ggml : AVX512 gemm for Q4_0_8_8 (llama/9532)

* AVX512 version of ggml_gemm_q4_0_8x8_q8_0

* Remove zero vector parameter passing

* Rename functions and rearrange order of macros

* Edit commments

* style : minor adjustments

* Update x to start from 0

---------

Co-authored-by: Georgi Gerganov <[email protected]>

ggml/src/ggml-aarch64.c

@@ -39,11 +39,44 @@
 //
 #if defined(__AVX__)
 #if defined(__F16C__)
+#if defined(__AVX512F__)
+#define GGML_F32Cx8x2_LOAD(x, y)     _mm512_cvtph_ps(_mm256_set_m128i(_mm_loadu_si128((const __m128i *)(y)), _mm_loadu_si128((const __m128i *)(x))))
+#define GGML_F32Cx16_REPEAT_LOAD(x)  _mm512_cvtph_ps(_mm256_set_m128i(x, x))
+#endif
 // the  _mm256_cvt intrinsics require F16C
 #define GGML_F32Cx8_LOAD(x)     _mm256_cvtph_ps(_mm_loadu_si128((const __m128i *)(x)))
 #define GGML_F32Cx8_REPEAT_LOAD(x, loadMask)     _mm256_cvtph_ps(_mm_shuffle_epi32(_mm_maskload_epi32((int const*)(x), loadMask), 68))
 #define GGML_F32Cx8_REARRANGE_LOAD(x, arrangeMask)     _mm256_cvtph_ps(_mm_shuffle_epi8(_mm_loadu_si128((const __m128i *) x), arrangeMask))
 #else
+#if defined(__AVX512F__)
+static inline __m512 __avx512_f32cx8x2_load(ggml_fp16_t *x, ggml_fp16_t *y) {
+    float tmp[16];
+
+    for (int i = 0; i < 8; i++) {
+        tmp[i] = GGML_FP16_TO_FP32(x[i]);
+    }
+
+    for (int i = 0; i < 8; i++) {
+        tmp[i + 8] = GGML_FP16_TO_FP32(y[i]);
+    }
+
+    return _mm512_loadu_ps(tmp);
+}
+static inline __m512 __avx512_repeat_f32cx16_load(__m128i x) {
+    float tmp[16];
+    uint16_t tmphalf[8];
+    _mm_storeu_si128((__m128i*)tmphalf, x);
+
+    for (int i = 0; i < 4; i++) {
+        tmp[i] = GGML_FP16_TO_FP32(tmphalf[i]);
+        tmp[i + 4] = GGML_FP16_TO_FP32(tmphalf[i]);
+        tmp[i + 8] = GGML_FP16_TO_FP32(tmphalf[i]);
+        tmp[i + 12] = GGML_FP16_TO_FP32(tmphalf[i]);
+    }
+
+    return _mm512_loadu_ps(tmp);
+}
+#endif
 static inline __m256 __avx_f32cx8_load(ggml_fp16_t *x) {
     float tmp[8];
 
@@ -78,30 +111,65 @@ static inline __m256 __avx_rearranged_f32cx8_load(ggml_fp16_t *x, __m128i arrang
 #define GGML_F32Cx8_LOAD(x)     __avx_f32cx8_load(x)
 #define GGML_F32Cx8_REPEAT_LOAD(x, loadMask)     __avx_repeat_f32cx8_load(x)
 #define GGML_F32Cx8_REARRANGE_LOAD(x, arrangeMask)     __avx_rearranged_f32cx8_load(x, arrangeMask)
+#if defined(__AVX512F__)
+#define GGML_F32Cx8x2_LOAD(x, y)     __avx512_f32cx8x2_load(x, y)
+#define GGML_F32Cx16_REPEAT_LOAD(x)  __avx512_repeat_f32cx16_load(x)
+#endif
 #endif
 #endif
 
 
 #if defined(__AVX2__) || defined(__AVX512F__)
-static inline __m256i sum_i16_pairs_int(const __m256i x) {
+#if defined(__AVX512F__)
+// add int16_t pairwise and return as 512 bit int vector
+static inline __m512i sum_i16_pairs_int_32x16(const __m512i x) {
+    const __m512i ones = _mm512_set1_epi16(1);
+    return _mm512_madd_epi16(ones, x);
+}
+
+static inline __m512i mul_sum_us8_pairs_int32x16(const __m512i ax, const __m512i sy) {
+#if defined(__AVXVNNI__) || (defined(__AVX512VNNI__) && defined(__AVX512VL__))
+    const __m512i zero = _mm512_setzero_si512();
+    return _mm512_dpbusd_epi32(zero, ax, sy);
+#else
+    // Perform multiplication and create 16-bit values
+    const __m512i dot = _mm512_maddubs_epi16(ax, sy);
+    return sum_i16_pairs_int_32x16(dot);
+#endif
+}
+
+// multiply int8_t, add results pairwise twice and return as 512 bit int vector
+static inline __m512i mul_sum_i8_pairs_int32x16(const __m512i x, const __m512i y) {
+    const __m512i zero = _mm512_setzero_si512();
+    // Get absolute values of x vectors
+    const __m512i ax = _mm512_abs_epi8(x);
+    // Sign the values of the y vectors
+    __mmask64 blt0 = _mm512_movepi8_mask(x);
+    const __m512i sy = _mm512_mask_sub_epi8(y, blt0, zero, y);
+    return mul_sum_us8_pairs_int32x16(ax, sy);
+}
+#endif
+
+// add int16_t pairwise and return as 256 bit int vector
+static inline __m256i sum_i16_pairs_int32x8(const __m256i x) {
     const __m256i ones = _mm256_set1_epi16(1);
     return _mm256_madd_epi16(ones, x);
 }
 
-static inline __m256i mul_sum_us8_pairs_int(const __m256i ax, const __m256i sy) {
+static inline __m256i mul_sum_us8_pairs_int32x8(const __m256i ax, const __m256i sy) {
 #if defined(__AVXVNNI__) || (defined(__AVX512VNNI__) && defined(__AVX512VL__))
     const __m256i zero = _mm256_setzero_si256();
     return _mm256_dpbusd_epi32(zero, ax, sy);
 #else
     // Perform multiplication and create 16-bit values
     const __m256i dot = _mm256_maddubs_epi16(ax, sy);
-    return sum_i16_pairs_int(dot);
+    return sum_i16_pairs_int32x8(dot);
 #endif
 }
 
 // Integer variant of the function defined in ggml-quants.c
-// multiply int8_t, add results pairwise twice and return as float vector
-static inline __m256i mul_sum_i8_pairs_int(const __m256i x, const __m256i y) {
+// multiply int8_t, add results pairwise twice and return as 256 bit int vector
+static inline __m256i mul_sum_i8_pairs_int32x8(const __m256i x, const __m256i y) {
 #if __AVXVNNIINT8__
     const __m256i zero = _mm256_setzero_si256();
     return _mm256_dpbssd_epi32(zero, x, y);
@@ -110,7 +178,7 @@ static inline __m256i mul_sum_i8_pairs_int(const __m256i x, const __m256i y) {
     const __m256i ax = _mm256_sign_epi8(x, x);
     // Sign the values of the y vectors
     const __m256i sy = _mm256_sign_epi8(y, x);
-    return mul_sum_us8_pairs_int(ax, sy);
+    return mul_sum_us8_pairs_int32x8(ax, sy);
 #endif
 }
 #endif
@@ -929,17 +997,17 @@ void ggml_gemv_q4_0_8x8_q8_0(int n, float * restrict s, size_t bs, const void *
                 // ...........................................................................
                 // B0(28-31) B4(28-31) B1(28-31) B5(28-31) B2(28-31) B6(28-31) B3(28-31) B7(28-31) with A0(28-31)
 
-                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int(_mm256_blend_epi32(rhs_vec_0123_0 ,_mm256_shuffle_epi32(rhs_vec_4567_0, 177), 170), _mm256_shuffle_epi32(lhs_vec_0, 0)));
-                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int(_mm256_blend_epi32(_mm256_shuffle_epi32(rhs_vec_0123_0, 177) ,rhs_vec_4567_0, 170), _mm256_shuffle_epi32(lhs_vec_0, 85)));
+                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int32x8(_mm256_blend_epi32(rhs_vec_0123_0 ,_mm256_shuffle_epi32(rhs_vec_4567_0, 177), 170), _mm256_shuffle_epi32(lhs_vec_0, 0)));
+                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int32x8(_mm256_blend_epi32(_mm256_shuffle_epi32(rhs_vec_0123_0, 177) ,rhs_vec_4567_0, 170), _mm256_shuffle_epi32(lhs_vec_0, 85)));
 
-                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int(_mm256_blend_epi32(rhs_vec_0123_1 ,_mm256_shuffle_epi32(rhs_vec_4567_1, 177), 170), _mm256_shuffle_epi32(lhs_vec_0, 170)));
-                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int(_mm256_blend_epi32(_mm256_shuffle_epi32(rhs_vec_0123_1, 177) ,rhs_vec_4567_1, 170), _mm256_shuffle_epi32(lhs_vec_0, 255)));
+                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int32x8(_mm256_blend_epi32(rhs_vec_0123_1 ,_mm256_shuffle_epi32(rhs_vec_4567_1, 177), 170), _mm256_shuffle_epi32(lhs_vec_0, 170)));
+                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int32x8(_mm256_blend_epi32(_mm256_shuffle_epi32(rhs_vec_0123_1, 177) ,rhs_vec_4567_1, 170), _mm256_shuffle_epi32(lhs_vec_0, 255)));
 
-                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int(_mm256_blend_epi32(rhs_vec_0123_2 ,_mm256_shuffle_epi32(rhs_vec_4567_2, 177), 170), _mm256_shuffle_epi32(lhs_vec_1, 0)));
-                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int(_mm256_blend_epi32(_mm256_shuffle_epi32(rhs_vec_0123_2, 177) ,rhs_vec_4567_2, 170), _mm256_shuffle_epi32(lhs_vec_1, 85)));
+                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int32x8(_mm256_blend_epi32(rhs_vec_0123_2 ,_mm256_shuffle_epi32(rhs_vec_4567_2, 177), 170), _mm256_shuffle_epi32(lhs_vec_1, 0)));
+                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int32x8(_mm256_blend_epi32(_mm256_shuffle_epi32(rhs_vec_0123_2, 177) ,rhs_vec_4567_2, 170), _mm256_shuffle_epi32(lhs_vec_1, 85)));
 
