whisper.rn 0.5.4 → 0.5.5

package/cpp/ggml-cpu/arch/x86/repack.cpp

@@ -646,7 +646,7 @@ static void gemm_q4_b32_8x8_q8_0_lut_avx(int n, float * WSP_GGML_RESTRICT s, siz
     __m256i requiredOrder = _mm256_set_epi32(3, 2, 1, 0, 7, 6, 5, 4);
     int64_t xstart = 0;
     int anr = nr - nr%16; // Used to align nr with boundary of 16
-#ifdef __AVX512F__
+#if defined(__AVX512BW__) && defined(__AVX512DQ__)
     int anc = nc - nc%16; // Used to align nc with boundary of 16
                           // Mask to mask out nibbles from packed bytes expanded to 512 bit length
     const __m512i m4bexpanded = _mm512_set1_epi8(0x0F);
@@ -1041,7 +1041,7 @@ static void gemm_q4_b32_8x8_q8_0_lut_avx(int n, float * WSP_GGML_RESTRICT s, siz
         xstart = anc/8;
         y = 0;
     }
-#endif // __AVX512F__
+#endif // __AVX512BW__ && __AVX512DQ__
 
     // Take group of four block_q8_0x4 structures at each pass of the loop and perform dot product operation
 
@@ -1989,7 +1989,7 @@ void wsp_ggml_gemm_q4_K_8x8_q8_K(int n, float * WSP_GGML_RESTRICT s, size_t bs,
     __m256i requiredOrder = _mm256_set_epi32(3, 2, 1, 0, 7, 6, 5, 4);
     int64_t xstart = 0;
     int anr = nr - nr % 16;; // Used to align nr with boundary of 16
-#ifdef __AVX512F__
+#if defined(__AVX512BW__) && defined(__AVX512DQ__)
     int anc = nc - nc % 16; // Used to align nc with boundary of 16
     // Mask to mask out nibbles from packed bytes expanded to 512 bit length
     const __m512i m4bexpanded = _mm512_set1_epi8(0x0F);
@@ -2727,7 +2727,7 @@ void wsp_ggml_gemm_q4_K_8x8_q8_K(int n, float * WSP_GGML_RESTRICT s, size_t bs,
         xstart = anc/8;
         y = 0;
     }
-#endif //AVX512F
+#endif // __AVX512BW__ && __AVX512DQ__
 
     // Take group of four block_q8_Kx4 structures at each pass of the loop and perform dot product operation
     for (; y < anr / 4; y += 4) {
@@ -3467,7 +3467,7 @@ void wsp_ggml_gemm_q2_K_8x8_q8_K(int n, float * WSP_GGML_RESTRICT s, size_t bs,
     __m256i scalesmask2 = _mm256_castsi128_si256(scalesmask2_sse);
     scalesmask2 = _mm256_permute2f128_si256(scalesmask2, scalesmask2, 0);
 
-#ifdef __AVX512F__
+#if defined(__AVX512BW__) && defined(__AVX512DQ__)
 
     int anc = nc - nc % 16; // Used to align nc with boundary of 16
 
@@ -4947,7 +4947,7 @@ void wsp_ggml_gemm_q2_K_8x8_q8_K(int n, float * WSP_GGML_RESTRICT s, size_t bs,
         y = 0;
     }
 
-#endif //AVX512F
+#endif // __AVX512BW__ && __AVX512DQ__
 
     // Take group of four block_q8_Kx4 structures at each pass of the loop and perform dot product operation
     for (; y < anr / 4; y += 4) {

