ggml-cpu: ARM64: repack version of q8_0 (dotprod and i8mm) (#18096)
* wip: skeleton for q8_0 repack * q8_0 repack GEMV implementations * GEMM implementations * Formatting * Fixed format consistency of repack gemm and gemv declarations * gemv and gemm generic location consistent with declarations * Removed non-correct unused variables statements * Cleanup, consistent style * Missing generic fallbacks for x86 and powerpc
This commit is contained in:
@@ -43,6 +43,8 @@
|
||||
#define ggml_gemv_q2_K_8x8_q8_K_generic ggml_gemv_q2_K_8x8_q8_K
|
||||
#define ggml_gemv_iq4_nl_4x4_q8_0_generic ggml_gemv_iq4_nl_4x4_q8_0
|
||||
#define ggml_gemv_iq4_nl_8x8_q8_0_generic ggml_gemv_iq4_nl_8x8_q8_0
|
||||
#define ggml_gemv_q8_0_4x4_q8_0_generic ggml_gemv_q8_0_4x4_q8_0
|
||||
#define ggml_gemv_q8_0_4x8_q8_0_generic ggml_gemv_q8_0_4x8_q8_0
|
||||
#define ggml_gemm_q4_0_4x4_q8_0_generic ggml_gemm_q4_0_4x4_q8_0
|
||||
#define ggml_gemm_q4_0_4x8_q8_0_generic ggml_gemm_q4_0_4x8_q8_0
|
||||
#define ggml_gemm_q4_0_8x8_q8_0_generic ggml_gemm_q4_0_8x8_q8_0
|
||||
@@ -51,6 +53,8 @@
|
||||
#define ggml_gemm_q2_K_8x8_q8_K_generic ggml_gemm_q2_K_8x8_q8_K
|
||||
#define ggml_gemm_iq4_nl_4x4_q8_0_generic ggml_gemm_iq4_nl_4x4_q8_0
|
||||
#define ggml_gemm_iq4_nl_8x8_q8_0_generic ggml_gemm_iq4_nl_8x8_q8_0
|
||||
#define ggml_gemm_q8_0_4x4_q8_0_generic ggml_gemm_q8_0_4x4_q8_0
|
||||
#define ggml_gemm_q8_0_4x8_q8_0_generic ggml_gemm_q8_0_4x8_q8_0
|
||||
#elif defined(__aarch64__) || defined(__arm__) || defined(_M_ARM) || defined(_M_ARM64)
|
||||
// repack.cpp
|
||||
#define ggml_quantize_mat_q8_K_4x4_generic ggml_quantize_mat_q8_K_4x4
|
||||
@@ -67,10 +71,14 @@
|
||||
#define ggml_gemv_q4_0_4x8_q8_0_generic ggml_gemv_q4_0_4x8_q8_0
|
||||
#define ggml_gemv_q4_K_8x4_q8_K_generic ggml_gemv_q4_K_8x4_q8_K
|
||||
#define ggml_gemv_iq4_nl_4x4_q8_0_generic ggml_gemv_iq4_nl_4x4_q8_0
|
||||
#define ggml_gemv_q8_0_4x4_q8_0_generic ggml_gemv_q8_0_4x4_q8_0
|
||||
#define ggml_gemv_q8_0_4x8_q8_0_generic ggml_gemv_q8_0_4x8_q8_0
|
||||
#define ggml_gemm_q4_0_4x4_q8_0_generic ggml_gemm_q4_0_4x4_q8_0
|
||||
#define ggml_gemm_q4_0_4x8_q8_0_generic ggml_gemm_q4_0_4x8_q8_0
|
||||
#define ggml_gemm_q4_K_8x4_q8_K_generic ggml_gemm_q4_K_8x4_q8_K
|
||||
#define ggml_gemm_iq4_nl_4x4_q8_0_generic ggml_gemm_iq4_nl_4x4_q8_0
|
||||
#define ggml_gemm_q8_0_4x4_q8_0_generic ggml_gemm_q8_0_4x4_q8_0
|
||||
#define ggml_gemm_q8_0_4x8_q8_0_generic ggml_gemm_q8_0_4x8_q8_0
|
||||
#elif defined(__POWERPC__) || defined(__powerpc__)
|
||||
// ref: https://github.com/ggml-org/llama.cpp/pull/14146#issuecomment-2972561679
|
||||
// quants.c
|
||||
@@ -91,6 +99,8 @@
|
||||
#define ggml_gemv_q2_K_8x8_q8_K_generic ggml_gemv_q2_K_8x8_q8_K