-                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int(_mm256_blend_epi32(rhs_vec_0123_3 ,_mm256_shuffle_epi32(rhs_vec_4567_3, 177), 170), _mm256_shuffle_epi32(lhs_vec_1, 170)));
-                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int(_mm256_blend_epi32(_mm256_shuffle_epi32(rhs_vec_0123_3, 177) ,rhs_vec_4567_3, 170), _mm256_shuffle_epi32(lhs_vec_1, 255)));
+                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int32x8(_mm256_blend_epi32(rhs_vec_0123_3 ,_mm256_shuffle_epi32(rhs_vec_4567_3, 177), 170), _mm256_shuffle_epi32(lhs_vec_1, 170)));
+                iacc = _mm256_add_epi32(iacc, mul_sum_i8_pairs_int32x8(_mm256_blend_epi32(_mm256_shuffle_epi32(rhs_vec_0123_3, 177) ,rhs_vec_4567_3, 170), _mm256_shuffle_epi32(lhs_vec_1, 255)));
 
                 // Accumulated values multipled with appropriate scales
                 acc_row = _mm256_fmadd_ps(_mm256_cvtepi32_ps(iacc), _mm256_mul_ps(col_scale_f32, row_scale_f32), acc_row);
@@ -2421,10 +2489,411 @@ void ggml_gemm_q4_0_8x8_q8_0(int n, float * restrict s, size_t bs, const void *
     __m256i signextendlut = _mm256_castsi128_si256(_mm_set_epi8(-1, -2, -3, -4, -5, -6, -7, -8, 7, 6, 5, 4, 3, 2, 1, 0));
     signextendlut = _mm256_permute2f128_si256(signextendlut, signextendlut, 0);
     // Permute mask used for easier vector processing at later stages
-    __m256i requiredOrder = _mm256_set_epi32(3 ,2 ,1 ,0, 7 ,6, 5, 4);
+    __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__
+    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);
+    // Lookup table to convert signed nibbles to signed bytes expanded to 512 bit length
+    __m512i signextendlutexpanded = _mm512_inserti32x8(_mm512_castsi256_si512(signextendlut), signextendlut, 1);
+
+    // Take group of four block_q8_0x4 structures at each pass of the loop and perform dot product operation
+    for (; y < anr / 4; y += 4) {
+
+        const block_q8_0x4 * a_ptrs[4];
+
+        a_ptrs[0] = a_ptr_start + (y * nb);
+        for (int i = 0; i < 3; ++i) {
+            a_ptrs[i + 1] = a_ptrs[i] + nb;
+        }
+
+        // Take group of two block_q4_0x8 structures at each pass of the loop and perform dot product operation
+        for (int64_t x = 0; x < anc / 8; x += 2) {
+
+            const block_q4_0x8 * b_ptr_0 = b_ptr_start + ((x)     * b_nb);
+            const block_q4_0x8 * b_ptr_1 = b_ptr_start + ((x + 1) * b_nb);
+
+            // Master FP accumulators
+            __m512 acc_rows[16];
+            for (int i = 0; i < 16; i++) {
+                acc_rows[i] = _mm512_setzero_ps();
+            }
+
+            for (int64_t b = 0; b < nb; b++) {
+                // Load the sixteen block_q4_0 quantized values interleaved with each other in chunks of eight - B0,B1 ....BE,BF
+                const __m256i rhs_raw_mat_0123_0 = _mm256_loadu_si256((const __m256i *)(b_ptr_0[b].qs));
+                const __m256i rhs_raw_mat_4567_0 = _mm256_loadu_si256((const __m256i *)(b_ptr_0[b].qs + 32));
+                const __m256i rhs_raw_mat_0123_1 = _mm256_loadu_si256((const __m256i *)(b_ptr_0[b].qs + 64));
+                const __m256i rhs_raw_mat_4567_1 = _mm256_loadu_si256((const __m256i *)(b_ptr_0[b].qs + 96));
+
+                const __m256i rhs_raw_mat_89AB_0 = _mm256_loadu_si256((const __m256i *)(b_ptr_1[b].qs));
+                const __m256i rhs_raw_mat_CDEF_0 = _mm256_loadu_si256((const __m256i *)(b_ptr_1[b].qs + 32));
+                const __m256i rhs_raw_mat_89AB_1 = _mm256_loadu_si256((const __m256i *)(b_ptr_1[b].qs + 64));
+                const __m256i rhs_raw_mat_CDEF_1 = _mm256_loadu_si256((const __m256i *)(b_ptr_1[b].qs + 96));
+
+                // Save the values in the following vectors in the formats B0B1B4B5B8B9BCBD, B2B3B6B7BABBBEBF for further processing and storing of values
+                const __m256i rhs_raw_mat_0145_0 = _mm256_blend_epi32(rhs_raw_mat_0123_0, _mm256_permutevar8x32_epi32(rhs_raw_mat_4567_0, requiredOrder), 240);
+                const __m256i rhs_raw_mat_2367_0 = _mm256_blend_epi32(_mm256_permutevar8x32_epi32(rhs_raw_mat_0123_0, requiredOrder), rhs_raw_mat_4567_0, 240);
+                const __m256i rhs_raw_mat_0145_1 = _mm256_blend_epi32(rhs_raw_mat_0123_1, _mm256_permutevar8x32_epi32(rhs_raw_mat_4567_1, requiredOrder), 240);
+                const __m256i rhs_raw_mat_2367_1 = _mm256_blend_epi32(_mm256_permutevar8x32_epi32(rhs_raw_mat_0123_1, requiredOrder), rhs_raw_mat_4567_1, 240);
+
+                const __m256i rhs_raw_mat_89CD_0 = _mm256_blend_epi32(rhs_raw_mat_89AB_0, _mm256_permutevar8x32_epi32(rhs_raw_mat_CDEF_0, requiredOrder), 240);
+                const __m256i rhs_raw_mat_ABEF_0 = _mm256_blend_epi32(_mm256_permutevar8x32_epi32(rhs_raw_mat_89AB_0, requiredOrder), rhs_raw_mat_CDEF_0, 240);
+                const __m256i rhs_raw_mat_89CD_1 = _mm256_blend_epi32(rhs_raw_mat_89AB_1, _mm256_permutevar8x32_epi32(rhs_raw_mat_CDEF_1, requiredOrder), 240);
+                const __m256i rhs_raw_mat_ABEF_1 = _mm256_blend_epi32(_mm256_permutevar8x32_epi32(rhs_raw_mat_89AB_1, requiredOrder), rhs_raw_mat_CDEF_1, 240);
+
+                const __m512i rhs_raw_mat_014589CD_0 = _mm512_inserti32x8(_mm512_castsi256_si512(rhs_raw_mat_0145_0), rhs_raw_mat_89CD_0, 1);
+                const __m512i rhs_raw_mat_2367ABEF_0 = _mm512_inserti32x8(_mm512_castsi256_si512(rhs_raw_mat_2367_0), rhs_raw_mat_ABEF_0, 1);
+                const __m512i rhs_raw_mat_014589CD_1 = _mm512_inserti32x8(_mm512_castsi256_si512(rhs_raw_mat_0145_1), rhs_raw_mat_89CD_1, 1);
+                const __m512i rhs_raw_mat_2367ABEF_1 = _mm512_inserti32x8(_mm512_castsi256_si512(rhs_raw_mat_2367_1), rhs_raw_mat_ABEF_1, 1);
+
+                // 4-bit -> 8-bit - Sign is maintained
+                const __m512i rhs_mat_014589CD_0 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(rhs_raw_mat_014589CD_0, m4bexpanded)); //B0(0-7) B1(0-7) B4(0-7) B5(0-7) B8(0-7) B9(0-7) BC(0-7) BD(0-7)
+                const __m512i rhs_mat_2367ABEF_0 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(rhs_raw_mat_2367ABEF_0, m4bexpanded)); //B2(0-7) B3(0-7) B6(0-7) B7(0-7) BA(0-7) BB(0-7) BE(0-7) BF(0-7)
+
+                const __m512i rhs_mat_014589CD_1 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(rhs_raw_mat_014589CD_1, m4bexpanded)); //B0(8-15) B1(8-15) B4(8-15) B5(8-15) B8(8-15) B9(8-15) BC(8-15) BD(8-15)
+                const __m512i rhs_mat_2367ABEF_1 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(rhs_raw_mat_2367ABEF_1, m4bexpanded)); //B2(8-15) B3(8-15) B6(8-15) B7(8-15) BA(8-15) BB(8-15) BE(8-15) BF(8-15)
+
+                const __m512i rhs_mat_014589CD_2 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(_mm512_srli_epi16(rhs_raw_mat_014589CD_0, 4), m4bexpanded)); //B0(16-23) B1(16-23) B4(16-23) B5(16-23) B8(16-23) B9(16-23) BC(16-23) BD(16-23)
+                const __m512i rhs_mat_2367ABEF_2 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(_mm512_srli_epi16(rhs_raw_mat_2367ABEF_0, 4), m4bexpanded)); //B2(16-23) B3(16-23) B6(16-23) B7(16-23) BA(16-23) BB(16-23) BE(16-23) BF(16-23)
+
+                const __m512i rhs_mat_014589CD_3 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(_mm512_srli_epi16(rhs_raw_mat_014589CD_1, 4), m4bexpanded)); //B0(24-31) B1(24-31) B4(24-31) B5(24-31) B8(24-31) B9(24-31) BC(24-31) BD(24-31)
+                const __m512i rhs_mat_2367ABEF_3 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(_mm512_srli_epi16(rhs_raw_mat_2367ABEF_1, 4), m4bexpanded)); //B2(24-31) B3(24-31) B6(24-31) B7(24-31) BA(24-31) BB(24-31) BE(24-31) BF(24-31)
+
+                // Shuffle pattern one - right side input
+                const __m512i rhs_mat_014589CD_0_sp1 = _mm512_shuffle_epi32(rhs_mat_014589CD_0, 136); //B0(0-3) B1(0-3) B0(0-3) B1(0-3) B4(0-3) B5(0-3) B4(0-3) B5(0-3) B8(0-3) B9(0-3) B8(0-3) B9(0-3) BC(0-3) BD(0-3) BC(0-3) BD(0-3)
