Skip to content

Metal implementatio for the trellis quants. #475

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 8 commits into from
Jun 1, 2025
Merged

Metal implementatio for the trellis quants. #475

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
983844e
iq2_kt: Metal dequantize
Kawrakow May 30, 2025
eeeca31
iq2_kt: Metal GEMV
Kawrakow May 30, 2025
2396cc3
iq3_kt: Metal dequantize
Kawrakow May 30, 2025
ad52554
iq3_kt: Metal GEMV
Kawrakow May 30, 2025
07663b2
iq4_kt: Metal dequantize - getting NaNs
Kawrakow May 30, 2025
b687b2b
iq4_kt: Metal GEMV - also not working
Kawrakow May 30, 2025
d14eb93
iq4_kt: Metal still not working
Kawrakow Jun 1, 2025
a7fa24a
Disable iq4_kt on Metal for now
Kawrakow Jun 1, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
90 changes: 88 additions & 2 deletions ggml/src/ggml-metal.m
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -115,6 +115,9 @@
GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ4_K,
GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ5_K,
GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ6_K,
GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ2_KT,
GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ3_KT,
//GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ4_KT,
GGML_METAL_KERNEL_TYPE_GET_ROWS_I32,
GGML_METAL_KERNEL_TYPE_RMS_NORM,
GGML_METAL_KERNEL_TYPE_FUSED_RMS_NORM,
Expand Down Expand Up @@ -158,6 +161,9 @@
GGML_METAL_KERNEL_TYPE_MUL_MV_IQ4_K_F32,
GGML_METAL_KERNEL_TYPE_MUL_MV_IQ5_K_F32,
GGML_METAL_KERNEL_TYPE_MUL_MV_IQ6_K_F32,
GGML_METAL_KERNEL_TYPE_MUL_MV_IQ2_KT_F32,
GGML_METAL_KERNEL_TYPE_MUL_MV_IQ3_KT_F32,
//GGML_METAL_KERNEL_TYPE_MUL_MV_IQ4_KT_F32,
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_F32_F32,
//GGML_METAL_KERNEL_TYPE_MUL_MV_ID_F16_F16,
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_F16_F32,
Expand Down Expand Up @@ -195,6 +201,9 @@
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ4_K_F32,
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ5_K_F32,
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ6_K_F32,
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ2_KT_F32,
GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ3_KT_F32,
//GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ4_KT_F32,
GGML_METAL_KERNEL_TYPE_MUL_MM_F32_F32,
GGML_METAL_KERNEL_TYPE_MUL_MM_F16_F32,
GGML_METAL_KERNEL_TYPE_MUL_MM_BF16_F32,
Expand Down Expand Up @@ -229,6 +238,9 @@
GGML_METAL_KERNEL_TYPE_MUL_MM_IQ4_K_F32,
GGML_METAL_KERNEL_TYPE_MUL_MM_IQ5_K_F32,
GGML_METAL_KERNEL_TYPE_MUL_MM_IQ6_K_F32,
GGML_METAL_KERNEL_TYPE_MUL_MM_IQ2_KT_F32,
GGML_METAL_KERNEL_TYPE_MUL_MM_IQ3_KT_F32,
//GGML_METAL_KERNEL_TYPE_MUL_MM_IQ4_KT_F32,
GGML_METAL_KERNEL_TYPE_MUL_MM_F32_F16,
GGML_METAL_KERNEL_TYPE_MUL_MM_F16_F16,
GGML_METAL_KERNEL_TYPE_MUL_MM_BF16_F16,
Expand Down Expand Up @@ -263,6 +275,9 @@
GGML_METAL_KERNEL_TYPE_MUL_MM_IQ4_K_F16,
GGML_METAL_KERNEL_TYPE_MUL_MM_IQ5_K_F16,
GGML_METAL_KERNEL_TYPE_MUL_MM_IQ6_K_F16,
GGML_METAL_KERNEL_TYPE_MUL_MM_IQ2_KT_F16,
GGML_METAL_KERNEL_TYPE_MUL_MM_IQ3_KT_F16,
//GGML_METAL_KERNEL_TYPE_MUL_MM_IQ4_KT_F16,
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_F32_F32,
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_F16_F32,
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_BF16_F32,
Expand Down Expand Up @@ -297,6 +312,9 @@
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ4_K_F32,
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ5_K_F32,
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ6_K_F32,
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ2_KT_F32,
GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ3_KT_F32,
//GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ4_KT_F32,
GGML_METAL_KERNEL_TYPE_ROPE_NORM_F32,
GGML_METAL_KERNEL_TYPE_ROPE_NORM_F16,
GGML_METAL_KERNEL_TYPE_ROPE_NEOX_F32,
Expand Down Expand Up @@ -747,6 +765,9 @@ static void ggml_metal_log(enum ggml_log_level level, const char * format, ...){
