mirror of
https://github.com/ikawrakow/ik_llama.cpp.git
synced 2026-04-28 02:11:50 +00:00
Fix iq4_k_r4 on NEON
This commit is contained in:
@@ -3225,6 +3225,8 @@ bool iqk_set_kernels_iqk_quants(int ne00, int typeA, int typeB, std::array<mul_m
|
|||||||
return false;
|
return false;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
func16 = nullptr;
|
||||||
|
|
||||||
switch (typeA) {
|
switch (typeA) {
|
||||||
case GGML_TYPE_IQ2_KS:
|
case GGML_TYPE_IQ2_KS:
|
||||||
IQK_SET_MUL_MAT_FUNCTIONS_T(mul_mat_qX_K_q8_K_T, DequantizerIQ2KS, kernels);
|
IQK_SET_MUL_MAT_FUNCTIONS_T(mul_mat_qX_K_q8_K_T, DequantizerIQ2KS, kernels);
|
||||||
@@ -3267,8 +3269,10 @@ bool iqk_set_kernels_iqk_quants(int ne00, int typeA, int typeB, std::array<mul_m
|
|||||||
break;
|
break;
|
||||||
case GGML_TYPE_IQ5_KS_R4:
|
case GGML_TYPE_IQ5_KS_R4:
|
||||||
IQK_SET_MUL_MAT_FUNCTIONS(mul_mat_iq5_ks_r4_q8_k, kernels);
|
IQK_SET_MUL_MAT_FUNCTIONS(mul_mat_iq5_ks_r4_q8_k, kernels);
|
||||||
|
break;
|
||||||
case GGML_TYPE_IQ5_K_R4:
|
case GGML_TYPE_IQ5_K_R4:
|
||||||
IQK_SET_MUL_MAT_FUNCTIONS(mul_mat_iq5_k_r4_q8_k, kernels);
|
IQK_SET_MUL_MAT_FUNCTIONS(mul_mat_iq5_k_r4_q8_k, kernels);
|
||||||
|
break;
|
||||||
default:
|
default:
|
||||||
return false;
|
return false;
|
||||||
}
|
}
|
||||||
|
|||||||
@@ -2836,6 +2836,7 @@ bool iqk_set_kernels_kquants(int ne00, int typeA, int typeB, std::array<mul_mat_
|
|||||||
IQK_SET_MUL_MAT_FUNCTIONS(mul_mat_q8_KV_q8_KV, kernels)
|
IQK_SET_MUL_MAT_FUNCTIONS(mul_mat_q8_KV_q8_KV, kernels)
|
||||||
kernels[0] = mul_mat_q8_KV_q8_KV_1;
|
kernels[0] = mul_mat_q8_KV_q8_KV_1;
|
||||||
func16 = mul_mat_q8_KV_q8_KV<16>;
|
func16 = mul_mat_q8_KV_q8_KV<16>;
|
||||||
|
break;
|
||||||
case GGML_TYPE_Q8_KV_R8:
|
case GGML_TYPE_Q8_KV_R8:
|
||||||
IQK_SET_MUL_MAT_FUNCTIONS(mul_mat_q8_KV_r8_q8_KV, kernels);
|
IQK_SET_MUL_MAT_FUNCTIONS(mul_mat_q8_KV_r8_q8_KV, kernels);
|
||||||
break;
|
break;
|
||||||
|
|||||||
Reference in New Issue
Block a user