package/cpp/ggml-cpu/arch-fallback.h

@@ -33,39 +33,52 @@
 // repack.cpp
 #define wsp_ggml_wsp_quantize_mat_q8_0_4x4_generic wsp_ggml_wsp_quantize_mat_q8_0_4x4
 #define wsp_ggml_wsp_quantize_mat_q8_0_4x8_generic wsp_ggml_wsp_quantize_mat_q8_0_4x8
+#define wsp_ggml_wsp_quantize_mat_q8_K_4x4_generic wsp_ggml_wsp_quantize_mat_q8_K_4x4
 #define wsp_ggml_wsp_quantize_mat_q8_K_4x8_generic wsp_ggml_wsp_quantize_mat_q8_K_4x8
 #define wsp_ggml_gemv_q4_0_4x4_q8_0_generic wsp_ggml_gemv_q4_0_4x4_q8_0
 #define wsp_ggml_gemv_q4_0_4x8_q8_0_generic wsp_ggml_gemv_q4_0_4x8_q8_0
 #define wsp_ggml_gemv_q4_0_8x8_q8_0_generic wsp_ggml_gemv_q4_0_8x8_q8_0
+#define wsp_ggml_gemv_q4_K_8x4_q8_K_generic wsp_ggml_gemv_q4_K_8x4_q8_K
 #define wsp_ggml_gemv_q4_K_8x8_q8_K_generic wsp_ggml_gemv_q4_K_8x8_q8_K
 #define wsp_ggml_gemv_q2_K_8x8_q8_K_generic wsp_ggml_gemv_q2_K_8x8_q8_K
 #define wsp_ggml_gemv_iq4_nl_4x4_q8_0_generic wsp_ggml_gemv_iq4_nl_4x4_q8_0
 #define wsp_ggml_gemv_iq4_nl_8x8_q8_0_generic wsp_ggml_gemv_iq4_nl_8x8_q8_0
+#define wsp_ggml_gemv_q8_0_4x4_q8_0_generic wsp_ggml_gemv_q8_0_4x4_q8_0
+#define wsp_ggml_gemv_q8_0_4x8_q8_0_generic wsp_ggml_gemv_q8_0_4x8_q8_0
 #define wsp_ggml_gemm_q4_0_4x4_q8_0_generic wsp_ggml_gemm_q4_0_4x4_q8_0
 #define wsp_ggml_gemm_q4_0_4x8_q8_0_generic wsp_ggml_gemm_q4_0_4x8_q8_0
 #define wsp_ggml_gemm_q4_0_8x8_q8_0_generic wsp_ggml_gemm_q4_0_8x8_q8_0
+#define wsp_ggml_gemm_q4_K_8x4_q8_K_generic wsp_ggml_gemm_q4_K_8x4_q8_K
 #define wsp_ggml_gemm_q4_K_8x8_q8_K_generic wsp_ggml_gemm_q4_K_8x8_q8_K
 #define wsp_ggml_gemm_q2_K_8x8_q8_K_generic wsp_ggml_gemm_q2_K_8x8_q8_K
 #define wsp_ggml_gemm_iq4_nl_4x4_q8_0_generic wsp_ggml_gemm_iq4_nl_4x4_q8_0
 #define wsp_ggml_gemm_iq4_nl_8x8_q8_0_generic wsp_ggml_gemm_iq4_nl_8x8_q8_0
+#define wsp_ggml_gemm_q8_0_4x4_q8_0_generic wsp_ggml_gemm_q8_0_4x4_q8_0
+#define wsp_ggml_gemm_q8_0_4x8_q8_0_generic wsp_ggml_gemm_q8_0_4x8_q8_0
 #elif defined(__aarch64__) || defined(__arm__) || defined(_M_ARM) || defined(_M_ARM64)
 // repack.cpp
+#define wsp_ggml_wsp_quantize_mat_q8_K_4x4_generic wsp_ggml_wsp_quantize_mat_q8_K_4x4
 #define wsp_ggml_wsp_quantize_mat_q8_K_4x8_generic wsp_ggml_wsp_quantize_mat_q8_K_4x8
-#define wsp_ggml_gemv_q4_K_8x8_q8_K_generic wsp_ggml_gemv_q4_K_8x8_q8_K
 #define wsp_ggml_gemv_iq4_nl_8x8_q8_0_generic wsp_ggml_gemv_iq4_nl_8x8_q8_0
 #define wsp_ggml_gemv_q2_K_8x8_q8_K_generic wsp_ggml_gemv_q2_K_8x8_q8_K
-#define wsp_ggml_gemm_q4_K_8x8_q8_K_generic wsp_ggml_gemm_q4_K_8x8_q8_K
 #define wsp_ggml_gemm_iq4_nl_8x8_q8_0_generic wsp_ggml_gemm_iq4_nl_8x8_q8_0
 #define wsp_ggml_gemm_q2_K_8x8_q8_K_generic wsp_ggml_gemm_q2_K_8x8_q8_K
 #elif defined(__x86_64__) || defined(__i386__) || defined(_M_IX86) || defined(_M_X64)
 // repack.cpp
 #define wsp_ggml_wsp_quantize_mat_q8_0_4x4_generic wsp_ggml_wsp_quantize_mat_q8_0_4x4
+#define wsp_ggml_wsp_quantize_mat_q8_K_4x4_generic wsp_ggml_wsp_quantize_mat_q8_K_4x4
 #define wsp_ggml_gemv_q4_0_4x4_q8_0_generic wsp_ggml_gemv_q4_0_4x4_q8_0
 #define wsp_ggml_gemv_q4_0_4x8_q8_0_generic wsp_ggml_gemv_q4_0_4x8_q8_0
+#define wsp_ggml_gemv_q4_K_8x4_q8_K_generic wsp_ggml_gemv_q4_K_8x4_q8_K
 #define wsp_ggml_gemv_iq4_nl_4x4_q8_0_generic wsp_ggml_gemv_iq4_nl_4x4_q8_0
+#define wsp_ggml_gemv_q8_0_4x4_q8_0_generic wsp_ggml_gemv_q8_0_4x4_q8_0
+#define wsp_ggml_gemv_q8_0_4x8_q8_0_generic wsp_ggml_gemv_q8_0_4x8_q8_0
 #define wsp_ggml_gemm_q4_0_4x4_q8_0_generic wsp_ggml_gemm_q4_0_4x4_q8_0
 #define wsp_ggml_gemm_q4_0_4x8_q8_0_generic wsp_ggml_gemm_q4_0_4x8_q8_0
+#define wsp_ggml_gemm_q4_K_8x4_q8_K_generic wsp_ggml_gemm_q4_K_8x4_q8_K
 #define wsp_ggml_gemm_iq4_nl_4x4_q8_0_generic wsp_ggml_gemm_iq4_nl_4x4_q8_0
+#define wsp_ggml_gemm_q8_0_4x4_q8_0_generic wsp_ggml_gemm_q8_0_4x4_q8_0
+#define wsp_ggml_gemm_q8_0_4x8_q8_0_generic wsp_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
@@ -76,21 +89,28 @@
 // repack.cpp
 #define wsp_ggml_wsp_quantize_mat_q8_0_4x4_generic wsp_ggml_wsp_quantize_mat_q8_0_4x4
 #define wsp_ggml_wsp_quantize_mat_q8_0_4x8_generic wsp_ggml_wsp_quantize_mat_q8_0_4x8
+#define wsp_ggml_wsp_quantize_mat_q8_K_4x4_generic wsp_ggml_wsp_quantize_mat_q8_K_4x4
 #define wsp_ggml_wsp_quantize_mat_q8_K_4x8_generic wsp_ggml_wsp_quantize_mat_q8_K_4x8
 #define wsp_ggml_gemv_q4_0_4x4_q8_0_generic wsp_ggml_gemv_q4_0_4x4_q8_0
 #define wsp_ggml_gemv_q4_0_4x8_q8_0_generic wsp_ggml_gemv_q4_0_4x8_q8_0
 #define wsp_ggml_gemv_q4_0_8x8_q8_0_generic wsp_ggml_gemv_q4_0_8x8_q8_0
+#define wsp_ggml_gemv_q4_K_8x4_q8_K_generic wsp_ggml_gemv_q4_K_8x4_q8_K