+                const __m512i rhs_mat_2367ABEF_0_sp1 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_0, 136); //B2(0-3) B3(0-3) B2(0-3) B3(0-3) B6(0-3) B7(0-3) B6(0-3) B7(0-3) BA(0-3) BB(0-3) BA(0-3) BB(0-3) BE(0-3) BF(0-3) BE(0-3) BF(0-3)
+
+                const __m512i rhs_mat_014589CD_1_sp1 = _mm512_shuffle_epi32(rhs_mat_014589CD_1, 136); //B0(8-11) B1(8-11) B0(8-11) B1(8-11) B4(8-11) B5(8-11) B4(8-11) B5(8-11) B8(8-11) B9(8-11) B8(8-11) B9(8-11) BC(8-11) BD(8-11) BC(8-11) BD(8-11)
+                const __m512i rhs_mat_2367ABEF_1_sp1 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_1, 136); //B2(8-11) B3(8-11) B2(8-11) B3(8-11) B6(8-11) B7(8-11) B6(8-11) B7(8-11) BA(8-11) BB(8-11) BA(8-11) BB(8-11) BE(8-11) BF(8-11) BE(8-11) BF(8-11)
+
+                const __m512i rhs_mat_014589CD_2_sp1 = _mm512_shuffle_epi32(rhs_mat_014589CD_2, 136); //B0(16-19) B1(16-19) B0(16-19) B1(16-19) B4(16-19) B5(16-19) B4(16-19) B5(16-19) B8(16-19) B9(16-19) B8(16-19) B9(16-19) BC(16-19) BD(16-19) BC(16-19) BD(16-19)
+                const __m512i rhs_mat_2367ABEF_2_sp1 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_2, 136); //B2(16-19) B3(16-19) B2(16-19) B3(16-19) B6(16-19) B7(16-19) B6(16-19) B7(16-19) BA(16-19) BB(16-19) BA(16-19) BB(16-19) BE(16-19) BF(16-19) BE(16-19) BF(16-19)
+
+                const __m512i rhs_mat_014589CD_3_sp1 = _mm512_shuffle_epi32(rhs_mat_014589CD_3, 136); //B0(24-27) B1(24-27) B0(24-27) B1(24-27) B4(24-27) B5(24-27) B4(24-27) B5(24-27) B8(24-27) B9(24-27) B8(24-27) B9(24-27) BC(24-27) BD(24-27) BC(24-27) BD(24-27)
+                const __m512i rhs_mat_2367ABEF_3_sp1 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_3, 136); //B2(24-27) B3(24-27) B2(24-27) B3(24-27) B6(24-27) B7(24-27) B6(24-27) B7(24-27) BA(24-27) BB(24-27) BA(24-27) BB(24-27) BE(24-27) BF(24-27) BE(24-27) BF(24-27)
+
+                // Shuffle pattern two - right side input
+
+                const __m512i rhs_mat_014589CD_0_sp2 = _mm512_shuffle_epi32(rhs_mat_014589CD_0, 221); //B0(4-7) B1(4-7) B0(4-7) B1(4-7) B4(4-7) B5(4-7) B4(4-7) B5(4-7) B8(4-7) B9(4-7) B8(4-7) B9(4-7) BC(4-7) BD(4-7) BC(4-7) BD(4-7)
+                const __m512i rhs_mat_2367ABEF_0_sp2 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_0, 221); //B2(4-7) B3(4-7) B2(4-7) B3(4-7) B6(4-7) B7(4-7) B6(4-7) B7(4-7) BA(4-7) BB(4-7) BA(4-7) BB(4-7) BE(4-7) BF(4-7) BE(4-7) BF(4-7)
+
+                const __m512i rhs_mat_014589CD_1_sp2 = _mm512_shuffle_epi32(rhs_mat_014589CD_1, 221); //B0(12-15) B1(12-15) B0(12-15) B1(12-15) B4(12-15) B5(12-15) B4(12-15) B5(12-15) B8(12-15) B9(12-15) B8(12-15) B9(12-15) BC(12-15) BD(12-15) BC(12-15) BD(12-15)
+                const __m512i rhs_mat_2367ABEF_1_sp2 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_1, 221); //B2(12-15) B3(12-15) B2(12-15) B3(12-15) B6(12-15) B7(12-15) B6(12-15) B7(12-15) BA(12-15) BB(12-15) BA(12-15) BB(12-15) BE(12-15) BF(12-15) BE(12-15) BF(12-15)
+
+                const __m512i rhs_mat_014589CD_2_sp2 = _mm512_shuffle_epi32(rhs_mat_014589CD_2, 221); //B0(20-23) B1(20-23) B0(20-23) B1(20-23) B4(20-23) B5(20-23) B4(20-23) B5(20-23) B8(20-23) B9(20-23) B8(20-23) B9(20-23) BC(20-23) BD(20-23) BC(20-23) BD(20-23)
+                const __m512i rhs_mat_2367ABEF_2_sp2 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_2, 221); //B2(20-23) B3(20-23) B2(20-23) B3(20-23) B6(20-23) B7(20-23) B6(20-23) B7(20-23) BA(20-23) BB(20-23) BA(20-23) BB(20-23) BE(20-23) BF(20-23) BE(20-23) BF(20-23)
+
+                const __m512i rhs_mat_014589CD_3_sp2 = _mm512_shuffle_epi32(rhs_mat_014589CD_3, 221); //B0(28-31) B1(28-31) B0(28-31) B1(28-31) B4(28-31) B5(28-31) B4(28-31) B5(28-31) B8(28-31) B9(28-31) B8(28-31) B9(28-31) BC(28-31) BD(28-31) BC(28-31) BD(28-31)
+                const __m512i rhs_mat_2367ABEF_3_sp2 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_3, 221); //B2(28-31) B3(28-31) B2(28-31) B3(28-31) B6(28-31) B7(28-31) B6(28-31) B7(28-31) BA(28-31) BB(28-31) BA(28-31) BB(28-31) BE(28-31) BF(28-31) BE(28-31) BF(28-31)
+
+                // Scale values - Load the weight scale values of two block_q4_0x8
+                const __m512 col_scale_f32 = GGML_F32Cx8x2_LOAD(b_ptr_0[b].d, b_ptr_1[b].d);
+
+                // Process LHS in pairs of rows
+                for (int rp = 0; rp < 4; rp++) {
+
+                    // Load the four block_q4_0 quantized values interleaved with each other in chunks of eight - A0,A1,A2,A3
+                    // Loaded as set of 128 bit vectors and repeated and stored into a 256 bit vector before again repeating into 512 bit vector
+                    __m256i lhs_mat_ymm_0123_0 = _mm256_loadu_si256((const __m256i *)((a_ptrs[rp][b].qs)));
+                    __m256i lhs_mat_ymm_01_0 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_0, lhs_mat_ymm_0123_0, 0);
+                    __m256i lhs_mat_ymm_23_0 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_0, lhs_mat_ymm_0123_0, 17);
+                    __m256i lhs_mat_ymm_0123_1 = _mm256_loadu_si256((const __m256i *)((a_ptrs[rp][b].qs + 32)));
+                    __m256i lhs_mat_ymm_01_1 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_1, lhs_mat_ymm_0123_1, 0);
+                    __m256i lhs_mat_ymm_23_1 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_1, lhs_mat_ymm_0123_1, 17);
+                    __m256i lhs_mat_ymm_0123_2 = _mm256_loadu_si256((const __m256i *)((a_ptrs[rp][b].qs + 64)));
+                    __m256i lhs_mat_ymm_01_2 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_2, lhs_mat_ymm_0123_2, 0);
+                    __m256i lhs_mat_ymm_23_2 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_2, lhs_mat_ymm_0123_2, 17);
+                    __m256i lhs_mat_ymm_0123_3 = _mm256_loadu_si256((const __m256i *)((a_ptrs[rp][b].qs + 96)));
+                    __m256i lhs_mat_ymm_01_3 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_3, lhs_mat_ymm_0123_3, 0);
+                    __m256i lhs_mat_ymm_23_3 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_3, lhs_mat_ymm_0123_3, 17);
+
+                    __m512i lhs_mat_01_0 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_01_0), lhs_mat_ymm_01_0, 1);
+                    __m512i lhs_mat_23_0 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_23_0), lhs_mat_ymm_23_0, 1);
+                    __m512i lhs_mat_01_1 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_01_1), lhs_mat_ymm_01_1, 1);
+                    __m512i lhs_mat_23_1 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_23_1), lhs_mat_ymm_23_1, 1);
+                    __m512i lhs_mat_01_2 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_01_2), lhs_mat_ymm_01_2, 1);
+                    __m512i lhs_mat_23_2 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_23_2), lhs_mat_ymm_23_2, 1);
+                    __m512i lhs_mat_01_3 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_01_3), lhs_mat_ymm_01_3, 1);
+                    __m512i lhs_mat_23_3 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_23_3), lhs_mat_ymm_23_3, 1);
+
+                    // Shuffle pattern one - left side input
+
+                    const __m512i lhs_mat_01_0_sp1 = _mm512_shuffle_epi32(lhs_mat_01_0, 160);  //A0(0-3) A0(0-3) A1(0-3) A1(0-3) A0(0-3) A0(0-3) A1(0-3) A1(0-3) A0(0-3) A0(0-3) A1(0-3) A1(0-3) A0(0-3) A0(0-3) A1(0-3) A1(0-3)
+                    const __m512i lhs_mat_23_0_sp1 = _mm512_shuffle_epi32(lhs_mat_23_0, 160);  //A2(0-3) A2(0-3) A3(0-3) A3(0-3) A2(0-3) A2(0-3) A3(0-3) A3(0-3) A2(0-3) A2(0-3) A3(0-3) A3(0-3) A2(0-3) A2(0-3) A3(0-3) A3(0-3)
+
+                    const __m512i lhs_mat_01_1_sp1 = _mm512_shuffle_epi32(lhs_mat_01_1, 160);  //A0(8-11) A0(8-11) A1(8-11) A1(8-11) A0(8-11) A0(8-11) A1(8-11) A1(8-11) A0(8-11) A0(8-11) A1(8-11) A1(8-11) A0(8-11) A0(8-11) A1(8-11) A1(8-11)
+                    const __m512i lhs_mat_23_1_sp1 = _mm512_shuffle_epi32(lhs_mat_23_1, 160);  //A2(8-11) A2(8-11) A3(8-11) A3(8-11) A2(8-11) A2(8-11) A3(8-11) A3(8-11) A2(8-11) A2(8-11) A3(8-11) A3(8-11) A2(8-11) A2(8-11) A3(8-11) A3(8-11)
