whispercpp 1.3.6 → 1.3.8

data/ext/sources/ggml/src/ggml-cpu/arch/riscv/repack.cpp

@@ -107,8 +107,7 @@ void ggml_quantize_mat_q8_0_4x8(const float * GGML_RESTRICT x, void * GGML_RESTR
     }
 #else
     UNUSED(nb);
-    UNUSED(y);
-    ggml_quantize_mat_q8_0_4x4_generic(x, vy, k);
+    ggml_quantize_mat_q8_0_4x8_generic(x, vy, k);
 #endif
 }
 
@@ -203,6 +202,7 @@ void ggml_gemv_q4_0_8x8_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const vo
     ggml_gemv_q4_0_8x8_q8_0_generic(n, s, bs, vx, vy, nr, nc);
 }
 
+#if defined __riscv_zvfh
 void ggml_gemv_q4_0_16x1_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc) {
     const int qk = QK8_0;
     const int nb = n / qk;
@@ -222,7 +222,6 @@ void ggml_gemv_q4_0_16x1_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const v
     UNUSED(ncols_interleaved);
     UNUSED(blocklen);
 
-#if defined __riscv_v_intrinsic
     const block_q8_0 * a_ptr = (const block_q8_0 *) vy;
     for (int x = 0; x < nc / ncols_interleaved; x++) {
         const block_q4_0x16 * b_ptr = (const block_q4_0x16 *) vx + (x * nb);
@@ -256,9 +255,6 @@ void ggml_gemv_q4_0_16x1_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const v
 
         __riscv_vse32_v_f32m2(s + x * 16, sumf, 16);
     }
-    return;
-#endif
-    ggml_gemv_q4_0_16x1_q8_0_generic(n, s, bs, vx, vy, nr, nc);
 }
 
 void ggml_gemv_q4_K_16x1_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc) {
@@ -280,7 +276,6 @@ void ggml_gemv_q4_K_16x1_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const v
     UNUSED(ncols_interleaved);
     UNUSED(blocklen);
 
-#if defined __riscv_v_intrinsic
     const block_q8_K * a_ptr = (const block_q8_K *) vy;
 
     for (int x = 0; x < nc / ncols_interleaved; x++) {
@@ -392,9 +387,6 @@ void ggml_gemv_q4_K_16x1_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const v
 
         __riscv_vse32_v_f32m2(s + x * 16, sumf, 16);
     }
-    return;
-#endif
-    ggml_gemv_q4_K_16x1_q8_K_generic(n, s, bs, vx, vy, nr, nc);
 }
 
 void ggml_gemv_iq4_nl_16x1_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc) {
@@ -416,7 +408,6 @@ void ggml_gemv_iq4_nl_16x1_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const
     UNUSED(ncols_interleaved);
     UNUSED(blocklen);
 
-#if defined __riscv_v_intrinsic
     const vint8mf2_t values = __riscv_vle8_v_i8mf2(kvalues_iq4nl, 16);
     const block_q8_0 * a_ptr = (const block_q8_0 *) vy;
     for (int x = 0; x < nc / ncols_interleaved; x++) {
@@ -451,9 +442,6 @@ void ggml_gemv_iq4_nl_16x1_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const
 
         __riscv_vse32_v_f32m2(s + x * 16, sumf, 16);
     }
-    return;
-#endif
-    ggml_gemv_iq4_nl_16x1_q8_0_generic(n, s, bs, vx, vy, nr, nc);
 }
 
 void ggml_gemv_q8_0_16x1_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc) {
@@ -476,7 +464,6 @@ void ggml_gemv_q8_0_16x1_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const v
     UNUSED(blocklen);
     UNUSED(bs);
 
-#if defined __riscv_v_intrinsic
     const block_q8_0 * a_ptr = (const block_q8_0 *) vy;
     for (int x = 0; x < nc / ncols_interleaved; x++) {
         const block_q8_0x16 * b_ptr = (const block_q8_0x16 *) vx + (x * nb);
@@ -505,9 +492,6 @@ void ggml_gemv_q8_0_16x1_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const v
 
         __riscv_vse32_v_f32m2(s + x * 16, sumf, 16);
     }
-    return;
-#endif
-    ggml_gemv_q8_0_16x1_q8_0_generic(n, s, bs, vx, vy, nr, nc);
 }
 
 void ggml_gemv_q2_K_16x1_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc) {
@@ -679,9 +663,9 @@ void ggml_gemv_q2_K_16x1_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const v
 