 #define wsp_ggml_gemv_q4_K_8x8_q8_K_generic wsp_ggml_gemv_q4_K_8x8_q8_K
 #define wsp_ggml_gemv_q2_K_8x8_q8_K_generic wsp_ggml_gemv_q2_K_8x8_q8_K
 #define wsp_ggml_gemv_iq4_nl_4x4_q8_0_generic wsp_ggml_gemv_iq4_nl_4x4_q8_0
 #define wsp_ggml_gemv_iq4_nl_8x8_q8_0_generic wsp_ggml_gemv_iq4_nl_8x8_q8_0
+#define wsp_ggml_gemv_q8_0_4x4_q8_0_generic wsp_ggml_gemv_q8_0_4x4_q8_0
+#define wsp_ggml_gemv_q8_0_4x8_q8_0_generic wsp_ggml_gemv_q8_0_4x8_q8_0
 #define wsp_ggml_gemm_q4_0_4x4_q8_0_generic wsp_ggml_gemm_q4_0_4x4_q8_0
 #define wsp_ggml_gemm_q4_0_4x8_q8_0_generic wsp_ggml_gemm_q4_0_4x8_q8_0
 #define wsp_ggml_gemm_q4_0_8x8_q8_0_generic wsp_ggml_gemm_q4_0_8x8_q8_0
+#define wsp_ggml_gemm_q4_K_8x4_q8_K_generic wsp_ggml_gemm_q4_K_8x4_q8_K
 #define wsp_ggml_gemm_q4_K_8x8_q8_K_generic wsp_ggml_gemm_q4_K_8x8_q8_K
 #define wsp_ggml_gemm_q2_K_8x8_q8_K_generic wsp_ggml_gemm_q2_K_8x8_q8_K
 #define wsp_ggml_gemm_iq4_nl_4x4_q8_0_generic wsp_ggml_gemm_iq4_nl_4x4_q8_0
 #define wsp_ggml_gemm_iq4_nl_8x8_q8_0_generic wsp_ggml_gemm_iq4_nl_8x8_q8_0
+#define wsp_ggml_gemm_q8_0_4x4_q8_0_generic wsp_ggml_gemm_q8_0_4x4_q8_0
+#define wsp_ggml_gemm_q8_0_4x8_q8_0_generic wsp_ggml_gemm_q8_0_4x8_q8_0
 #elif defined(__loongarch64)
 // quants.c
 #define wsp_quantize_row_q8_K_generic wsp_quantize_row_q8_K
@@ -101,21 +121,28 @@
 // repack.cpp
 #define wsp_ggml_wsp_quantize_mat_q8_0_4x4_generic wsp_ggml_wsp_quantize_mat_q8_0_4x4
 #define wsp_ggml_wsp_quantize_mat_q8_0_4x8_generic wsp_ggml_wsp_quantize_mat_q8_0_4x8
+#define wsp_ggml_wsp_quantize_mat_q8_K_4x4_generic wsp_ggml_wsp_quantize_mat_q8_K_4x4
 #define wsp_ggml_wsp_quantize_mat_q8_K_4x8_generic wsp_ggml_wsp_quantize_mat_q8_K_4x8
 #define wsp_ggml_gemv_q4_0_4x4_q8_0_generic wsp_ggml_gemv_q4_0_4x4_q8_0
 #define wsp_ggml_gemv_q4_0_4x8_q8_0_generic wsp_ggml_gemv_q4_0_4x8_q8_0
 #define wsp_ggml_gemv_q4_0_8x8_q8_0_generic wsp_ggml_gemv_q4_0_8x8_q8_0
+#define wsp_ggml_gemv_q4_K_8x4_q8_K_generic wsp_ggml_gemv_q4_K_8x4_q8_K
 #define wsp_ggml_gemv_q4_K_8x8_q8_K_generic wsp_ggml_gemv_q4_K_8x8_q8_K
 #define wsp_ggml_gemv_q2_K_8x8_q8_K_generic wsp_ggml_gemv_q2_K_8x8_q8_K
 #define wsp_ggml_gemv_iq4_nl_4x4_q8_0_generic wsp_ggml_gemv_iq4_nl_4x4_q8_0
 #define wsp_ggml_gemv_iq4_nl_8x8_q8_0_generic wsp_ggml_gemv_iq4_nl_8x8_q8_0
+#define wsp_ggml_gemv_q8_0_4x4_q8_0_generic wsp_ggml_gemv_q8_0_4x4_q8_0
+#define wsp_ggml_gemv_q8_0_4x8_q8_0_generic wsp_ggml_gemv_q8_0_4x8_q8_0
 #define wsp_ggml_gemm_q4_0_4x4_q8_0_generic wsp_ggml_gemm_q4_0_4x4_q8_0
 #define wsp_ggml_gemm_q4_0_4x8_q8_0_generic wsp_ggml_gemm_q4_0_4x8_q8_0
 #define wsp_ggml_gemm_q4_0_8x8_q8_0_generic wsp_ggml_gemm_q4_0_8x8_q8_0
+#define wsp_ggml_gemm_q4_K_8x4_q8_K_generic wsp_ggml_gemm_q4_K_8x4_q8_K
 #define wsp_ggml_gemm_q4_K_8x8_q8_K_generic wsp_ggml_gemm_q4_K_8x8_q8_K
 #define wsp_ggml_gemm_q2_K_8x8_q8_K_generic wsp_ggml_gemm_q2_K_8x8_q8_K
 #define wsp_ggml_gemm_iq4_nl_4x4_q8_0_generic wsp_ggml_gemm_iq4_nl_4x4_q8_0
 #define wsp_ggml_gemm_iq4_nl_8x8_q8_0_generic wsp_ggml_gemm_iq4_nl_8x8_q8_0
+#define wsp_ggml_gemm_q8_0_4x4_q8_0_generic wsp_ggml_gemm_q8_0_4x4_q8_0
+#define wsp_ggml_gemm_q8_0_4x8_q8_0_generic wsp_ggml_gemm_q8_0_4x8_q8_0
 #elif defined(__riscv)
 // quants.c
 #define wsp_quantize_row_q8_K_generic wsp_quantize_row_q8_K
@@ -134,19 +161,26 @@
 // repack.cpp
 #define wsp_ggml_wsp_quantize_mat_q8_0_4x4_generic wsp_ggml_wsp_quantize_mat_q8_0_4x4
 #define wsp_ggml_wsp_quantize_mat_q8_0_4x8_generic wsp_ggml_wsp_quantize_mat_q8_0_4x8
+#define wsp_ggml_wsp_quantize_mat_q8_K_4x4_generic wsp_ggml_wsp_quantize_mat_q8_K_4x4
 #define wsp_ggml_wsp_quantize_mat_q8_K_4x8_generic wsp_ggml_wsp_quantize_mat_q8_K_4x8
 #define wsp_ggml_gemv_q4_0_4x4_q8_0_generic wsp_ggml_gemv_q4_0_4x4_q8_0
 #define wsp_ggml_gemv_q4_0_4x8_q8_0_generic wsp_ggml_gemv_q4_0_4x8_q8_0
+#define wsp_ggml_gemv_q4_K_8x4_q8_K_generic wsp_ggml_gemv_q4_K_8x4_q8_K
 #define wsp_ggml_gemv_q4_K_8x8_q8_K_generic wsp_ggml_gemv_q4_K_8x8_q8_K
 #define wsp_ggml_gemv_q2_K_8x8_q8_K_generic wsp_ggml_gemv_q2_K_8x8_q8_K
 #define wsp_ggml_gemv_iq4_nl_4x4_q8_0_generic wsp_ggml_gemv_iq4_nl_4x4_q8_0
 #define wsp_ggml_gemv_iq4_nl_8x8_q8_0_generic wsp_ggml_gemv_iq4_nl_8x8_q8_0
+#define wsp_ggml_gemv_q8_0_4x4_q8_0_generic wsp_ggml_gemv_q8_0_4x4_q8_0
+#define wsp_ggml_gemv_q8_0_4x8_q8_0_generic wsp_ggml_gemv_q8_0_4x8_q8_0
 #define wsp_ggml_gemm_q4_0_4x4_q8_0_generic wsp_ggml_gemm_q4_0_4x4_q8_0
 #define wsp_ggml_gemm_q4_0_4x8_q8_0_generic wsp_ggml_gemm_q4_0_4x8_q8_0