+
+                    const __m512i lhs_mat_01_2_sp1 = _mm512_shuffle_epi32(lhs_mat_01_2, 160);  //A0(16-19) A0(16-19) A1(16-19) A1(16-19) A0(16-19) A0(16-19) A1(16-19) A1(16-19) A0(16-19) A0(16-19) A1(16-19) A1(16-19) A0(16-19) A0(16-19) A1(16-19) A1(16-19)
+                    const __m512i lhs_mat_23_2_sp1 = _mm512_shuffle_epi32(lhs_mat_23_2, 160);  //A2(16-19) A2(16-19) A3(16-19) A3(16-19) A2(16-19) A2(16-19) A3(16-19) A3(16-19) A2(16-19) A2(16-19) A3(16-19) A3(16-19) A2(16-19) A2(16-19) A3(16-19) A3(16-19)
+
+                    const __m512i lhs_mat_01_3_sp1 = _mm512_shuffle_epi32(lhs_mat_01_3, 160);  //A0(24-27) A0(24-27) A1(24-27) A1(24-27) A0(24-27) A0(24-27) A1(24-27) A1(24-27) A0(24-27) A0(24-27) A1(24-27) A1(24-27) A0(24-27) A0(24-27) A1(24-27) A1(24-27)
+                    const __m512i lhs_mat_23_3_sp1 = _mm512_shuffle_epi32(lhs_mat_23_3, 160);  //A2(24-27) A2(24-27) A3(24-27) A3(24-27) A2(24-27) A2(24-27) A3(24-27) A3(24-27) A2(24-27) A2(24-27) A3(24-27) A3(24-27) A2(24-27) A2(24-27) A3(24-27) A3(24-27)
+
+                    // Shuffle pattern two - left side input
+
+                    const __m512i lhs_mat_01_0_sp2 = _mm512_shuffle_epi32(lhs_mat_01_0, 245);  //A0(4-7) A0(4-7) A1(4-7) A1(4-7) A0(4-7) A0(4-7) A1(4-7) A1(4-7) A0(4-7) A0(4-7) A1(4-7) A1(4-7) A0(4-7) A0(4-7) A1(4-7) A1(4-7)
+                    const __m512i lhs_mat_23_0_sp2 = _mm512_shuffle_epi32(lhs_mat_23_0, 245);  //A2(4-7) A2(4-7) A3(4-7) A3(4-7) A2(4-7) A2(4-7) A3(4-7) A3(4-7) A2(4-7) A2(4-7) A3(4-7) A3(4-7) A2(4-7) A2(4-7) A3(4-7) A3(4-7)
+
+                    const __m512i lhs_mat_01_1_sp2 = _mm512_shuffle_epi32(lhs_mat_01_1, 245);  //A0(12-15) A0(12-15) A1(12-15) A1(12-15) A0(12-15) A0(12-15) A1(12-15) A1(12-15) A0(12-15) A0(12-15) A1(12-15) A1(12-15) A0(12-15) A0(12-15) A1(12-15) A1(12-15)
+                    const __m512i lhs_mat_23_1_sp2 = _mm512_shuffle_epi32(lhs_mat_23_1, 245);  //A2(12-15) A2(12-15) A3(12-15) A3(12-15) A2(12-15) A2(12-15) A3(12-15) A3(12-15) A2(12-15) A2(12-15) A3(12-15) A3(12-15) A2(12-15) A2(12-15) A3(12-15) A3(12-15)
+
+                    const __m512i lhs_mat_01_2_sp2 = _mm512_shuffle_epi32(lhs_mat_01_2, 245);  //A0(20-23) A0(20-23) A1(20-23) A1(20-23) A0(20-23) A0(20-23) A1(20-23) A1(20-23) A0(20-23) A0(20-23) A1(20-23) A1(20-23) A0(20-23) A0(20-23) A1(20-23) A1(20-23)
+                    const __m512i lhs_mat_23_2_sp2 = _mm512_shuffle_epi32(lhs_mat_23_2, 245);  //A2(20-23) A2(20-23) A3(20-23) A3(20-23) A2(20-23) A2(20-23) A3(20-23) A3(20-23) A2(20-23) A2(20-23) A3(20-23) A3(20-23) A2(20-23) A2(20-23) A3(20-23) A3(20-23)
+
+                    const __m512i lhs_mat_01_3_sp2 = _mm512_shuffle_epi32(lhs_mat_01_3, 245);  //A0(28-31) A0(28-31) A1(28-31) A1(28-31) A0(28-31) A0(28-31) A1(28-31) A1(28-31) A0(28-31) A0(28-31) A1(28-31) A1(28-31) A0(28-31) A0(28-31) A1(28-31) A1(28-31)
+                    const __m512i lhs_mat_23_3_sp2 = _mm512_shuffle_epi32(lhs_mat_23_3, 245);  //A2(28-31) A2(28-31) A3(28-31) A3(28-31) A2(28-31) A2(28-31) A3(28-31) A3(28-31) A2(28-31) A2(28-31) A3(28-31) A3(28-31) A2(28-31) A2(28-31) A3(28-31) A3(28-31)
+
+                    // The values arranged in shuffle patterns are operated with dot product operation within 32 bit lane i.e corresponding bytes and multiplied and added into 32 bit integers within 32 bit lane
+                    // Resembles MMLAs into 2x2 matrices in ARM Version
+                    __m512i iacc_mat_00_sp1 =
+                        _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_01_3_sp1, rhs_mat_014589CD_3_sp1), mul_sum_i8_pairs_int32x16(lhs_mat_01_2_sp1, rhs_mat_014589CD_2_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_01_1_sp1, rhs_mat_014589CD_1_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_01_0_sp1, rhs_mat_014589CD_0_sp1));
+                    __m512i iacc_mat_01_sp1 =
+                        _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_01_3_sp1, rhs_mat_2367ABEF_3_sp1), mul_sum_i8_pairs_int32x16(lhs_mat_01_2_sp1, rhs_mat_2367ABEF_2_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_01_1_sp1, rhs_mat_2367ABEF_1_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_01_0_sp1, rhs_mat_2367ABEF_0_sp1));
+                    __m512i iacc_mat_10_sp1 =
+                        _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_23_3_sp1, rhs_mat_014589CD_3_sp1), mul_sum_i8_pairs_int32x16(lhs_mat_23_2_sp1, rhs_mat_014589CD_2_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_23_1_sp1, rhs_mat_014589CD_1_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_23_0_sp1, rhs_mat_014589CD_0_sp1));
+                    __m512i iacc_mat_11_sp1 =
+                        _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_23_3_sp1, rhs_mat_2367ABEF_3_sp1), mul_sum_i8_pairs_int32x16(lhs_mat_23_2_sp1, rhs_mat_2367ABEF_2_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_23_1_sp1, rhs_mat_2367ABEF_1_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_23_0_sp1, rhs_mat_2367ABEF_0_sp1));
+                    __m512i iacc_mat_00_sp2 =
+                        _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_01_3_sp2, rhs_mat_014589CD_3_sp2), mul_sum_i8_pairs_int32x16(lhs_mat_01_2_sp2, rhs_mat_014589CD_2_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_01_1_sp2, rhs_mat_014589CD_1_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_01_0_sp2, rhs_mat_014589CD_0_sp2));
+                    __m512i iacc_mat_01_sp2 =
+                        _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_01_3_sp2, rhs_mat_2367ABEF_3_sp2), mul_sum_i8_pairs_int32x16(lhs_mat_01_2_sp2, rhs_mat_2367ABEF_2_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_01_1_sp2, rhs_mat_2367ABEF_1_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_01_0_sp2, rhs_mat_2367ABEF_0_sp2));
+                    __m512i iacc_mat_10_sp2 =
+                        _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_23_3_sp2, rhs_mat_014589CD_3_sp2), mul_sum_i8_pairs_int32x16(lhs_mat_23_2_sp2, rhs_mat_014589CD_2_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_23_1_sp2, rhs_mat_014589CD_1_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_23_0_sp2, rhs_mat_014589CD_0_sp2));
+                    __m512i iacc_mat_11_sp2 =
+                        _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_23_3_sp2, rhs_mat_2367ABEF_3_sp2), mul_sum_i8_pairs_int32x16(lhs_mat_23_2_sp2, rhs_mat_2367ABEF_2_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_23_1_sp2, rhs_mat_2367ABEF_1_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_23_0_sp2, rhs_mat_2367ABEF_0_sp2));
+
+                    // Output of both shuffle patterns are added in order to sum dot product outputs of all 32 values in block
+                    __m512i iacc_mat_00 = _mm512_add_epi32(iacc_mat_00_sp1, iacc_mat_00_sp2);
+                    __m512i iacc_mat_01 = _mm512_add_epi32(iacc_mat_01_sp1, iacc_mat_01_sp2);
+                    __m512i iacc_mat_10 = _mm512_add_epi32(iacc_mat_10_sp1, iacc_mat_10_sp2);
+                    __m512i iacc_mat_11 = _mm512_add_epi32(iacc_mat_11_sp1, iacc_mat_11_sp2);
+
+
+                    // Straighten out to make 4 row vectors
+                    __m512i iacc_row_0 = _mm512_mask_blend_epi32(0xCCCC, iacc_mat_00, _mm512_shuffle_epi32(iacc_mat_01, 78));
+                    __m512i iacc_row_1 = _mm512_mask_blend_epi32(0xCCCC, _mm512_shuffle_epi32(iacc_mat_00, 78), iacc_mat_01);
+                    __m512i iacc_row_2 = _mm512_mask_blend_epi32(0xCCCC, iacc_mat_10, _mm512_shuffle_epi32(iacc_mat_11, 78));
+                    __m512i iacc_row_3 = _mm512_mask_blend_epi32(0xCCCC, _mm512_shuffle_epi32(iacc_mat_10, 78), iacc_mat_11);
+
+                    // Load the scale(d) values for all the 4 Q8_0 blocks and repeat it across lanes
+                    const __m128i row_scale_f16 = _mm_shuffle_epi32(_mm_maskload_epi32((int const*)(a_ptrs[rp][b].d), loadMask), 68);
