vdotq_s32 -> ggml_vdotq_s32

Signed-off-by: Alberto Cabrera <[email protected]>

Author: Alcpz

ggml/src/ggml-cpu/arch/arm/repack.cpp

@@ -525,7 +525,7 @@ void ggml_gemv_q4_K_8x8_q8_K(int                        n,
     UNUSED(ncols_interleaved);
     UNUSED(blocklen);
 
-#if ! ((defined(_MSC_VER)) && ! defined(__clang__)) && defined(__aarch64__) && defined(__ARM_NEON) && defined(__ARM_FEATURE_DOTPROD)
+#if ! ((defined(_MSC_VER)) && ! defined(__clang__)) && defined(__aarch64__) && defined(__ARM_NEON)
     constexpr int col_pairs = ncols_interleaved / 2;
     const uint8x16_t m4b = vdupq_n_u8(0x0f);
 
@@ -596,15 +596,15 @@ void ggml_gemv_q4_K_8x8_q8_K(int                        n,
                     uint8x16_t q4_qs_cp_2 = vld1q_u8(q4_base + 16 * cp + 128);
                     uint8x16_t q4_qs_cp_3 = vld1q_u8(q4_base + 16 * cp + 192);
 
-                    acc_lo[cp] = vdotq_s32(acc_lo[cp], vreinterpretq_s8_u8(vandq_u8(q4_qs_cp_0, m4b)), q8_qs[0]); // 0 .. 7
-                    acc_lo[cp] = vdotq_s32(acc_lo[cp], vreinterpretq_s8_u8(vandq_u8(q4_qs_cp_1, m4b)), q8_qs[1]); // 8 ..15
-                    acc_lo[cp] = vdotq_s32(acc_lo[cp], vreinterpretq_s8_u8(vandq_u8(q4_qs_cp_2, m4b)), q8_qs[2]); // 16..23
-                    acc_lo[cp] = vdotq_s32(acc_lo[cp], vreinterpretq_s8_u8(vandq_u8(q4_qs_cp_3, m4b)), q8_qs[3]); // 24..31
+                    acc_lo[cp] = ggml_vdotq_s32(acc_lo[cp], vreinterpretq_s8_u8(vandq_u8(q4_qs_cp_0, m4b)), q8_qs[0]); // 0 .. 7
+                    acc_lo[cp] = ggml_vdotq_s32(acc_lo[cp], vreinterpretq_s8_u8(vandq_u8(q4_qs_cp_1, m4b)), q8_qs[1]); // 8 ..15
+                    acc_lo[cp] = ggml_vdotq_s32(acc_lo[cp], vreinterpretq_s8_u8(vandq_u8(q4_qs_cp_2, m4b)), q8_qs[2]); // 16..23
+                    acc_lo[cp] = ggml_vdotq_s32(acc_lo[cp], vreinterpretq_s8_u8(vandq_u8(q4_qs_cp_3, m4b)), q8_qs[3]); // 24..31
 
-                    acc_hi[cp] = vdotq_s32(acc_hi[cp], vreinterpretq_s8_u8(vshrq_n_u8(q4_qs_cp_0, 4)), q8_qs[4]); // 32..39
-                    acc_hi[cp] = vdotq_s32(acc_hi[cp], vreinterpretq_s8_u8(vshrq_n_u8(q4_qs_cp_1, 4)), q8_qs[5]); // 40..47
-                    acc_hi[cp] = vdotq_s32(acc_hi[cp], vreinterpretq_s8_u8(vshrq_n_u8(q4_qs_cp_2, 4)), q8_qs[6]); // 48..55
-                    acc_hi[cp] = vdotq_s32(acc_hi[cp], vreinterpretq_s8_u8(vshrq_n_u8(q4_qs_cp_3, 4)), q8_qs[7]); // 56..63
+                    acc_hi[cp] = ggml_vdotq_s32(acc_hi[cp], vreinterpretq_s8_u8(vshrq_n_u8(q4_qs_cp_0, 4)), q8_qs[4]); // 32..39
+                    acc_hi[cp] = ggml_vdotq_s32(acc_hi[cp], vreinterpretq_s8_u8(vshrq_n_u8(q4_qs_cp_1, 4)), q8_qs[5]); // 40..47
+                    acc_hi[cp] = ggml_vdotq_s32(acc_hi[cp], vreinterpretq_s8_u8(vshrq_n_u8(q4_qs_cp_2, 4)), q8_qs[6]); // 48..55
+                    acc_hi[cp] = ggml_vdotq_s32(acc_hi[cp], vreinterpretq_s8_u8(vshrq_n_u8(q4_qs_cp_3, 4)), q8_qs[7]); // 56..63
                 }
 
 
@@ -652,7 +652,7 @@ void ggml_gemv_q4_K_8x8_q8_K(int                        n,
         vst1q_f32(s + base + 4, acc_f32[1]);
     }  // for x
     return;
-#endif
+#endif // #if ! ((defined(_MSC_VER)) && ! defined(__clang__)) && defined(__aarch64__) && defined(__ARM_NEON)
     ggml_gemv_q4_K_8x8_q8_K_generic(n, s, bs, vx, vy, nr, nc);
 }
 

ggml/src/ggml-cpu/repack.cpp

@@ -1847,7 +1847,7 @@ static const ggml::cpu::tensor_traits * ggml_repack_get_optimal_repack_type(cons
                 return &q4_K_8x8_q8_K;
             }
         }
-        if (ggml_cpu_has_neon() && ggml_cpu_has_matmul_int8() && ggml_cpu_has_dotprod()) {
+        if (ggml_cpu_has_neon() && ggml_cpu_has_matmul_int8()) {
             if (cur->ne[1] % 8 == 0) {
                 return &q4_K_8x8_q8_K;
             }