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ4_K, get_rows_iq4_k, true);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ5_K, get_rows_iq5_k, true);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ6_K, get_rows_iq6_k, true);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ2_KT, get_rows_iq2_kt, true);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ3_KT, get_rows_iq3_kt, true);
//GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ4_KT, get_rows_iq4_kt, true);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_GET_ROWS_I32, get_rows_i32, true);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_RMS_NORM, rms_norm, ctx->support_simdgroup_reduction);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_FUSED_RMS_NORM, fused_rms_norm, ctx->support_simdgroup_reduction);
Expand Down Expand Up @@ -790,6 +811,9 @@ static void ggml_metal_log(enum ggml_log_level level, const char * format, ...){
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_IQ4_K_F32, mul_mv_iq4_k_f32, ctx->support_simdgroup_reduction);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_IQ5_K_F32, mul_mv_iq5_k_f32, ctx->support_simdgroup_reduction);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_IQ6_K_F32, mul_mv_iq6_k_f32, ctx->support_simdgroup_reduction);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_IQ2_KT_F32, mul_mv_iq2_kt_f32, ctx->support_simdgroup_reduction);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_IQ3_KT_F32, mul_mv_iq3_kt_f32, ctx->support_simdgroup_reduction);
//GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_IQ4_KT_F32, mul_mv_iq4_kt_f32, ctx->support_simdgroup_reduction);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_F32_F32, mul_mv_id_f32_f32, ctx->support_simdgroup_reduction);
//GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_F16_F16, mul_mv_id_f16_f16, ctx->support_simdgroup_reduction);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_F16_F32, mul_mv_id_f16_f32, ctx->support_simdgroup_reduction);
Expand Down Expand Up @@ -827,6 +851,9 @@ static void ggml_metal_log(enum ggml_log_level level, const char * format, ...){
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ4_K_F32, mul_mv_id_iq4_k_f32, ctx->support_simdgroup_reduction);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ5_K_F32, mul_mv_id_iq5_k_f32, ctx->support_simdgroup_reduction);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ6_K_F32, mul_mv_id_iq6_k_f32, ctx->support_simdgroup_reduction);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ2_KT_F32, mul_mv_id_iq2_kt_f32, ctx->support_simdgroup_reduction);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ3_KT_F32, mul_mv_id_iq3_kt_f32, ctx->support_simdgroup_reduction);
//GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ4_KT_F32, mul_mv_id_iq4_kt_f32, ctx->support_simdgroup_reduction);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_F32_F32, mul_mm_f32_f32, ctx->support_simdgroup_mm);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_F16_F32, mul_mm_f16_f32, ctx->support_simdgroup_mm);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_BF16_F32, mul_mm_bf16_f32, ctx->support_simdgroup_mm);
Expand Down Expand Up @@ -861,6 +888,9 @@ static void ggml_metal_log(enum ggml_log_level level, const char * format, ...){
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_IQ4_K_F32, mul_mm_iq4_k_f32, ctx->support_simdgroup_mm);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_IQ5_K_F32, mul_mm_iq5_k_f32, ctx->support_simdgroup_mm);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_IQ6_K_F32, mul_mm_iq6_k_f32, ctx->support_simdgroup_mm);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_IQ2_KT_F32, mul_mm_iq2_kt_f32, ctx->support_simdgroup_mm);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_IQ3_KT_F32, mul_mm_iq3_kt_f32, ctx->support_simdgroup_mm);
//GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_IQ4_KT_F32, mul_mm_iq4_kt_f32, ctx->support_simdgroup_mm);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_F32_F16, mul_mm_f32_f16, ctx->support_simdgroup_mm);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_F16_F16, mul_mm_f16_f16, ctx->support_simdgroup_mm);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_BF16_F16, mul_mm_bf16_f16, ctx->support_simdgroup_mm);