+                    const __m512 row_scale_f32 = GGML_F32Cx16_REPEAT_LOAD(row_scale_f16);
+
+                    // Multiply with appropiate scales and accumulate
+                    acc_rows[rp * 4]     = _mm512_fmadd_ps(_mm512_cvtepi32_ps(iacc_row_0), _mm512_mul_ps(col_scale_f32, _mm512_shuffle_ps(row_scale_f32, row_scale_f32, 0)),   acc_rows[rp * 4]);
+                    acc_rows[rp * 4 + 1] = _mm512_fmadd_ps(_mm512_cvtepi32_ps(iacc_row_1), _mm512_mul_ps(col_scale_f32, _mm512_shuffle_ps(row_scale_f32, row_scale_f32, 85)),  acc_rows[rp * 4 + 1]);
+                    acc_rows[rp * 4 + 2] = _mm512_fmadd_ps(_mm512_cvtepi32_ps(iacc_row_2), _mm512_mul_ps(col_scale_f32, _mm512_shuffle_ps(row_scale_f32, row_scale_f32, 170)), acc_rows[rp * 4 + 2]);
+                    acc_rows[rp * 4 + 3] = _mm512_fmadd_ps(_mm512_cvtepi32_ps(iacc_row_3), _mm512_mul_ps(col_scale_f32, _mm512_shuffle_ps(row_scale_f32, row_scale_f32, 255)), acc_rows[rp * 4 + 3]);
+                }
+            }
+
+            // Store the accumulated values
+            for (int i = 0; i < 16; i++) {
+                _mm512_storeu_ps((float *)(s + ((y * 4 + i) * bs + x * 8)), acc_rows[i]);
+            }
+        }
+    }
+    // Take a block_q8_0x4 structures at each pass of the loop and perform dot product operation
+    for (; y < nr / 4; y ++) {
+
+        const block_q8_0x4 * a_ptr = a_ptr_start + (y * nb);
+
+        // Take group of two block_q4_0x8 structures at each pass of the loop and perform dot product operation
+        for (int64_t x = 0; x < anc / 8; x += 2) {
+
+            const block_q4_0x8 * b_ptr_0 = b_ptr_start + ((x)     * b_nb);
+            const block_q4_0x8 * b_ptr_1 = b_ptr_start + ((x + 1) * b_nb);
+
+            // Master FP accumulators
+            __m512 acc_rows[4];
+            for (int i = 0; i < 4; i++) {
+                acc_rows[i] = _mm512_setzero_ps();
+            }
+
+            for (int64_t b = 0; b < nb; b++) {
+                // Load the sixteen block_q4_0 quantized values interleaved with each other in chunks of eight - B0,B1 ....BE,BF
+                const __m256i rhs_raw_mat_0123_0 = _mm256_loadu_si256((const __m256i *)(b_ptr_0[b].qs));
+                const __m256i rhs_raw_mat_4567_0 = _mm256_loadu_si256((const __m256i *)(b_ptr_0[b].qs + 32));
+                const __m256i rhs_raw_mat_0123_1 = _mm256_loadu_si256((const __m256i *)(b_ptr_0[b].qs + 64));
+                const __m256i rhs_raw_mat_4567_1 = _mm256_loadu_si256((const __m256i *)(b_ptr_0[b].qs + 96));
+
+                const __m256i rhs_raw_mat_89AB_0 = _mm256_loadu_si256((const __m256i *)(b_ptr_1[b].qs));
+                const __m256i rhs_raw_mat_CDEF_0 = _mm256_loadu_si256((const __m256i *)(b_ptr_1[b].qs + 32));
+                const __m256i rhs_raw_mat_89AB_1 = _mm256_loadu_si256((const __m256i *)(b_ptr_1[b].qs + 64));
+                const __m256i rhs_raw_mat_CDEF_1 = _mm256_loadu_si256((const __m256i *)(b_ptr_1[b].qs + 96));
+
+                // Save the values in the following vectors in the formats B0B1B4B5, B2B3B6B7 for further processing and storing of valuess
+                const __m256i rhs_raw_mat_0145_0 = _mm256_blend_epi32(rhs_raw_mat_0123_0, _mm256_permutevar8x32_epi32(rhs_raw_mat_4567_0, requiredOrder), 240);
+                const __m256i rhs_raw_mat_2367_0 = _mm256_blend_epi32(_mm256_permutevar8x32_epi32(rhs_raw_mat_0123_0, requiredOrder), rhs_raw_mat_4567_0, 240);
+                const __m256i rhs_raw_mat_0145_1 = _mm256_blend_epi32(rhs_raw_mat_0123_1, _mm256_permutevar8x32_epi32(rhs_raw_mat_4567_1, requiredOrder), 240);
+                const __m256i rhs_raw_mat_2367_1 = _mm256_blend_epi32(_mm256_permutevar8x32_epi32(rhs_raw_mat_0123_1, requiredOrder), rhs_raw_mat_4567_1, 240);
+
+                const __m256i rhs_raw_mat_89CD_0 = _mm256_blend_epi32(rhs_raw_mat_89AB_0, _mm256_permutevar8x32_epi32(rhs_raw_mat_CDEF_0, requiredOrder), 240);
+                const __m256i rhs_raw_mat_ABEF_0 = _mm256_blend_epi32(_mm256_permutevar8x32_epi32(rhs_raw_mat_89AB_0, requiredOrder), rhs_raw_mat_CDEF_0, 240);
+                const __m256i rhs_raw_mat_89CD_1 = _mm256_blend_epi32(rhs_raw_mat_89AB_1, _mm256_permutevar8x32_epi32(rhs_raw_mat_CDEF_1, requiredOrder), 240);
+                const __m256i rhs_raw_mat_ABEF_1 = _mm256_blend_epi32(_mm256_permutevar8x32_epi32(rhs_raw_mat_89AB_1, requiredOrder), rhs_raw_mat_CDEF_1, 240);
+
+                const __m512i rhs_raw_mat_014589CD_0 = _mm512_inserti32x8(_mm512_castsi256_si512(rhs_raw_mat_0145_0), rhs_raw_mat_89CD_0, 1);
+                const __m512i rhs_raw_mat_2367ABEF_0 = _mm512_inserti32x8(_mm512_castsi256_si512(rhs_raw_mat_2367_0), rhs_raw_mat_ABEF_0, 1);
+                const __m512i rhs_raw_mat_014589CD_1 = _mm512_inserti32x8(_mm512_castsi256_si512(rhs_raw_mat_0145_1), rhs_raw_mat_89CD_1, 1);
+                const __m512i rhs_raw_mat_2367ABEF_1 = _mm512_inserti32x8(_mm512_castsi256_si512(rhs_raw_mat_2367_1), rhs_raw_mat_ABEF_1, 1);
+
+                // 4-bit -> 8-bit - Sign is maintained
+                const __m512i rhs_mat_014589CD_0 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(rhs_raw_mat_014589CD_0, m4bexpanded)); //B0(0-7) B1(0-7) B4(0-7) B5(0-7) B8(0-7) B9(0-7) BC(0-7) BD(0-7)
+                const __m512i rhs_mat_2367ABEF_0 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(rhs_raw_mat_2367ABEF_0, m4bexpanded)); //B2(0-7) B3(0-7) B6(0-7) B7(0-7) BA(0-7) BB(0-7) BE(0-7) BF(0-7)
+
+                const __m512i rhs_mat_014589CD_1 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(rhs_raw_mat_014589CD_1, m4bexpanded)); //B0(8-15) B1(8-15) B4(8-15) B5(8-15) B8(8-15) B9(8-15) BC(8-15) BD(8-15)
+                const __m512i rhs_mat_2367ABEF_1 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(rhs_raw_mat_2367ABEF_1, m4bexpanded)); //B2(8-15) B3(8-15) B6(8-15) B7(8-15) BA(8-15) BB(8-15) BE(8-15) BF(8-15)
+
+                const __m512i rhs_mat_014589CD_2 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(_mm512_srli_epi16(rhs_raw_mat_014589CD_0, 4), m4bexpanded)); //B0(16-23) B1(16-23) B4(16-23) B5(16-23) B8(16-23) B9(16-23) BC(16-23) BD(16-23)
+                const __m512i rhs_mat_2367ABEF_2 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(_mm512_srli_epi16(rhs_raw_mat_2367ABEF_0, 4), m4bexpanded)); //B2(16-23) B3(16-23) B6(16-23) B7(16-23) BA(16-23) BB(16-23) BE(16-23) BF(16-23)
+
+                const __m512i rhs_mat_014589CD_3 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(_mm512_srli_epi16(rhs_raw_mat_014589CD_1, 4), m4bexpanded)); //B0(24-31) B1(24-31) B4(24-31) B5(24-31) B8(24-31) B9(24-31) BC(24-31) BD(24-31)
+                const __m512i rhs_mat_2367ABEF_3 = _mm512_shuffle_epi8(signextendlutexpanded, _mm512_and_si512(_mm512_srli_epi16(rhs_raw_mat_2367ABEF_1, 4), m4bexpanded)); //B2(24-31) B3(24-31) B6(24-31) B7(24-31) BA(24-31) BB(24-31) BE(24-31) BF(24-31)
+
+                // Shuffle pattern one - right side input
+                const __m512i rhs_mat_014589CD_0_sp1 = _mm512_shuffle_epi32(rhs_mat_014589CD_0, 136); //B0(0-3) B1(0-3) B0(0-3) B1(0-3) B4(0-3) B5(0-3) B4(0-3) B5(0-3) B8(0-3) B9(0-3) B8(0-3) B9(0-3) BC(0-3) BD(0-3) BC(0-3) BD(0-3)
+                const __m512i rhs_mat_2367ABEF_0_sp1 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_0, 136); //B2(0-3) B3(0-3) B2(0-3) B3(0-3) B6(0-3) B7(0-3) B6(0-3) B7(0-3) BA(0-3) BB(0-3) BA(0-3) BB(0-3) BE(0-3) BF(0-3) BE(0-3) BF(0-3)
+
+                const __m512i rhs_mat_014589CD_1_sp1 = _mm512_shuffle_epi32(rhs_mat_014589CD_1, 136); //B0(8-11) B1(8-11) B0(8-11) B1(8-11) B4(8-11) B5(8-11) B4(8-11) B5(8-11) B8(8-11) B9(8-11) B8(8-11) B9(8-11) BC(8-11) BD(8-11) BC(8-11) BD(8-11)
+                const __m512i rhs_mat_2367ABEF_1_sp1 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_1, 136); //B2(8-11) B3(8-11) B2(8-11) B3(8-11) B6(8-11) B7(8-11) B6(8-11) B7(8-11) BA(8-11) BB(8-11) BA(8-11) BB(8-11) BE(8-11) BF(8-11) BE(8-11) BF(8-11)
+