+#define wsp_ggml_gemm_q4_K_8x4_q8_K_generic wsp_ggml_gemm_q4_K_8x4_q8_K
 #define wsp_ggml_gemm_q4_K_8x8_q8_K_generic wsp_ggml_gemm_q4_K_8x8_q8_K
 #define wsp_ggml_gemm_q2_K_8x8_q8_K_generic wsp_ggml_gemm_q2_K_8x8_q8_K
 #define wsp_ggml_gemm_iq4_nl_4x4_q8_0_generic wsp_ggml_gemm_iq4_nl_4x4_q8_0
 #define wsp_ggml_gemm_iq4_nl_8x8_q8_0_generic wsp_ggml_gemm_iq4_nl_8x8_q8_0
+#define wsp_ggml_gemm_q8_0_4x4_q8_0_generic wsp_ggml_gemm_q8_0_4x4_q8_0
+#define wsp_ggml_gemm_q8_0_4x8_q8_0_generic wsp_ggml_gemm_q8_0_4x8_q8_0
 #elif defined(__s390x__)
 // quants.c
 #define wsp_quantize_row_q8_K_generic wsp_quantize_row_q8_K
@@ -163,21 +197,28 @@
 // repack.cpp
 #define wsp_ggml_wsp_quantize_mat_q8_0_4x4_generic wsp_ggml_wsp_quantize_mat_q8_0_4x4
 #define wsp_ggml_wsp_quantize_mat_q8_0_4x8_generic wsp_ggml_wsp_quantize_mat_q8_0_4x8
+#define wsp_ggml_wsp_quantize_mat_q8_K_4x4_generic wsp_ggml_wsp_quantize_mat_q8_K_4x4
 #define wsp_ggml_wsp_quantize_mat_q8_K_4x8_generic wsp_ggml_wsp_quantize_mat_q8_K_4x8
 #define wsp_ggml_gemv_q4_0_4x4_q8_0_generic wsp_ggml_gemv_q4_0_4x4_q8_0
 #define wsp_ggml_gemv_q4_0_4x8_q8_0_generic wsp_ggml_gemv_q4_0_4x8_q8_0
 #define wsp_ggml_gemv_q4_0_8x8_q8_0_generic wsp_ggml_gemv_q4_0_8x8_q8_0
+#define wsp_ggml_gemv_q4_K_8x4_q8_K_generic wsp_ggml_gemv_q4_K_8x4_q8_K
 #define wsp_ggml_gemv_q4_K_8x8_q8_K_generic wsp_ggml_gemv_q4_K_8x8_q8_K
 #define wsp_ggml_gemv_q2_K_8x8_q8_K_generic wsp_ggml_gemv_q2_K_8x8_q8_K
 #define wsp_ggml_gemv_iq4_nl_4x4_q8_0_generic wsp_ggml_gemv_iq4_nl_4x4_q8_0
 #define wsp_ggml_gemv_iq4_nl_8x8_q8_0_generic wsp_ggml_gemv_iq4_nl_8x8_q8_0
+#define wsp_ggml_gemv_q8_0_4x4_q8_0_generic wsp_ggml_gemv_q8_0_4x4_q8_0
+#define wsp_ggml_gemv_q8_0_4x8_q8_0_generic wsp_ggml_gemv_q8_0_4x8_q8_0
 #define wsp_ggml_gemm_q4_0_4x4_q8_0_generic wsp_ggml_gemm_q4_0_4x4_q8_0
 #define wsp_ggml_gemm_q4_0_4x8_q8_0_generic wsp_ggml_gemm_q4_0_4x8_q8_0
 #define wsp_ggml_gemm_q4_0_8x8_q8_0_generic wsp_ggml_gemm_q4_0_8x8_q8_0
+#define wsp_ggml_gemm_q4_K_8x4_q8_K_generic wsp_ggml_gemm_q4_K_8x4_q8_K
 #define wsp_ggml_gemm_q4_K_8x8_q8_K_generic wsp_ggml_gemm_q4_K_8x8_q8_K
 #define wsp_ggml_gemm_q2_K_8x8_q8_K_generic wsp_ggml_gemm_q2_K_8x8_q8_K
 #define wsp_ggml_gemm_iq4_nl_4x4_q8_0_generic wsp_ggml_gemm_iq4_nl_4x4_q8_0
 #define wsp_ggml_gemm_iq4_nl_8x8_q8_0_generic wsp_ggml_gemm_iq4_nl_8x8_q8_0
+#define wsp_ggml_gemm_q8_0_4x4_q8_0_generic wsp_ggml_gemm_q8_0_4x4_q8_0
+#define wsp_ggml_gemm_q8_0_4x8_q8_0_generic wsp_ggml_gemm_q8_0_4x8_q8_0
 #elif defined(__wasm__)
 // quants.c
 #define wsp_ggml_vec_dot_q4_1_q8_1_generic wsp_ggml_vec_dot_q4_1_q8_1
@@ -196,19 +237,26 @@
 // repack.cpp
 #define wsp_ggml_wsp_quantize_mat_q8_0_4x4_generic wsp_ggml_wsp_quantize_mat_q8_0_4x4
 #define wsp_ggml_wsp_quantize_mat_q8_0_4x8_generic wsp_ggml_wsp_quantize_mat_q8_0_4x8
+#define wsp_ggml_wsp_quantize_mat_q8_K_4x4_generic wsp_ggml_wsp_quantize_mat_q8_K_4x4
 #define wsp_ggml_wsp_quantize_mat_q8_K_4x8_generic wsp_ggml_wsp_quantize_mat_q8_K_4x8
 #define wsp_ggml_gemv_q4_0_4x4_q8_0_generic wsp_ggml_gemv_q4_0_4x4_q8_0
 #define wsp_ggml_gemv_q4_0_4x8_q8_0_generic wsp_ggml_gemv_q4_0_4x8_q8_0
 #define wsp_ggml_gemv_q4_0_8x8_q8_0_generic wsp_ggml_gemv_q4_0_8x8_q8_0
+#define wsp_ggml_gemv_q4_K_8x4_q8_K_generic wsp_ggml_gemv_q4_K_8x4_q8_K
 #define wsp_ggml_gemv_q4_K_8x8_q8_K_generic wsp_ggml_gemv_q4_K_8x8_q8_K
 #define wsp_ggml_gemv_q2_K_8x8_q8_K_generic wsp_ggml_gemv_q2_K_8x8_q8_K
 #define wsp_ggml_gemv_iq4_nl_4x4_q8_0_generic wsp_ggml_gemv_iq4_nl_4x4_q8_0
 #define wsp_ggml_gemv_iq4_nl_8x8_q8_0_generic wsp_ggml_gemv_iq4_nl_8x8_q8_0
+#define wsp_ggml_gemv_q8_0_4x4_q8_0_generic wsp_ggml_gemv_q8_0_4x4_q8_0
+#define wsp_ggml_gemv_q8_0_4x8_q8_0_generic wsp_ggml_gemv_q8_0_4x8_q8_0
 #define wsp_ggml_gemm_q4_0_4x4_q8_0_generic wsp_ggml_gemm_q4_0_4x4_q8_0
 #define wsp_ggml_gemm_q4_0_4x8_q8_0_generic wsp_ggml_gemm_q4_0_4x8_q8_0
 #define wsp_ggml_gemm_q4_0_8x8_q8_0_generic wsp_ggml_gemm_q4_0_8x8_q8_0
+#define wsp_ggml_gemm_q4_K_8x4_q8_K_generic wsp_ggml_gemm_q4_K_8x4_q8_K
 #define wsp_ggml_gemm_q4_K_8x8_q8_K_generic wsp_ggml_gemm_q4_K_8x8_q8_K
 #define wsp_ggml_gemm_q2_K_8x8_q8_K_generic wsp_ggml_gemm_q2_K_8x8_q8_K
 #define wsp_ggml_gemm_iq4_nl_4x4_q8_0_generic wsp_ggml_gemm_iq4_nl_4x4_q8_0
 #define wsp_ggml_gemm_iq4_nl_8x8_q8_0_generic wsp_ggml_gemm_iq4_nl_8x8_q8_0
+#define wsp_ggml_gemm_q8_0_4x4_q8_0_generic wsp_ggml_gemm_q8_0_4x4_q8_0
+#define wsp_ggml_gemm_q8_0_4x8_q8_0_generic wsp_ggml_gemm_q8_0_4x8_q8_0
 #endif