Expand Down Expand Up @@ -895,6 +925,9 @@ static void ggml_metal_log(enum ggml_log_level level, const char * format, ...){
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_IQ4_K_F16, mul_mm_iq4_k_f16, ctx->support_simdgroup_mm);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_IQ5_K_F16, mul_mm_iq5_k_f16, ctx->support_simdgroup_mm);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_IQ6_K_F16, mul_mm_iq6_k_f16, ctx->support_simdgroup_mm);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_IQ2_KT_F16, mul_mm_iq2_kt_f16, ctx->support_simdgroup_mm);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_IQ3_KT_F16, mul_mm_iq3_kt_f16, ctx->support_simdgroup_mm);
//GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_IQ4_KT_F16, mul_mm_iq4_kt_f16, ctx->support_simdgroup_mm);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_F32_F32, mul_mm_id_f32_f32, ctx->support_simdgroup_mm);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_F16_F32, mul_mm_id_f16_f32, ctx->support_simdgroup_mm);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_BF16_F32, mul_mm_id_bf16_f32, ctx->support_simdgroup_mm);
Expand Down Expand Up @@ -929,6 +962,9 @@ static void ggml_metal_log(enum ggml_log_level level, const char * format, ...){
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ4_K_F32, mul_mm_id_iq4_k_f32, ctx->support_simdgroup_mm);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ5_K_F32, mul_mm_id_iq5_k_f32, ctx->support_simdgroup_mm);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ6_K_F32, mul_mm_id_iq6_k_f32, ctx->support_simdgroup_mm);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ2_KT_F32, mul_mm_id_iq2_kt_f32, ctx->support_simdgroup_mm);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ3_KT_F32, mul_mm_id_iq3_kt_f32, ctx->support_simdgroup_mm);
//GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ4_KT_F32, mul_mm_id_iq4_kt_f32, ctx->support_simdgroup_mm);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_ROPE_NORM_F32, rope_norm_f32, true);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_ROPE_NORM_F16, rope_norm_f16, true);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_ROPE_NEOX_F32, rope_neox_f32, true);
Expand Down Expand Up @@ -2142,6 +2178,9 @@ static void ggml_metal_encode_node(
case GGML_TYPE_IQ4_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_IQ4_K_F32 ].pipeline; break;
case GGML_TYPE_IQ5_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_IQ5_K_F32 ].pipeline; break;
case GGML_TYPE_IQ6_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_IQ6_K_F32 ].pipeline; break;
case GGML_TYPE_IQ2_KT: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_IQ2_KT_F32 ].pipeline; break;
case GGML_TYPE_IQ3_KT: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_IQ3_KT_F32 ].pipeline; break;
//case GGML_TYPE_IQ4_KT: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_IQ4_KT_F32 ].pipeline; break;
default: GGML_ABORT("MUL MAT-MAT not implemented");
}
}
Expand Down Expand Up @@ -2181,6 +2220,9 @@ static void ggml_metal_encode_node(
case GGML_TYPE_IQ4_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_IQ4_K_F16 ].pipeline; break;
case GGML_TYPE_IQ5_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_IQ5_K_F16 ].pipeline; break;
case GGML_TYPE_IQ6_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_IQ6_K_F16 ].pipeline; break;
case GGML_TYPE_IQ2_KT: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_IQ2_KT_F16 ].pipeline; break;
case GGML_TYPE_IQ3_KT: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_IQ3_KT_F16 ].pipeline; break;
//case GGML_TYPE_IQ4_KT: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_IQ4_KT_F16 ].pipeline; break;
default: GGML_ABORT("MUL MAT-MAT not implemented");
}
}
Expand Down Expand Up @@ -2440,6 +2482,24 @@ static void ggml_metal_encode_node(
nth1 = 16;
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_IQ6_K_F32].pipeline;
} break;
case GGML_TYPE_IQ2_KT:
{
nth0 = 4;
nth1 = 16;
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_IQ2_KT_F32].pipeline;
} break;
case GGML_TYPE_IQ3_KT:
{
nth0 = 4;
nth1 = 16;
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_IQ3_KT_F32].pipeline;
