequantize_q5_1>;
template [[host_name("kernel_cpy_q8_0_f16")]] kernel cpy_q_f_t kernel_cpy_q_f32<half4x4, block_q8_0, 2, dequantize_q8_0>;
template<typename T>
kernel void kernel_concat(
constant ggml_metal_kargs_concat & args,
device const char * src0,
device const char * src1,
device char * dst,
uint3 tgpig[[threadgroup_position_in_grid]],
ushort3 tpitg[[thread_position_in_threadgroup]],
ushort3 ntg[[threads_per_threadgroup]]) {
constant ggml_metal_kargs_concat & args,
device const char * src0,
device const char * src1,
device char * dst,
uint3 tgpig[[threadgroup_position_in_grid]],
ushort3 tpitg[[thread_position_in_threadgroup]],
ushort3 ntg[[threads_per_threadgroup]]) {
const int i3 = tgpig.z;
const int i2 = tgpig.y;
@@ -7533,21 +7534,31 @@ kernel void kernel_concat(
int o[4] = {0, 0, 0, 0};
o[args.dim] = args.dim == 0 ? args.ne00 : (args.dim == 1 ? args.ne01 : (args.dim == 2 ? args.ne02 : args.ne03));
device const float * x;
for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
device const T * x;
if (i0 < args.ne00 && i1 < args.ne01 && i2 < args.ne02 && i3 < args.ne03) {
x = (device const float *)(src0 + (i3 )*args.nb03 + (i2 )*args.nb02 + (i1 )*args.nb01 + (i0 )*args.nb00);
x = (device const T *)(src0 + (i3 )*args.nb03 + (i2 )*args.nb02 + (i1 )*args.nb01 + (i0 )*args.nb00);
} else {
x = (device const float *)(src1 + (i3 - o[3])*args.nb13 + (i2 - o[2])*args.nb12 + (i1 - o[1])*args.nb11 + (i0 - o[0])*args.nb10);
x = (device const T *)(src1 + (i3 - o[3])*args.nb13 + (i2 - o[2])*args.nb12 + (i1 - o[1])*args.nb11 + (i0 - o[0])*args.nb10);
}
device float * y = (device float *)(dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + i0*args.nb0);
device T * y = (device T *)(dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + i0*args.nb0);
*y = *x;
}
}
typedef decltype(kernel_concat<float>) kernel_concat_t;
template [[host_name("kernel_concat_f32")]] kernel kernel_concat_t kernel_concat<float>;
template [[host_name("kernel_concat_f16")]] kernel kernel_concat_t kernel_concat<half>;
template [[host_name("kernel_concat_bf16")]] kernel kernel_concat_t kernel_concat<bfloat>;
template [[host_name("kernel_concat_i8")]] kernel kernel_concat_t kernel_concat<char>;
template [[host_name("kernel_concat_i16")]] kernel kernel_concat_t kernel_concat<short>;
template [[host_name("kernel_concat_i32")]] kernel kernel_concat_t kernel_concat<int>;
template [[host_name("kernel_concat_i64")]] kernel kernel_concat_t kernel_concat<long>;
template<int nr0, typename args_t>
void kernel_mul_mv_q2_K_f32_impl(
args_t args,