package/cpp/ggml-cpu/ggml-cpu-impl.h

@@ -328,7 +328,7 @@ inline static int32x4_t wsp_ggml_vdotq_s32(int32x4_t acc, int8x16_t a, int8x16_t
 
 #if defined(_MSC_VER) || defined(__MINGW32__)
 #include <intrin.h>
-#elif defined(__AVX__) || defined(__AVX2__) || defined(__AVX512F__) || defined(__SSSE3__) || defined(__SSE3__) || defined(__SSE__)
+#elif defined(__SSE__) || defined(__SSE3__) || defined(__SSSE3__) || defined(__AVX__) || defined(__F16C__) || defined(__AVX2__) || defined(__AVX512F__) || defined(__AVX512BF16__)
 #include <immintrin.h>
 #endif
 

package/cpp/ggml-cpu/ggml-cpu.c

@@ -81,6 +81,11 @@ struct wsp_ggml_arm_arch_features_type {
 } wsp_ggml_arm_arch_features = { 0 };
 #endif
 
+#if defined(__riscv)
+struct wsp_ggml_riscv_arch_features_type {
+    int rvv_vlen;
+} wsp_ggml_riscv_arch_features = { 0 };
+#endif
 
 #if defined(_WIN32)
 
@@ -187,6 +192,9 @@ typedef void * thread_ret_t;
 
 typedef pthread_t wsp_ggml_thread_t;
 
+#define WSP_GGML_THREADPOOL_N_THREADS_MASK (0xffffU)
+#define WSP_GGML_THREADPOOL_N_THREADS_BITS (16)
+
 #if defined(__APPLE__)
 #include <unistd.h>
 #include <mach/mach.h>
@@ -449,7 +457,7 @@ struct wsp_ggml_threadpool {
     struct wsp_ggml_cplan  * cplan;
 
     // synchronization primitives
-    atomic_int n_graph;       // incremented when there is work to be done (i.e each graph)
+    atomic_int n_graph;       // updated when there is work to be done (i.e each graph) holds graph and active thread counts.
     atomic_int WSP_GGML_CACHE_ALIGN n_barrier;
     atomic_int WSP_GGML_CACHE_ALIGN n_barrier_passed;
     atomic_int WSP_GGML_CACHE_ALIGN current_chunk; // currently processing chunk during Mat_Mul, shared between all the threads.
@@ -457,12 +465,10 @@ struct wsp_ggml_threadpool {
     // these are atomic as an annotation for thread-sanitizer
     atomic_bool stop;         // Used for stopping the threadpool altogether
     atomic_bool pause;        // Used for pausing the threadpool or individual threads
-    atomic_int abort;         // Used for aborting processing of a graph
+    atomic_int  abort;        // Used for aborting processing of a graph
 
     struct wsp_ggml_compute_state * workers;   // per thread state
-    int          n_threads_max; // number of threads in the pool
-    atomic_int   n_threads_cur; // number of threads used in the current graph
-
+    int          n_threads;   // Number of threads in the pool
     int32_t      prio;        // Scheduling priority
     uint32_t     poll;        // Polling level (0 - no polling)
 