         } // End K-Block
         __riscv_vse32_v_f32m2(s + col_tile, v_sumf, vl);
-
     }
 }
+#endif
 
 void ggml_gemm_q4_0_8x8_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc) {
     const int qk = QK8_0;
@@ -909,6 +893,7 @@ void ggml_gemm_q4_0_8x8_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const vo
     ggml_gemm_q4_0_8x8_q8_0_generic(n, s, bs, vx, vy, nr, nc);
 }
 
+#if defined __riscv_zvfh
 void ggml_gemm_q4_0_16x1_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc) {
     const int qk = QK8_0;
     const int nb = n / qk;
@@ -929,7 +914,6 @@ void ggml_gemm_q4_0_16x1_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const v
     UNUSED(ncols_interleaved);
     UNUSED(blocklen);
 
-#if defined __riscv_v_intrinsic
     for (int y = 0; y < nr / 4; y++) {
         const block_q8_0x4 * a_ptr = (const block_q8_0x4 *) vy + (y * nb);
         for (int x = 0; x < nc / ncols_interleaved; x++) {
@@ -994,9 +978,6 @@ void ggml_gemm_q4_0_16x1_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const v
             __riscv_vse32_v_f32m2(s + (y * 4 + 3) * bs + x * 16, sumf_3, 16);
         }
     }
-    return;
-#endif
-    ggml_gemm_q4_0_16x1_q8_0_generic(n, s, bs, vx, vy, nr, nc);
 }
 
 void ggml_gemm_q4_K_16x1_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc) {
@@ -1019,7 +1000,6 @@ void ggml_gemm_q4_K_16x1_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const v
     UNUSED(ncols_interleaved);
     UNUSED(blocklen);
 
-#if defined __riscv_v_intrinsic
     for (int y = 0; y < nr / 4; y++) {
         const block_q8_Kx4 * a_ptr = (const block_q8_Kx4 *) vy + (y * nb);
         for (int x = 0; x < nc / ncols_interleaved; x++) {
@@ -1267,9 +1247,6 @@ void ggml_gemm_q4_K_16x1_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const v
             __riscv_vse32_v_f32m2(s + (y * 4 + 3) * bs + x * 16, sumf_3, 16);
         }
     }
-    return;
-#endif
-    ggml_gemm_q4_K_16x1_q8_K_generic(n, s, bs, vx, vy, nr, nc);
 }
 
 void ggml_gemm_iq4_nl_16x1_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc) {
@@ -1292,7 +1269,6 @@ void ggml_gemm_iq4_nl_16x1_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const
     UNUSED(ncols_interleaved);
     UNUSED(blocklen);
 
-#if defined __riscv_v_intrinsic
     const vint8mf2_t values = __riscv_vle8_v_i8mf2(kvalues_iq4nl, 16);
     for (int y = 0; y < nr / 4; y++) {
         const block_q8_0x4 * a_ptr = (const block_q8_0x4 *) vy + (y * nb);
@@ -1355,9 +1331,6 @@ void ggml_gemm_iq4_nl_16x1_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const
             __riscv_vse32_v_f32m2(s + (y * 4 + 3) * bs + x * 16, sumf_3, 16);
         }
     }
-    return;
-#endif
-    ggml_gemm_iq4_nl_16x1_q8_0_generic(n, s, bs, vx, vy, nr, nc);
 }
 
 void ggml_gemm_q8_0_16x1_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc) {
@@ -1380,7 +1353,6 @@ void ggml_gemm_q8_0_16x1_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const v
     UNUSED(ncols_interleaved);
     UNUSED(blocklen);
 
-#if defined __riscv_v_intrinsic
     for (int y = 0; y < nr / 4; y++) {
         const block_q8_0x4 * a_ptr = (const block_q8_0x4 *) vy + (y * nb);
         for (int x = 0; x < nc / ncols_interleaved; x++) {
@@ -1429,9 +1401,6 @@ void ggml_gemm_q8_0_16x1_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const v
             __riscv_vse32_v_f32m2(s + (y * 4 + 3) * bs + x * 16, sumf_3, 16);
         }
     }
-    return;
-#endif
-    ggml_gemm_q8_0_16x1_q8_0_generic(n, s, bs, vx, vy, nr, nc);
 }
 
 void ggml_gemm_q2_K_16x1_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, const void * GGML_RESTRICT vy, int nr, int nc) {
@@ -1731,3 +1700,4 @@ void ggml_gemm_q2_K_16x1_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const v
         }
     }
 }
+#endif

data/ext/sources/ggml/src/ggml-cpu/arch/s390/quants.c

@@ -1463,4 +1463,3 @@ void ggml_vec_dot_iq4_xs_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const v
     ggml_vec_dot_iq4_xs_q8_K_generic(n, s, bs, vx, bx, vy, by, nrc);
 #endif
 }
-

data/ext/sources/ggml/src/ggml-cpu/arch/wasm/quants.c

@@ -355,6 +355,78 @@ void ggml_vec_dot_q4_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const voi
     *s = sumf;
 }
 