+                const __m512i rhs_mat_014589CD_2_sp1 = _mm512_shuffle_epi32(rhs_mat_014589CD_2, 136); //B0(16-19) B1(16-19) B0(16-19) B1(16-19) B4(16-19) B5(16-19) B4(16-19) B5(16-19) B8(16-19) B9(16-19) B8(16-19) B9(16-19) BC(16-19) BD(16-19) BC(16-19) BD(16-19)
+                const __m512i rhs_mat_2367ABEF_2_sp1 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_2, 136); //B2(16-19) B3(16-19) B2(16-19) B3(16-19) B6(16-19) B7(16-19) B6(16-19) B7(16-19) BA(16-19) BB(16-19) BA(16-19) BB(16-19) BE(16-19) BF(16-19) BE(16-19) BF(16-19)
+
+                const __m512i rhs_mat_014589CD_3_sp1 = _mm512_shuffle_epi32(rhs_mat_014589CD_3, 136); //B0(24-27) B1(24-27) B0(24-27) B1(24-27) B4(24-27) B5(24-27) B4(24-27) B5(24-27) B8(24-27) B9(24-27) B8(24-27) B9(24-27) BC(24-27) BD(24-27) BC(24-27) BD(24-27)
+                const __m512i rhs_mat_2367ABEF_3_sp1 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_3, 136); //B2(24-27) B3(24-27) B2(24-27) B3(24-27) B6(24-27) B7(24-27) B6(24-27) B7(24-27) BA(24-27) BB(24-27) BA(24-27) BB(24-27) BE(24-27) BF(24-27) BE(24-27) BF(24-27)
+
+                // Shuffle pattern two - right side input
+
+                const __m512i rhs_mat_014589CD_0_sp2 = _mm512_shuffle_epi32(rhs_mat_014589CD_0, 221); //B0(4-7) B1(4-7) B0(4-7) B1(4-7) B4(4-7) B5(4-7) B4(4-7) B5(4-7) B8(4-7) B9(4-7) B8(4-7) B9(4-7) BC(4-7) BD(4-7) BC(4-7) BD(4-7)
+                const __m512i rhs_mat_2367ABEF_0_sp2 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_0, 221); //B2(4-7) B3(4-7) B2(4-7) B3(4-7) B6(4-7) B7(4-7) B6(4-7) B7(4-7) BA(4-7) BB(4-7) BA(4-7) BB(4-7) BE(4-7) BF(4-7) BE(4-7) BF(4-7)
+
+                const __m512i rhs_mat_014589CD_1_sp2 = _mm512_shuffle_epi32(rhs_mat_014589CD_1, 221); //B0(12-15) B1(12-15) B0(12-15) B1(12-15) B4(12-15) B5(12-15) B4(12-15) B5(12-15) B8(12-15) B9(12-15) B8(12-15) B9(12-15) BC(12-15) BD(12-15) BC(12-15) BD(12-15)
+                const __m512i rhs_mat_2367ABEF_1_sp2 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_1, 221); //B2(12-15) B3(12-15) B2(12-15) B3(12-15) B6(12-15) B7(12-15) B6(12-15) B7(12-15) BA(12-15) BB(12-15) BA(12-15) BB(12-15) BE(12-15) BF(12-15) BE(12-15) BF(12-15)
+
+                const __m512i rhs_mat_014589CD_2_sp2 = _mm512_shuffle_epi32(rhs_mat_014589CD_2, 221); //B0(20-23) B1(20-23) B0(20-23) B1(20-23) B4(20-23) B5(20-23) B4(20-23) B5(20-23) B8(20-23) B9(20-23) B8(20-23) B9(20-23) BC(20-23) BD(20-23) BC(20-23) BD(20-23)
+                const __m512i rhs_mat_2367ABEF_2_sp2 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_2, 221); //B2(20-23) B3(20-23) B2(20-23) B3(20-23) B6(20-23) B7(20-23) B6(20-23) B7(20-23) BA(20-23) BB(20-23) BA(20-23) BB(20-23) BE(20-23) BF(20-23) BE(20-23) BF(20-23)
+
+                const __m512i rhs_mat_014589CD_3_sp2 = _mm512_shuffle_epi32(rhs_mat_014589CD_3, 221); //B0(28-31) B1(28-31) B0(28-31) B1(28-31) B4(28-31) B5(28-31) B4(28-31) B5(28-31) B8(28-31) B9(28-31) B8(28-31) B9(28-31) BC(28-31) BD(28-31) BC(28-31) BD(28-31)
+                const __m512i rhs_mat_2367ABEF_3_sp2 = _mm512_shuffle_epi32(rhs_mat_2367ABEF_3, 221); //B2(28-31) B3(28-31) B2(28-31) B3(28-31) B6(28-31) B7(28-31) B6(28-31) B7(28-31) BA(28-31) BB(28-31) BA(28-31) BB(28-31) BE(28-31) BF(28-31) BE(28-31) BF(28-31)
+
+
+                // Scale values - Load the weight scale values of two block_q4_0x8
+                const __m512 col_scale_f32 = GGML_F32Cx8x2_LOAD(b_ptr_0[b].d, b_ptr_1[b].d);
+
+                // Load the four block_q4_0 quantized values interleaved with each other in chunks of eight - A0,A1,A2,A3
+                // Loaded as set of 128 bit vectors and repeated and stored into a 256 bit vector before again repeating into 512 bit vector
+                __m256i lhs_mat_ymm_0123_0 = _mm256_loadu_si256((const __m256i *)((a_ptr[b].qs)));
+                __m256i lhs_mat_ymm_01_0 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_0, lhs_mat_ymm_0123_0, 0);
+                __m256i lhs_mat_ymm_23_0 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_0, lhs_mat_ymm_0123_0, 17);
+                __m256i lhs_mat_ymm_0123_1 = _mm256_loadu_si256((const __m256i *)((a_ptr[b].qs + 32)));
+                __m256i lhs_mat_ymm_01_1 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_1, lhs_mat_ymm_0123_1, 0);
+                __m256i lhs_mat_ymm_23_1 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_1, lhs_mat_ymm_0123_1, 17);
+                __m256i lhs_mat_ymm_0123_2 = _mm256_loadu_si256((const __m256i *)((a_ptr[b].qs + 64)));
+                __m256i lhs_mat_ymm_01_2 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_2, lhs_mat_ymm_0123_2, 0);
+                __m256i lhs_mat_ymm_23_2 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_2, lhs_mat_ymm_0123_2, 17);
+                __m256i lhs_mat_ymm_0123_3 = _mm256_loadu_si256((const __m256i *)((a_ptr[b].qs + 96)));
+                __m256i lhs_mat_ymm_01_3 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_3, lhs_mat_ymm_0123_3, 0);
+                __m256i lhs_mat_ymm_23_3 = _mm256_permute2f128_si256(lhs_mat_ymm_0123_3, lhs_mat_ymm_0123_3, 17);
+
+                __m512i lhs_mat_01_0 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_01_0), lhs_mat_ymm_01_0, 1);
+                __m512i lhs_mat_23_0 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_23_0), lhs_mat_ymm_23_0, 1);
+                __m512i lhs_mat_01_1 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_01_1), lhs_mat_ymm_01_1, 1);
+                __m512i lhs_mat_23_1 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_23_1), lhs_mat_ymm_23_1, 1);
+                __m512i lhs_mat_01_2 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_01_2), lhs_mat_ymm_01_2, 1);
+                __m512i lhs_mat_23_2 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_23_2), lhs_mat_ymm_23_2, 1);
+                __m512i lhs_mat_01_3 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_01_3), lhs_mat_ymm_01_3, 1);
+                __m512i lhs_mat_23_3 = _mm512_inserti32x8(_mm512_castsi256_si512(lhs_mat_ymm_23_3), lhs_mat_ymm_23_3, 1);
+
+                // Shuffle pattern one - left side input
+
+                const __m512i lhs_mat_01_0_sp1 = _mm512_shuffle_epi32(lhs_mat_01_0, 160);  //A0(0-3) A0(0-3) A1(0-3) A1(0-3) A0(0-3) A0(0-3) A1(0-3) A1(0-3) A0(0-3) A0(0-3) A1(0-3) A1(0-3) A0(0-3) A0(0-3) A1(0-3) A1(0-3)
+                const __m512i lhs_mat_23_0_sp1 = _mm512_shuffle_epi32(lhs_mat_23_0, 160);  //A2(0-3) A2(0-3) A3(0-3) A3(0-3) A2(0-3) A2(0-3) A3(0-3) A3(0-3) A2(0-3) A2(0-3) A3(0-3) A3(0-3) A2(0-3) A2(0-3) A3(0-3) A3(0-3)
+
+                const __m512i lhs_mat_01_1_sp1 = _mm512_shuffle_epi32(lhs_mat_01_1, 160);  //A0(8-11) A0(8-11) A1(8-11) A1(8-11) A0(8-11) A0(8-11) A1(8-11) A1(8-11) A0(8-11) A0(8-11) A1(8-11) A1(8-11) A0(8-11) A0(8-11) A1(8-11) A1(8-11)
+                const __m512i lhs_mat_23_1_sp1 = _mm512_shuffle_epi32(lhs_mat_23_1, 160);  //A2(8-11) A2(8-11) A3(8-11) A3(8-11) A2(8-11) A2(8-11) A3(8-11) A3(8-11) A2(8-11) A2(8-11) A3(8-11) A3(8-11) A2(8-11) A2(8-11) A3(8-11) A3(8-11)
+
+                const __m512i lhs_mat_01_2_sp1 = _mm512_shuffle_epi32(lhs_mat_01_2, 160);  //A0(16-19) A0(16-19) A1(16-19) A1(16-19) A0(16-19) A0(16-19) A1(16-19) A1(16-19) A0(16-19) A0(16-19) A1(16-19) A1(16-19) A0(16-19) A0(16-19) A1(16-19) A1(16-19)
+                const __m512i lhs_mat_23_2_sp1 = _mm512_shuffle_epi32(lhs_mat_23_2, 160);  //A2(16-19) A2(16-19) A3(16-19) A3(16-19) A2(16-19) A2(16-19) A3(16-19) A3(16-19) A2(16-19) A2(16-19) A3(16-19) A3(16-19) A2(16-19) A2(16-19) A3(16-19) A3(16-19)
+
+                const __m512i lhs_mat_01_3_sp1 = _mm512_shuffle_epi32(lhs_mat_01_3, 160);  //A0(24-27) A0(24-27) A1(24-27) A1(24-27) A0(24-27) A0(24-27) A1(24-27) A1(24-27) A0(24-27) A0(24-27) A1(24-27) A1(24-27) A0(24-27) A0(24-27) A1(24-27) A1(24-27)
+                const __m512i lhs_mat_23_3_sp1 = _mm512_shuffle_epi32(lhs_mat_23_3, 160);  //A2(24-27) A2(24-27) A3(24-27) A3(24-27) A2(24-27) A2(24-27) A3(24-27) A3(24-27) A2(24-27) A2(24-27) A3(24-27) A3(24-27) A2(24-27) A2(24-27) A3(24-27) A3(24-27)