|
||||
#define ggml_gemv_iq4_nl_4x4_q8_0_generic ggml_gemv_iq4_nl_4x4_q8_0
|
||||
#define ggml_gemv_iq4_nl_8x8_q8_0_generic ggml_gemv_iq4_nl_8x8_q8_0
|
||||
#define ggml_gemv_q8_0_4x4_q8_0_generic ggml_gemv_q8_0_4x4_q8_0
|
||||
#define ggml_gemv_q8_0_4x8_q8_0_generic ggml_gemv_q8_0_4x8_q8_0
|
||||
#define ggml_gemm_q4_0_4x4_q8_0_generic ggml_gemm_q4_0_4x4_q8_0
|
||||
#define ggml_gemm_q4_0_4x8_q8_0_generic ggml_gemm_q4_0_4x8_q8_0
|
||||
#define ggml_gemm_q4_0_8x8_q8_0_generic ggml_gemm_q4_0_8x8_q8_0
|
||||
@@ -99,6 +109,8 @@
|
||||
#define ggml_gemm_q2_K_8x8_q8_K_generic ggml_gemm_q2_K_8x8_q8_K
|
||||
#define ggml_gemm_iq4_nl_4x4_q8_0_generic ggml_gemm_iq4_nl_4x4_q8_0
|
||||
#define ggml_gemm_iq4_nl_8x8_q8_0_generic ggml_gemm_iq4_nl_8x8_q8_0
|
||||
#define ggml_gemm_q8_0_4x4_q8_0_generic ggml_gemm_q8_0_4x4_q8_0
|
||||
#define ggml_gemm_q8_0_4x8_q8_0_generic ggml_gemm_q8_0_4x8_q8_0
|
||||
#elif defined(__loongarch64)
|
||||
// quants.c
|
||||
#define quantize_row_q8_K_generic quantize_row_q8_K
|
||||
@@ -119,6 +131,8 @@
|
||||
#define ggml_gemv_q2_K_8x8_q8_K_generic ggml_gemv_q2_K_8x8_q8_K
|
||||
#define ggml_gemv_iq4_nl_4x4_q8_0_generic ggml_gemv_iq4_nl_4x4_q8_0
|
||||
#define ggml_gemv_iq4_nl_8x8_q8_0_generic ggml_gemv_iq4_nl_8x8_q8_0
|
||||
#define ggml_gemv_q8_0_4x4_q8_0_generic ggml_gemv_q8_0_4x4_q8_0
|
||||
#define ggml_gemv_q8_0_4x8_q8_0_generic ggml_gemv_q8_0_4x8_q8_0
|
||||
#define ggml_gemm_q4_0_4x4_q8_0_generic ggml_gemm_q4_0_4x4_q8_0
|
||||
#define ggml_gemm_q4_0_4x8_q8_0_generic ggml_gemm_q4_0_4x8_q8_0
|
||||
#define ggml_gemm_q4_0_8x8_q8_0_generic ggml_gemm_q4_0_8x8_q8_0
|
||||
@@ -127,6 +141,8 @@
|
||||
#define ggml_gemm_q2_K_8x8_q8_K_generic ggml_gemm_q2_K_8x8_q8_K
|
||||
#define ggml_gemm_iq4_nl_4x4_q8_0_generic ggml_gemm_iq4_nl_4x4_q8_0
|
||||
#define ggml_gemm_iq4_nl_8x8_q8_0_generic ggml_gemm_iq4_nl_8x8_q8_0
|
||||
#define ggml_gemm_q8_0_4x4_q8_0_generic ggml_gemm_q8_0_4x4_q8_0
|
||||
#define ggml_gemm_q8_0_4x8_q8_0_generic ggml_gemm_q8_0_4x8_q8_0
|
||||
#elif defined(__riscv)
|
||||
// quants.c
|
||||
#define quantize_row_q8_K_generic quantize_row_q8_K
|
||||
@@ -154,6 +170,8 @@
|
||||
#define ggml_gemv_q2_K_8x8_q8_K_generic ggml_gemv_q2_K_8x8_q8_K
|
||||
#define ggml_gemv_iq4_nl_4x4_q8_0_generic ggml_gemv_iq4_nl_4x4_q8_0
|
||||
#define ggml_gemv_iq4_nl_8x8_q8_0_generic ggml_gemv_iq4_nl_8x8_q8_0
|
||||
#define ggml_gemv_q8_0_4x4_q8_0_generic ggml_gemv_q8_0_4x4_q8_0
|
||||
#define ggml_gemv_q8_0_4x8_q8_0_generic ggml_gemv_q8_0_4x8_q8_0
|
||||
#define ggml_gemm_q4_0_4x4_q8_0_generic ggml_gemm_q4_0_4x4_q8_0
|
||||
#define ggml_gemm_q4_0_4x8_q8_0_generic ggml_gemm_q4_0_4x8_q8_0
|
||||
#define ggml_gemm_q4_K_8x4_q8_K_generic ggml_gemm_q4_K_8x4_q8_K