@@ -490,6 +496,15 @@ static inline void wsp_ggml_thread_cpu_relax(void) {
 static inline void wsp_ggml_thread_cpu_relax(void) {
     _mm_pause();
 }
+#elif defined(__riscv)
+static inline void wsp_ggml_thread_cpu_relax(void) {
+    #ifdef __riscv_zihintpause
+        __asm__ __volatile__ ("pause");
+    #else
+        /* Encoding of the pause instruction */
+        __asm__ __volatile__ (".4byte 0x100000F");
+    #endif
+}
 #else
 static inline void wsp_ggml_thread_cpu_relax(void) {;}
 #endif
@@ -530,7 +545,7 @@ struct wsp_ggml_state {
 static struct wsp_ggml_state g_state = {0};
 
 void wsp_ggml_barrier(struct wsp_ggml_threadpool * tp) {
-    int n_threads = atomic_load_explicit(&tp->n_threads_cur, memory_order_relaxed);
+    int n_threads = atomic_load_explicit(&tp->n_graph, memory_order_relaxed) & WSP_GGML_THREADPOOL_N_THREADS_MASK;
     if (n_threads == 1) {
         return;
     }
@@ -547,7 +562,7 @@ void wsp_ggml_barrier(struct wsp_ggml_threadpool * tp) {
         // last thread
         atomic_store_explicit(&tp->n_barrier, 0, memory_order_relaxed);
 
-        // exit barrier (fill seq-cst fence)
+        // exit barrier (full seq-cst fence)
         atomic_fetch_add_explicit(&tp->n_barrier_passed, 1, memory_order_seq_cst);
         return;
     }
@@ -683,24 +698,25 @@ bool wsp_ggml_is_numa(void) {
 }
 
 #if defined(__ARM_ARCH)
-
-#if defined(__linux__) && defined(__aarch64__)
-#include <sys/auxv.h>
-#endif
-
-static void wsp_ggml_init_arm_arch_features(void) {
 #if defined(__aarch64__) && defined(__ARM_FEATURE_SVE)
-#if defined(__linux__)
-    wsp_ggml_arm_arch_features.sve_cnt = PR_SVE_VL_LEN_MASK & prctl(PR_SVE_GET_VL);
+#include <arm_sve.h>
+static void wsp_ggml_init_arm_arch_features(void) {
+    wsp_ggml_arm_arch_features.sve_cnt = svcntb();
+}
 #else
-    // TODO: add support of SVE for non-linux systems
-#error "TODO: SVE is not supported on this platform. To use SVE, sve_cnt needs to be initialized here."
-#endif
+static void wsp_ggml_init_arm_arch_features(void) {}
 #endif
-}
-
 #endif // __ARM_ARCH
 
+#if defined(__riscv) && defined(__riscv_v_intrinsic)
+#include <riscv_vector.h>
+static void wsp_ggml_init_riscv_arch_features(void) {
+    wsp_ggml_riscv_arch_features.rvv_vlen = __riscv_vlenb();
+}
+#else
+static void wsp_ggml_init_riscv_arch_features(void) {}
+#endif
+
 struct wsp_ggml_tensor * wsp_ggml_new_i32(struct wsp_ggml_context * ctx, int32_t value) {
     WSP_GGML_ASSERT(!wsp_ggml_get_no_alloc(ctx));
 
@@ -1927,6 +1943,10 @@ static void wsp_ggml_compute_forward(struct wsp_ggml_compute_params * params, st
             {
                 wsp_ggml_compute_forward_argsort(params, tensor);
             } break;
+        case WSP_GGML_OP_TOP_K:
+            {
+                wsp_ggml_compute_forward_top_k(params, tensor);
+            } break;
         case WSP_GGML_OP_LEAKY_RELU:
             {
                 wsp_ggml_compute_forward_leaky_relu(params, tensor);
@@ -2311,6 +2331,7 @@ static int wsp_ggml_get_n_tasks(struct wsp_ggml_tensor * node, int n_threads) {
         case WSP_GGML_OP_ARANGE:
         case WSP_GGML_OP_TIMESTEP_EMBEDDING:
         case WSP_GGML_OP_ARGSORT:
+        case WSP_GGML_OP_TOP_K:
         case WSP_GGML_OP_FLASH_ATTN_EXT:
         case WSP_GGML_OP_FLASH_ATTN_BACK:
         case WSP_GGML_OP_SSM_CONV:
@@ -2622,7 +2643,7 @@ static void wsp_ggml_thread_cpumask_next(const bool * global_mask, bool * local_
 void wsp_ggml_threadpool_free(struct wsp_ggml_threadpool* threadpool) {
     if (!threadpool) return;
 
-    const int n_threads = threadpool->n_threads_max;
+    const int n_threads = threadpool->n_threads;
 
 #ifndef WSP_GGML_USE_OPENMP
     struct wsp_ggml_compute_state* workers = threadpool->workers;
@@ -2698,9 +2719,14 @@ struct wsp_ggml_cplan wsp_ggml_graph_plan(
         //WSP_GGML_PRINT_DEBUG("Threadpool is not specified. Will create a disposable threadpool : n_threads %d\n", n_threads);
     }
     if (n_threads <= 0) {
-        n_threads = threadpool ? threadpool->n_threads_max : WSP_GGML_DEFAULT_N_THREADS;
+        n_threads = threadpool ? threadpool->n_threads : WSP_GGML_DEFAULT_N_THREADS;
     }
 
+#if defined(__EMSCRIPTEN__) && !defined(__EMSCRIPTEN_PTHREADS__)
+    // Emscripten without pthreads support can only use a single thread
+    n_threads = 1;
+#endif
+
     size_t work_size = 0;
 
     struct wsp_ggml_cplan cplan;
@@ -2834,6 +2860,10 @@ struct wsp_ggml_cplan wsp_ggml_graph_plan(
                         cur += sizeof(wsp_ggml_fp16_t)*ne00*ne01*ne02*ne03;
                         cur += sizeof(wsp_ggml_fp16_t)*ne10*ne11*ne12;
                     } break;
+                case WSP_GGML_OP_TOP_K:
+                    {
+                        cur += sizeof(int32_t)*node->src[0]->ne[0]*n_tasks;
+                    } break;
                 case WSP_GGML_OP_FLASH_ATTN_EXT:
                     {
                         const int64_t ne10 = node->src[1]->ne[0]; // DK
@@ -2897,12 +2927,14 @@ static thread_ret_t wsp_ggml_graph_compute_thread(void * data) {
 
     struct wsp_ggml_compute_params params = {
         /*.ith       =*/ state->ith,
-        /*.nth       =*/ atomic_load_explicit(&tp->n_threads_cur, memory_order_relaxed),
+        /*.nth       =*/ atomic_load_explicit(&tp->n_graph, memory_order_relaxed) & WSP_GGML_THREADPOOL_N_THREADS_MASK,
         /*.wsize     =*/ cplan->work_size,
         /*.wdata     =*/ cplan->work_data,
         /*.threadpool=*/ tp,
     };
 