} break;
//case GGML_TYPE_IQ4_KT:
// {
// nth0 = 4;
// nth1 = 16;
// pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_IQ4_KT_F32].pipeline;
// } break;
default:
{
GGML_METAL_LOG_ERROR("Asserting on type %d\n", (int)src0t);
Expand Down Expand Up @@ -2471,7 +2531,8 @@ static void ggml_metal_encode_node(
if (src0t == GGML_TYPE_Q4_0 || src0t == GGML_TYPE_Q4_1 || src0t == GGML_TYPE_Q5_0 ||
src0t == GGML_TYPE_Q5_1 || src0t == GGML_TYPE_Q8_0 || src0t == GGML_TYPE_Q2_K ||
src0t == GGML_TYPE_IQ1_S || src0t == GGML_TYPE_IQ1_M || src0t == GGML_TYPE_IQ2_S||
src0t == GGML_TYPE_IQ1_BN|| src0t == GGML_TYPE_IQ2_BN|| src0t == GGML_TYPE_Q6_0) {
src0t == GGML_TYPE_IQ1_BN|| src0t == GGML_TYPE_IQ2_BN|| src0t == GGML_TYPE_Q6_0 ||
src0t == GGML_TYPE_IQ2_KT|| src0t == GGML_TYPE_IQ3_KT) { //|| src0t == GGML_TYPE_IQ4_KT) {
[encoder dispatchThreadgroups:MTLSizeMake((ne01 + 7)/8, ne11, ne12*ne13) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
}
else if (src0t == GGML_TYPE_IQ2_KS || src0t == GGML_TYPE_IQ2_K || src0t == GGML_TYPE_IQ3_K) {
Expand Down Expand Up @@ -2596,6 +2657,9 @@ static void ggml_metal_encode_node(
case GGML_TYPE_IQ4_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ4_K_F32 ].pipeline; break;
case GGML_TYPE_IQ5_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ5_K_F32 ].pipeline; break;
case GGML_TYPE_IQ6_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ6_K_F32 ].pipeline; break;
case GGML_TYPE_IQ2_KT: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ2_KT_F32 ].pipeline; break;
case GGML_TYPE_IQ3_KT: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ3_KT_F32 ].pipeline; break;
//case GGML_TYPE_IQ4_KT: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MM_ID_IQ4_KT_F32 ].pipeline; break;
default: GGML_ABORT("MUL_MAT_ID not implemented");
}

Expand Down Expand Up @@ -2839,6 +2903,24 @@ static void ggml_metal_encode_node(
nth1 = 16;
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ6_K_F32].pipeline;
} break;
case GGML_TYPE_IQ2_KT:
{
nth0 = 4;
nth1 = 16;
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ2_KT_F32].pipeline;
} break;
case GGML_TYPE_IQ3_KT:
{
nth0 = 4;
nth1 = 16;
pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ3_KT_F32].pipeline;
} break;
//case GGML_TYPE_IQ4_KT:
// {
// nth0 = 4;
// nth1 = 16;
// pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ4_KT_F32].pipeline;
// } break;
default:
{
GGML_METAL_LOG_ERROR("Asserting on type %d\n", (int)src2t);
Expand Down Expand Up @@ -2881,7 +2963,8 @@ static void ggml_metal_encode_node(
if (src0t == GGML_TYPE_Q4_0 || src0t == GGML_TYPE_Q4_1 || src0t == GGML_TYPE_Q5_0 ||
src0t == GGML_TYPE_Q5_1 || src0t == GGML_TYPE_Q8_0 || src0t == GGML_TYPE_Q2_K ||
src0t == GGML_TYPE_IQ1_S || src0t == GGML_TYPE_IQ1_M || src0t == GGML_TYPE_Q6_0 ||
src0t == GGML_TYPE_IQ1_BN|| src0t == GGML_TYPE_IQ2_BN|| src0t == GGML_TYPE_IQ2_K) {
src0t == GGML_TYPE_IQ1_BN|| src0t == GGML_TYPE_IQ2_BN|| src0t == GGML_TYPE_IQ2_K||
src0t == GGML_TYPE_IQ2_KT|| src0t == GGML_TYPE_IQ3_KT) { //|| src0t == GGML_TYPE_IQ4_KT) {
[encoder dispatchThreadgroups:MTLSizeMake((ne01 + 7)/8, _ne1, tgz) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
}
else if (src0t == GGML_TYPE_IQ2_KS || src0t == GGML_TYPE_IQ2_K || src0t == GGML_TYPE_IQ3_K) {
Expand Down Expand Up @@ -2962,6 +3045,9 @@ static void ggml_metal_encode_node(
case GGML_TYPE_IQ4_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ4_K ].pipeline; break;
case GGML_TYPE_IQ5_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ5_K ].pipeline; break;
case GGML_TYPE_IQ6_K: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ6_K ].pipeline; break;
case GGML_TYPE_IQ2_KT: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ2_KT ].pipeline; break;
case GGML_TYPE_IQ3_KT: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ3_KT ].pipeline; break;
//case GGML_TYPE_IQ4_KT: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_IQ4_KT ].pipeline; break;
case GGML_TYPE_I32: pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GET_ROWS_I32 ].pipeline; break;
default: GGML_ABORT("not implemented");
}
Expand Down
Loading