+
+                // Shuffle pattern two - left side input
+
+                const __m512i lhs_mat_01_0_sp2 = _mm512_shuffle_epi32(lhs_mat_01_0, 245);  //A0(4-7) A0(4-7) A1(4-7) A1(4-7) A0(4-7) A0(4-7) A1(4-7) A1(4-7) A0(4-7) A0(4-7) A1(4-7) A1(4-7) A0(4-7) A0(4-7) A1(4-7) A1(4-7)
+                const __m512i lhs_mat_23_0_sp2 = _mm512_shuffle_epi32(lhs_mat_23_0, 245);  //A2(4-7) A2(4-7) A3(4-7) A3(4-7) A2(4-7) A2(4-7) A3(4-7) A3(4-7) A2(4-7) A2(4-7) A3(4-7) A3(4-7) A2(4-7) A2(4-7) A3(4-7) A3(4-7)
+
+                const __m512i lhs_mat_01_1_sp2 = _mm512_shuffle_epi32(lhs_mat_01_1, 245);  //A0(12-15) A0(12-15) A1(12-15) A1(12-15) A0(12-15) A0(12-15) A1(12-15) A1(12-15) A0(12-15) A0(12-15) A1(12-15) A1(12-15) A0(12-15) A0(12-15) A1(12-15) A1(12-15)
+                const __m512i lhs_mat_23_1_sp2 = _mm512_shuffle_epi32(lhs_mat_23_1, 245);  //A2(12-15) A2(12-15) A3(12-15) A3(12-15) A2(12-15) A2(12-15) A3(12-15) A3(12-15) A2(12-15) A2(12-15) A3(12-15) A3(12-15) A2(12-15) A2(12-15) A3(12-15) A3(12-15)
+
+                const __m512i lhs_mat_01_2_sp2 = _mm512_shuffle_epi32(lhs_mat_01_2, 245);  //A0(20-23) A0(20-23) A1(20-23) A1(20-23) A0(20-23) A0(20-23) A1(20-23) A1(20-23) A0(20-23) A0(20-23) A1(20-23) A1(20-23) A0(20-23) A0(20-23) A1(20-23) A1(20-23)
+                const __m512i lhs_mat_23_2_sp2 = _mm512_shuffle_epi32(lhs_mat_23_2, 245);  //A2(20-23) A2(20-23) A3(20-23) A3(20-23) A2(20-23) A2(20-23) A3(20-23) A3(20-23) A2(20-23) A2(20-23) A3(20-23) A3(20-23) A2(20-23) A2(20-23) A3(20-23) A3(20-23)
+
+                const __m512i lhs_mat_01_3_sp2 = _mm512_shuffle_epi32(lhs_mat_01_3, 245);  //A0(28-31) A0(28-31) A1(28-31) A1(28-31) A0(28-31) A0(28-31) A1(28-31) A1(28-31) A0(28-31) A0(28-31) A1(28-31) A1(28-31) A0(28-31) A0(28-31) A1(28-31) A1(28-31)
+                const __m512i lhs_mat_23_3_sp2 = _mm512_shuffle_epi32(lhs_mat_23_3, 245);  //A2(28-31) A2(28-31) A3(28-31) A3(28-31) A2(28-31) A2(28-31) A3(28-31) A3(28-31) A2(28-31) A2(28-31) A3(28-31) A3(28-31) A2(28-31) A2(28-31) A3(28-31) A3(28-31)
+
+                // The values arranged in shuffle patterns are operated with dot product operation within 32 bit lane i.e corresponding bytes and multiplied and added into 32 bit integers within 32 bit lane
+                // Resembles MMLAs into 2x2 matrices in ARM Version
+                __m512i iacc_mat_00_sp1 =
+                    _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_01_3_sp1, rhs_mat_014589CD_3_sp1), mul_sum_i8_pairs_int32x16(lhs_mat_01_2_sp1, rhs_mat_014589CD_2_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_01_1_sp1, rhs_mat_014589CD_1_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_01_0_sp1, rhs_mat_014589CD_0_sp1));
+                __m512i iacc_mat_01_sp1 =
+                    _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_01_3_sp1, rhs_mat_2367ABEF_3_sp1), mul_sum_i8_pairs_int32x16(lhs_mat_01_2_sp1, rhs_mat_2367ABEF_2_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_01_1_sp1, rhs_mat_2367ABEF_1_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_01_0_sp1, rhs_mat_2367ABEF_0_sp1));
+                __m512i iacc_mat_10_sp1 =
+                    _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_23_3_sp1, rhs_mat_014589CD_3_sp1), mul_sum_i8_pairs_int32x16(lhs_mat_23_2_sp1, rhs_mat_014589CD_2_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_23_1_sp1, rhs_mat_014589CD_1_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_23_0_sp1, rhs_mat_014589CD_0_sp1));
+                __m512i iacc_mat_11_sp1 =
+                    _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_23_3_sp1, rhs_mat_2367ABEF_3_sp1), mul_sum_i8_pairs_int32x16(lhs_mat_23_2_sp1, rhs_mat_2367ABEF_2_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_23_1_sp1, rhs_mat_2367ABEF_1_sp1)), mul_sum_i8_pairs_int32x16(lhs_mat_23_0_sp1, rhs_mat_2367ABEF_0_sp1));
+                __m512i iacc_mat_00_sp2 =
+                    _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_01_3_sp2, rhs_mat_014589CD_3_sp2), mul_sum_i8_pairs_int32x16(lhs_mat_01_2_sp2, rhs_mat_014589CD_2_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_01_1_sp2, rhs_mat_014589CD_1_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_01_0_sp2, rhs_mat_014589CD_0_sp2));
+                __m512i iacc_mat_01_sp2 =
+                    _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_01_3_sp2, rhs_mat_2367ABEF_3_sp2), mul_sum_i8_pairs_int32x16(lhs_mat_01_2_sp2, rhs_mat_2367ABEF_2_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_01_1_sp2, rhs_mat_2367ABEF_1_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_01_0_sp2, rhs_mat_2367ABEF_0_sp2));
+                __m512i iacc_mat_10_sp2 =
+                    _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_23_3_sp2, rhs_mat_014589CD_3_sp2), mul_sum_i8_pairs_int32x16(lhs_mat_23_2_sp2, rhs_mat_014589CD_2_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_23_1_sp2, rhs_mat_014589CD_1_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_23_0_sp2, rhs_mat_014589CD_0_sp2));
+                __m512i iacc_mat_11_sp2 =
+                    _mm512_add_epi32(_mm512_add_epi32(_mm512_add_epi32(mul_sum_i8_pairs_int32x16(lhs_mat_23_3_sp2, rhs_mat_2367ABEF_3_sp2), mul_sum_i8_pairs_int32x16(lhs_mat_23_2_sp2, rhs_mat_2367ABEF_2_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_23_1_sp2, rhs_mat_2367ABEF_1_sp2)), mul_sum_i8_pairs_int32x16(lhs_mat_23_0_sp2, rhs_mat_2367ABEF_0_sp2));
+
+                // Output of both shuffle patterns are added in order to sum dot product outputs of all 32 values in block
+                __m512i iacc_mat_00 = _mm512_add_epi32(iacc_mat_00_sp1, iacc_mat_00_sp2);
+                __m512i iacc_mat_01 = _mm512_add_epi32(iacc_mat_01_sp1, iacc_mat_01_sp2);
+                __m512i iacc_mat_10 = _mm512_add_epi32(iacc_mat_10_sp1, iacc_mat_10_sp2);
+                __m512i iacc_mat_11 = _mm512_add_epi32(iacc_mat_11_sp1, iacc_mat_11_sp2);
+
+
+                // Straighten out to make 4 row vectors
+                __m512i iacc_row_0 = _mm512_mask_blend_epi32(0xCCCC, iacc_mat_00, _mm512_shuffle_epi32(iacc_mat_01, 78));
+                __m512i iacc_row_1 = _mm512_mask_blend_epi32(0xCCCC, _mm512_shuffle_epi32(iacc_mat_00, 78), iacc_mat_01);
+                __m512i iacc_row_2 = _mm512_mask_blend_epi32(0xCCCC, iacc_mat_10, _mm512_shuffle_epi32(iacc_mat_11, 78));
+                __m512i iacc_row_3 = _mm512_mask_blend_epi32(0xCCCC, _mm512_shuffle_epi32(iacc_mat_10, 78), iacc_mat_11);
+
+                // Load the scale(d) values for all the 4 Q8_0 blocks and repeat it across lanes
+                const __m128i row_scale_f16 = _mm_shuffle_epi32(_mm_maskload_epi32((int const*)(a_ptr[b].d), loadMask), 68);
+                const __m512 row_scale_f32 = GGML_F32Cx16_REPEAT_LOAD(row_scale_f16);
+
+                // Multiply with appropiate scales and accumulate
+                acc_rows[0] = _mm512_fmadd_ps(_mm512_cvtepi32_ps(iacc_row_0), _mm512_mul_ps(col_scale_f32, _mm512_shuffle_ps(row_scale_f32, row_scale_f32, 0)),   acc_rows[0]);
+                acc_rows[1] = _mm512_fmadd_ps(_mm512_cvtepi32_ps(iacc_row_1), _mm512_mul_ps(col_scale_f32, _mm512_shuffle_ps(row_scale_f32, row_scale_f32, 85)),  acc_rows[1]);
+                acc_rows[2] = _mm512_fmadd_ps(_mm512_cvtepi32_ps(iacc_row_2), _mm512_mul_ps(col_scale_f32, _mm512_shuffle_ps(row_scale_f32, row_scale_f32, 170)), acc_rows[2]);
+                acc_rows[3] = _mm512_fmadd_ps(_mm512_cvtepi32_ps(iacc_row_3), _mm512_mul_ps(col_scale_f32, _mm512_shuffle_ps(row_scale_f32, row_scale_f32, 255)), acc_rows[3]);
+            }
+
+            // Store the accumulated values
+            for (int i = 0; i < 4; i++) {
+                _mm512_storeu_ps((float *)(s + ((y * 4 + i) * bs + x * 8)), acc_rows[i]);
+            }
+        }
+    }
+    if (anc != nc) {
+        xstart = anc/8;
+        y = 0;
+    }
+#endif // __AVX512F__
 