+    WSP_GGML_PRINT_DEBUG("thread #%d compute-start cplan %p last-graph %d \n", state->ith, cplan, state->last_graph);
+
     for (int node_n = 0; node_n < cgraph->n_nodes && atomic_load_explicit(&tp->abort, memory_order_relaxed) != node_n; node_n++) {
         struct wsp_ggml_tensor * node = cgraph->nodes[node_n];
 
@@ -2924,6 +2956,8 @@ static thread_ret_t wsp_ggml_graph_compute_thread(void * data) {
         }
     }
 
+    WSP_GGML_PRINT_DEBUG("thread #%d compute-done cplan %p last-graph %d \n", state->ith, cplan, state->last_graph);
+
     wsp_ggml_barrier(state->threadpool);
 
     return 0;
@@ -2931,27 +2965,23 @@ static thread_ret_t wsp_ggml_graph_compute_thread(void * data) {
 
 #ifndef WSP_GGML_USE_OPENMP
 
-// check if thread is active
-static inline bool wsp_ggml_graph_compute_thread_active(struct wsp_ggml_compute_state * state) {
-    struct wsp_ggml_threadpool * threadpool = state->threadpool;
-    int n_threads = atomic_load_explicit(&threadpool->n_threads_cur, memory_order_relaxed);
-    return (state->ith < n_threads);
-}
-
 // check if thread is ready to proceed (exit from polling or sleeping)
+// returns true if loops should exit, sets state->pending to indicate new work
 static inline bool wsp_ggml_graph_compute_thread_ready(struct wsp_ggml_compute_state * state) {
     struct wsp_ggml_threadpool * threadpool = state->threadpool;
 
     if (state->pending || threadpool->stop || threadpool->pause) { return true; }
 
     // check for new graph/work
-    int new_graph = atomic_load_explicit(&threadpool->n_graph, memory_order_relaxed);
-    if (new_graph != state->last_graph) {
-        state->pending    = wsp_ggml_graph_compute_thread_active(state);
-        state->last_graph = new_graph;
+    int n_graph   = atomic_load_explicit(&threadpool->n_graph, memory_order_relaxed);
+    int n_threads = n_graph & WSP_GGML_THREADPOOL_N_THREADS_MASK;
+    if (n_graph != state->last_graph) {
+        state->pending    = (state->ith < n_threads);
+        state->last_graph = n_graph;
+        return true;
     }
 
-    return state->pending;
+    return false;
 }
 
 // sync thread state after polling
@@ -2968,11 +2998,6 @@ static inline void wsp_ggml_graph_compute_thread_sync(struct wsp_ggml_compute_st
 static inline bool wsp_ggml_graph_compute_poll_for_work(struct wsp_ggml_compute_state * state) {
     struct wsp_ggml_threadpool * threadpool = state->threadpool;
 
-    // Skip polling for unused threads
-    if (!wsp_ggml_graph_compute_thread_active(state)) {
-        return state->pending;
-    }
-
     // This seems to make 0 ... 100 a decent range for polling level across modern processors.
     // Perhaps, we can adjust it dynamically based on load and things.
     const uint64_t n_rounds = 1024UL * 128 * threadpool->poll;
@@ -3034,7 +3059,6 @@ static thread_ret_t wsp_ggml_graph_compute_secondary_thread(void* data) {
         wsp_ggml_graph_compute_check_for_work(state);
         if (state->pending) {
             state->pending = false;
-
             wsp_ggml_graph_compute_thread(state);
         }
     }
@@ -3049,14 +3073,15 @@ static void wsp_ggml_graph_compute_kickoff(struct wsp_ggml_threadpool * threadpo
 
     wsp_ggml_mutex_lock(&threadpool->mutex);
 
-    WSP_GGML_PRINT_DEBUG("threadpool: n_threads_cur %d n_threads %d\n", threadpool->n_threads_cur, n_threads);
+    // Update the number of active threads and the graph count
+    int n_graph = atomic_load_explicit(&threadpool->n_graph, memory_order_relaxed) >> WSP_GGML_THREADPOOL_N_THREADS_BITS;
+    n_graph = ((n_graph + 1) << WSP_GGML_THREADPOOL_N_THREADS_BITS) | (n_threads & WSP_GGML_THREADPOOL_N_THREADS_MASK);
 
-    // Update the number of active threads
-    atomic_store_explicit(&threadpool->n_threads_cur, n_threads, memory_order_relaxed);
+    WSP_GGML_PRINT_DEBUG("compute-kickoff: n_threads %d n_graph %d\n", n_threads, n_graph);
 
     // Indicate the graph is ready to be processed
     // We need the full seq-cst fence here because of the polling threads (used in thread_sync)
-    atomic_fetch_add_explicit(&threadpool->n_graph, 1, memory_order_seq_cst);
+    atomic_store_explicit(&threadpool->n_graph, n_graph, memory_order_seq_cst);
 
     if (threadpool->pause) {
        // Update main thread prio and affinity to match the threadpool settings
@@ -3094,8 +3119,7 @@ static struct wsp_ggml_threadpool * wsp_ggml_threadpool_new_impl(
         threadpool->pause            = tpp->paused;
         threadpool->abort            = -1;
         threadpool->workers          = NULL;
-        threadpool->n_threads_max    = tpp->n_threads;
-        threadpool->n_threads_cur    = tpp->n_threads;
+        threadpool->n_threads        = tpp->n_threads;
         threadpool->poll             = tpp->poll;
         threadpool->prio             = tpp->prio;
         threadpool->ec               = WSP_GGML_STATUS_SUCCESS;
@@ -3190,7 +3214,7 @@ enum wsp_ggml_status wsp_ggml_graph_compute(struct wsp_ggml_cgraph * cgraph, str
             {
                 // update the number of threads from the actual number of threads that we got from OpenMP
                 n_threads = omp_get_num_threads();
-                atomic_store_explicit(&threadpool->n_threads_cur, n_threads, memory_order_relaxed);
+                atomic_store_explicit(&threadpool->n_graph, n_threads, memory_order_relaxed);
             }
 