|
||||
@@ -161,6 +179,8 @@
|
||||
#define ggml_gemm_q2_K_8x8_q8_K_generic ggml_gemm_q2_K_8x8_q8_K
|
||||
#define ggml_gemm_iq4_nl_4x4_q8_0_generic ggml_gemm_iq4_nl_4x4_q8_0
|
||||
#define ggml_gemm_iq4_nl_8x8_q8_0_generic ggml_gemm_iq4_nl_8x8_q8_0
|
||||
#define ggml_gemm_q8_0_4x4_q8_0_generic ggml_gemm_q8_0_4x4_q8_0
|
||||
#define ggml_gemm_q8_0_4x8_q8_0_generic ggml_gemm_q8_0_4x8_q8_0
|
||||
#elif defined(__s390x__)
|
||||
// quants.c
|
||||
#define quantize_row_q8_K_generic quantize_row_q8_K
|
||||
@@ -187,6 +207,8 @@
|
||||
#define ggml_gemv_q2_K_8x8_q8_K_generic ggml_gemv_q2_K_8x8_q8_K
|
||||
#define ggml_gemv_iq4_nl_4x4_q8_0_generic ggml_gemv_iq4_nl_4x4_q8_0
|
||||
#define ggml_gemv_iq4_nl_8x8_q8_0_generic ggml_gemv_iq4_nl_8x8_q8_0
|
||||
#define ggml_gemv_q8_0_4x4_q8_0_generic ggml_gemv_q8_0_4x4_q8_0
|
||||
#define ggml_gemv_q8_0_4x8_q8_0_generic ggml_gemv_q8_0_4x8_q8_0
|
||||
#define ggml_gemm_q4_0_4x4_q8_0_generic ggml_gemm_q4_0_4x4_q8_0
|
||||
#define ggml_gemm_q4_0_4x8_q8_0_generic ggml_gemm_q4_0_4x8_q8_0
|
||||
#define ggml_gemm_q4_0_8x8_q8_0_generic ggml_gemm_q4_0_8x8_q8_0
|
||||
@@ -195,6 +217,8 @@
|
||||
#define ggml_gemm_q2_K_8x8_q8_K_generic ggml_gemm_q2_K_8x8_q8_K
|
||||
#define ggml_gemm_iq4_nl_4x4_q8_0_generic ggml_gemm_iq4_nl_4x4_q8_0
|
||||
#define ggml_gemm_iq4_nl_8x8_q8_0_generic ggml_gemm_iq4_nl_8x8_q8_0
|
||||
#define ggml_gemm_q8_0_4x4_q8_0_generic ggml_gemm_q8_0_4x4_q8_0
|
||||
#define ggml_gemm_q8_0_4x8_q8_0_generic ggml_gemm_q8_0_4x8_q8_0
|
||||
#elif defined(__wasm__)
|
||||
// quants.c
|
||||
#define ggml_vec_dot_q4_1_q8_1_generic ggml_vec_dot_q4_1_q8_1
|
||||
@@ -223,6 +247,8 @@
|
||||
#define ggml_gemv_q2_K_8x8_q8_K_generic ggml_gemv_q2_K_8x8_q8_K
|
||||
#define ggml_gemv_iq4_nl_4x4_q8_0_generic ggml_gemv_iq4_nl_4x4_q8_0
|
||||
#define ggml_gemv_iq4_nl_8x8_q8_0_generic ggml_gemv_iq4_nl_8x8_q8_0
|
||||
#define ggml_gemv_q8_0_4x4_q8_0_generic ggml_gemv_q8_0_4x4_q8_0
|
||||
#define ggml_gemv_q8_0_4x8_q8_0_generic ggml_gemv_q8_0_4x8_q8_0
|
||||
#define ggml_gemm_q4_0_4x4_q8_0_generic ggml_gemm_q4_0_4x4_q8_0
|
||||
#define ggml_gemm_q4_0_4x8_q8_0_generic ggml_gemm_q4_0_4x8_q8_0
|
||||
#define ggml_gemm_q4_0_8x8_q8_0_generic ggml_gemm_q4_0_8x8_q8_0
|
||||
@@ -231,4 +257,6 @@
|
||||
#define ggml_gemm_q2_K_8x8_q8_K_generic ggml_gemm_q2_K_8x8_q8_K
|
||||
#define ggml_gemm_iq4_nl_4x4_q8_0_generic ggml_gemm_iq4_nl_4x4_q8_0
|
||||
#define ggml_gemm_iq4_nl_8x8_q8_0_generic ggml_gemm_iq4_nl_8x8_q8_0
|
||||
#define ggml_gemm_q8_0_4x4_q8_0_generic ggml_gemm_q8_0_4x4_q8_0
|
||||
#define ggml_gemm_q8_0_4x8_q8_0_generic ggml_gemm_q8_0_4x8_q8_0
|
||||
#endif
|
||||
|
||||
Reference in New Issue
Block a user