+void ggml_vec_dot_q4_1_q8_1(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc) {
+    const int qk = QK8_1;
+    const int nb = n / qk;
+
+    assert(n % qk == 0);
+    assert(nrc == 1);
+    UNUSED(nrc);
+    UNUSED(bx);
+    UNUSED(by);
+    UNUSED(bs);
+
+    const block_q4_1 * GGML_RESTRICT x = vx;
+    const block_q8_1 * GGML_RESTRICT y = vy;
+
+    float sumf = 0;
+
+#if defined __wasm_simd128__
+    v128_t sumv = wasm_f32x4_splat(0.0f);
+    float summs = 0.0f;
+
+    for (int ib = 0; ib < nb; ++ib) {
+        const block_q4_1 * GGML_RESTRICT x0 = &x[ib];
+        const block_q8_1 * GGML_RESTRICT y0 = &y[ib];
+
+        summs += GGML_CPU_FP16_TO_FP32(x0->m) * GGML_CPU_FP16_TO_FP32(y0->s);
+
+        const v128_t raw  = wasm_v128_load(x0->qs);
+        const v128_t v0s  = wasm_v128_and(raw, wasm_i8x16_splat(0x0F));
+        const v128_t v1s  = wasm_u8x16_shr(raw, 4);
+
+        const v128_t ys_lo = wasm_v128_load(y0->qs);
+        const v128_t ys_hi = wasm_v128_load(y0->qs + 16);
+
+        const v128_t v0s_l = wasm_u16x8_extend_low_u8x16(v0s);
+        const v128_t v0s_h = wasm_u16x8_extend_high_u8x16(v0s);
+        const v128_t ylo_l = wasm_i16x8_extend_low_i8x16(ys_lo);
+        const v128_t ylo_h = wasm_i16x8_extend_high_i8x16(ys_lo);
+        const v128_t v1s_l = wasm_u16x8_extend_low_u8x16(v1s);
+        const v128_t v1s_h = wasm_u16x8_extend_high_u8x16(v1s);
+        const v128_t yhi_l = wasm_i16x8_extend_low_i8x16(ys_hi);
+        const v128_t yhi_h = wasm_i16x8_extend_high_i8x16(ys_hi);
+
+        const v128_t acc = wasm_i32x4_add(
+            wasm_i32x4_add(
+                wasm_i32x4_dot_i16x8(v0s_l, ylo_l),
+                wasm_i32x4_dot_i16x8(v0s_h, ylo_h)),
+            wasm_i32x4_add(
+                wasm_i32x4_dot_i16x8(v1s_l, yhi_l),
+                wasm_i32x4_dot_i16x8(v1s_h, yhi_h)));
+
+        sumv = wasm_f32x4_add(sumv,
+            wasm_f32x4_mul(
+                wasm_f32x4_convert_i32x4(acc),
+                wasm_f32x4_splat(GGML_CPU_FP16_TO_FP32(x0->d) * GGML_CPU_FP16_TO_FP32(y0->d))));
+    }
+
+    sumf = wasm_f32x4_extract_lane(sumv, 0) + wasm_f32x4_extract_lane(sumv, 1) +
+           wasm_f32x4_extract_lane(sumv, 2) + wasm_f32x4_extract_lane(sumv, 3) + summs;
+
+    *s = sumf;
+
+#else
+    UNUSED(nb);
+    UNUSED(x);
+    UNUSED(y);
+    UNUSED(sumf);
+
+    ggml_vec_dot_q4_1_q8_1_generic(
+        n, s, bs, vx, bx, vy, by, nrc);
+#endif
+}
+
 void ggml_vec_dot_q5_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc) {
     const int qk = QK8_0;
     const int nb = n / qk;
@@ -1218,4 +1290,3 @@ void ggml_vec_dot_q6_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const voi
     ggml_vec_dot_q6_K_q8_K_generic(n, s, bs, vx, bx, vy, by, nrc);
 #endif
 }
-