             // Apply thread CPU mask and priority
@@ -3203,13 +3227,13 @@ enum wsp_ggml_status wsp_ggml_graph_compute(struct wsp_ggml_cgraph * cgraph, str
             wsp_ggml_graph_compute_thread(&threadpool->workers[ith]);
         }
     } else {
-        atomic_store_explicit(&threadpool->n_threads_cur, 1, memory_order_relaxed);
+        atomic_store_explicit(&threadpool->n_graph, 1, memory_order_relaxed);
         wsp_ggml_graph_compute_thread(&threadpool->workers[0]);
     }
 #else
-    if (n_threads > threadpool->n_threads_max) {
-        WSP_GGML_LOG_WARN("cplan requested more threads (%d) than available (%d)\n", n_threads, threadpool->n_threads_max);
-        n_threads = threadpool->n_threads_max;
+    if (n_threads > threadpool->n_threads) {
+        WSP_GGML_LOG_WARN("cplan requested more threads (%d) than available (%d)\n", n_threads, threadpool->n_threads);
+        n_threads = threadpool->n_threads;
     }
 
     // Kick all threads to start the new graph
@@ -3296,13 +3320,33 @@ void wsp_ggml_cpu_fp16_to_fp32(const wsp_ggml_fp16_t * x, float * y, int64_t n)
         __m128 y_vec = _mm_cvtph_ps(x_vec);
         _mm_storeu_ps(y + i, y_vec);
     }
-#elif defined(__riscv_zvfh)
-    for (int vl; i < n; i += vl) {
-        vl = __riscv_vsetvl_e16m1(n - i);
-        vfloat16m1_t vx = __riscv_vle16_v_f16m1((_Float16 *)&x[i], vl);
-        vfloat32m2_t vy = __riscv_vfwcvt_f_f_v_f32m2(vx, vl);
-        __riscv_vse32_v_f32m2(&y[i], vy, vl);
+
+#elif defined(__riscv_v_intrinsic) && defined(__riscv_zvfhmin)
+    // calculate step size
+    const int epr = __riscv_vsetvlmax_e16m2();
+    const int step = epr * 2;
+    const int np = (n & ~(step - 1));
+
+    // unroll by 2
+    for (; i < np; i += step) {
+        vfloat16m2_t ax0 = __riscv_vle16_v_f16m2((const _Float16*)x + i, epr);
+        vfloat32m4_t ay0 = __riscv_vfwcvt_f_f_v_f32m4(ax0, epr);
+        __riscv_vse32_v_f32m4(y + i, ay0, epr);
+
+        vfloat16m2_t ax1 = __riscv_vle16_v_f16m2((const _Float16*)x + i + epr, epr);
+        vfloat32m4_t ay1 = __riscv_vfwcvt_f_f_v_f32m4(ax1, epr);
+        __riscv_vse32_v_f32m4(y + i + epr, ay1, epr);
     }
+
+    // leftovers
+    int vl;
+    for (i = np; i < n; i += vl) {
+        vl = __riscv_vsetvl_e16m2(n - i);
+        vfloat16m2_t ax0 = __riscv_vle16_v_f16m2((const _Float16*)x + i, vl);
+        vfloat32m4_t ay0 = __riscv_vfwcvt_f_f_v_f32m4(ax0, vl);
+        __riscv_vse32_v_f32m4(y + i, ay0, vl);
+    }
+
 #endif
 
     for (; i < n; ++i) {
@@ -3347,6 +3391,31 @@ void wsp_ggml_cpu_bf16_to_fp32(const wsp_ggml_bf16_t * x, float * y, int64_t n)
                                         (const __m128i *)(x + i))),
                                 16)));
     }
+#elif defined(__riscv_v_intrinsic) && defined(__riscv_zvfbfmin)
+    // calculate step size
+    const int epr = __riscv_vsetvlmax_e16m2();
+    const int step = epr * 2;
+    const int np = (n & ~(step - 1));
+
+    // unroll by 2
+    for (; i < np; i += step) {
+        vbfloat16m2_t ax0 = __riscv_vle16_v_bf16m2((const __bf16*)x + i, epr);
+        vfloat32m4_t ay0 = __riscv_vfwcvtbf16_f_f_v_f32m4(ax0, epr);
+        __riscv_vse32_v_f32m4(y + i, ay0, epr);
+
+        vbfloat16m2_t ax1 = __riscv_vle16_v_bf16m2((const __bf16*)x + i + epr, epr);
+        vfloat32m4_t ay1 = __riscv_vfwcvtbf16_f_f_v_f32m4(ax1, epr);
+        __riscv_vse32_v_f32m4(y + i + epr, ay1, epr);
+    }
+
+    // leftovers
+    int vl;
+    for (i = np; i < n; i += vl) {
+        vl = __riscv_vsetvl_e16m2(n - i);
+        vbfloat16m2_t ax0 = __riscv_vle16_v_bf16m2((const __bf16*)x + i, vl);
+        vfloat32m4_t ay0 = __riscv_vfwcvtbf16_f_f_v_f32m4(ax0, vl);
+        __riscv_vse32_v_f32m4(y + i, ay0, vl);
+    }
 #endif
     for (; i < n; i++) {
         y[i] = WSP_GGML_BF16_TO_FP32(x[i]);
@@ -3449,6 +3518,14 @@ int wsp_ggml_cpu_has_riscv_v(void) {
 #endif
 }
 
+int wsp_ggml_cpu_get_rvv_vlen(void) {
+#if defined(__riscv) && defined(__riscv_v_intrinsic)
+    return wsp_ggml_riscv_arch_features.rvv_vlen;
+#else
+    return 0;
+#endif
+}
+
 int wsp_ggml_cpu_has_f16c(void) {
 #if defined(__F16C__)
     return 1;
@@ -3615,6 +3692,10 @@ void wsp_ggml_cpu_init(void) {
         wsp_ggml_init_arm_arch_features();
 #endif
 
+#if defined(__riscv)
+        wsp_ggml_init_riscv_arch_features();
+#endif
+
         is_first_call = false;
     }