     // Take group of four block_q8_0x4 structures at each pass of the loop and perform dot product operation
-    int anr = nr - nr %16; // Used to align nr with boundary of 16
 
     for (; y < anr / 4; y += 4) {
         const block_q8_0x4 * a_ptrs[4];
@@ -2435,7 +2904,7 @@ void ggml_gemm_q4_0_8x8_q8_0(int n, float * restrict s, size_t bs, const void *
         }
 
         // Take group of eight block_q4_0x8 structures at each pass of the loop and perform dot product operation
-        for (int64_t x = 0; x < nc / 8; x++) {
+        for (int64_t x = xstart; x < nc / 8; x++) {
 
             const block_q4_0x8 * b_ptr = b_ptr_start + (x * b_nb);
 
@@ -2547,21 +3016,21 @@ void ggml_gemm_q4_0_8x8_q8_0(int n, float * restrict s, size_t bs, const void *
                     // The values arranged in shuffle patterns are operated with dot product operation within 32 bit lane i.e corresponding bytes and multiplied and added into 32 bit integers within 32 bit lane
                     // Resembles MMLAs into 2x2 matrices in ARM Version
                     __m256i iacc_mat_00_sp1 =
-                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_01_3_sp1, rhs_mat_0145_3_sp1), mul_sum_i8_pairs_int(lhs_mat_01_2_sp1, rhs_mat_0145_2_sp1)), mul_sum_i8_pairs_int(lhs_mat_01_1_sp1, rhs_mat_0145_1_sp1)), mul_sum_i8_pairs_int(lhs_mat_01_0_sp1, rhs_mat_0145_0_sp1));
+                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_01_3_sp1, rhs_mat_0145_3_sp1), mul_sum_i8_pairs_int32x8(lhs_mat_01_2_sp1, rhs_mat_0145_2_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_01_1_sp1, rhs_mat_0145_1_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_01_0_sp1, rhs_mat_0145_0_sp1));
                     __m256i iacc_mat_01_sp1 =
-                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_01_3_sp1, rhs_mat_2367_3_sp1), mul_sum_i8_pairs_int(lhs_mat_01_2_sp1, rhs_mat_2367_2_sp1)), mul_sum_i8_pairs_int(lhs_mat_01_1_sp1, rhs_mat_2367_1_sp1)), mul_sum_i8_pairs_int(lhs_mat_01_0_sp1, rhs_mat_2367_0_sp1));
+                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_01_3_sp1, rhs_mat_2367_3_sp1), mul_sum_i8_pairs_int32x8(lhs_mat_01_2_sp1, rhs_mat_2367_2_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_01_1_sp1, rhs_mat_2367_1_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_01_0_sp1, rhs_mat_2367_0_sp1));
                     __m256i iacc_mat_10_sp1 =
-                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_23_3_sp1, rhs_mat_0145_3_sp1), mul_sum_i8_pairs_int(lhs_mat_23_2_sp1, rhs_mat_0145_2_sp1)), mul_sum_i8_pairs_int(lhs_mat_23_1_sp1, rhs_mat_0145_1_sp1)), mul_sum_i8_pairs_int(lhs_mat_23_0_sp1, rhs_mat_0145_0_sp1));
+                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_23_3_sp1, rhs_mat_0145_3_sp1), mul_sum_i8_pairs_int32x8(lhs_mat_23_2_sp1, rhs_mat_0145_2_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_23_1_sp1, rhs_mat_0145_1_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_23_0_sp1, rhs_mat_0145_0_sp1));
                     __m256i iacc_mat_11_sp1 =
-                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_23_3_sp1, rhs_mat_2367_3_sp1), mul_sum_i8_pairs_int(lhs_mat_23_2_sp1, rhs_mat_2367_2_sp1)), mul_sum_i8_pairs_int(lhs_mat_23_1_sp1, rhs_mat_2367_1_sp1)), mul_sum_i8_pairs_int(lhs_mat_23_0_sp1, rhs_mat_2367_0_sp1));
+                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_23_3_sp1, rhs_mat_2367_3_sp1), mul_sum_i8_pairs_int32x8(lhs_mat_23_2_sp1, rhs_mat_2367_2_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_23_1_sp1, rhs_mat_2367_1_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_23_0_sp1, rhs_mat_2367_0_sp1));
                     __m256i iacc_mat_00_sp2 =
-                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_01_3_sp2, rhs_mat_0145_3_sp2), mul_sum_i8_pairs_int(lhs_mat_01_2_sp2, rhs_mat_0145_2_sp2)), mul_sum_i8_pairs_int(lhs_mat_01_1_sp2, rhs_mat_0145_1_sp2)), mul_sum_i8_pairs_int(lhs_mat_01_0_sp2, rhs_mat_0145_0_sp2));
+                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_01_3_sp2, rhs_mat_0145_3_sp2), mul_sum_i8_pairs_int32x8(lhs_mat_01_2_sp2, rhs_mat_0145_2_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_01_1_sp2, rhs_mat_0145_1_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_01_0_sp2, rhs_mat_0145_0_sp2));
                     __m256i iacc_mat_01_sp2 =
-                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_01_3_sp2, rhs_mat_2367_3_sp2), mul_sum_i8_pairs_int(lhs_mat_01_2_sp2, rhs_mat_2367_2_sp2)), mul_sum_i8_pairs_int(lhs_mat_01_1_sp2, rhs_mat_2367_1_sp2)), mul_sum_i8_pairs_int(lhs_mat_01_0_sp2, rhs_mat_2367_0_sp2));
+                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_01_3_sp2, rhs_mat_2367_3_sp2), mul_sum_i8_pairs_int32x8(lhs_mat_01_2_sp2, rhs_mat_2367_2_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_01_1_sp2, rhs_mat_2367_1_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_01_0_sp2, rhs_mat_2367_0_sp2));
                     __m256i iacc_mat_10_sp2 =
-                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_23_3_sp2, rhs_mat_0145_3_sp2), mul_sum_i8_pairs_int(lhs_mat_23_2_sp2, rhs_mat_0145_2_sp2)), mul_sum_i8_pairs_int(lhs_mat_23_1_sp2, rhs_mat_0145_1_sp2)), mul_sum_i8_pairs_int(lhs_mat_23_0_sp2, rhs_mat_0145_0_sp2));
+                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_23_3_sp2, rhs_mat_0145_3_sp2), mul_sum_i8_pairs_int32x8(lhs_mat_23_2_sp2, rhs_mat_0145_2_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_23_1_sp2, rhs_mat_0145_1_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_23_0_sp2, rhs_mat_0145_0_sp2));
                     __m256i iacc_mat_11_sp2 =
-                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_23_3_sp2, rhs_mat_2367_3_sp2), mul_sum_i8_pairs_int(lhs_mat_23_2_sp2, rhs_mat_2367_2_sp2)), mul_sum_i8_pairs_int(lhs_mat_23_1_sp2, rhs_mat_2367_1_sp2)), mul_sum_i8_pairs_int(lhs_mat_23_0_sp2, rhs_mat_2367_0_sp2));
+                        _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_23_3_sp2, rhs_mat_2367_3_sp2), mul_sum_i8_pairs_int32x8(lhs_mat_23_2_sp2, rhs_mat_2367_2_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_23_1_sp2, rhs_mat_2367_1_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_23_0_sp2, rhs_mat_2367_0_sp2));
 
                     // Output of both shuffle patterns are added in order to sum dot product outputs of all 32 values in block
                     __m256i iacc_mat_00 = _mm256_add_epi32(iacc_mat_00_sp1, iacc_mat_00_sp2);
@@ -2599,7 +3068,7 @@ void ggml_gemm_q4_0_8x8_q8_0(int n, float * restrict s, size_t bs, const void *
         const block_q8_0x4 * a_ptr = a_ptr_start + (y * nb);
 
         // Load the eight block_q4_0 quantized values interleaved with each other in chunks of eight - B0,B1 ....B6,B7
-        for (int64_t x = 0; x < nc / 8; x++) {
+        for (int64_t x = xstart; x < nc / 8; x++) {
 
             const block_q4_0x8 * b_ptr = b_ptr_start + (x * b_nb);
 
@@ -2711,21 +3180,21 @@ void ggml_gemm_q4_0_8x8_q8_0(int n, float * restrict s, size_t bs, const void *
                 // The values arranged in shuffle patterns are operated with dot product operation within 32 bit lane i.e corresponding bytes and multiplied and added into 32 bit integers within 32 bit lane
                 // Resembles MMLAs into 2x2 matrices in ARM Version
                 __m256i iacc_mat_00_sp1 =
-                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_01_3_sp1, rhs_mat_0145_3_sp1), mul_sum_i8_pairs_int(lhs_mat_01_2_sp1, rhs_mat_0145_2_sp1)), mul_sum_i8_pairs_int(lhs_mat_01_1_sp1, rhs_mat_0145_1_sp1)), mul_sum_i8_pairs_int(lhs_mat_01_0_sp1, rhs_mat_0145_0_sp1));
+                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_01_3_sp1, rhs_mat_0145_3_sp1), mul_sum_i8_pairs_int32x8(lhs_mat_01_2_sp1, rhs_mat_0145_2_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_01_1_sp1, rhs_mat_0145_1_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_01_0_sp1, rhs_mat_0145_0_sp1));
                 __m256i iacc_mat_01_sp1 =
-                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_01_3_sp1, rhs_mat_2367_3_sp1), mul_sum_i8_pairs_int(lhs_mat_01_2_sp1, rhs_mat_2367_2_sp1)), mul_sum_i8_pairs_int(lhs_mat_01_1_sp1, rhs_mat_2367_1_sp1)), mul_sum_i8_pairs_int(lhs_mat_01_0_sp1, rhs_mat_2367_0_sp1));
+                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_01_3_sp1, rhs_mat_2367_3_sp1), mul_sum_i8_pairs_int32x8(lhs_mat_01_2_sp1, rhs_mat_2367_2_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_01_1_sp1, rhs_mat_2367_1_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_01_0_sp1, rhs_mat_2367_0_sp1));
                 __m256i iacc_mat_10_sp1 =
-                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_23_3_sp1, rhs_mat_0145_3_sp1), mul_sum_i8_pairs_int(lhs_mat_23_2_sp1, rhs_mat_0145_2_sp1)), mul_sum_i8_pairs_int(lhs_mat_23_1_sp1, rhs_mat_0145_1_sp1)), mul_sum_i8_pairs_int(lhs_mat_23_0_sp1, rhs_mat_0145_0_sp1));
+                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_23_3_sp1, rhs_mat_0145_3_sp1), mul_sum_i8_pairs_int32x8(lhs_mat_23_2_sp1, rhs_mat_0145_2_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_23_1_sp1, rhs_mat_0145_1_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_23_0_sp1, rhs_mat_0145_0_sp1));
                 __m256i iacc_mat_11_sp1 =
-                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_23_3_sp1, rhs_mat_2367_3_sp1), mul_sum_i8_pairs_int(lhs_mat_23_2_sp1, rhs_mat_2367_2_sp1)), mul_sum_i8_pairs_int(lhs_mat_23_1_sp1, rhs_mat_2367_1_sp1)), mul_sum_i8_pairs_int(lhs_mat_23_0_sp1, rhs_mat_2367_0_sp1));
+                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_23_3_sp1, rhs_mat_2367_3_sp1), mul_sum_i8_pairs_int32x8(lhs_mat_23_2_sp1, rhs_mat_2367_2_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_23_1_sp1, rhs_mat_2367_1_sp1)), mul_sum_i8_pairs_int32x8(lhs_mat_23_0_sp1, rhs_mat_2367_0_sp1));
                 __m256i iacc_mat_00_sp2 =
-                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_01_3_sp2, rhs_mat_0145_3_sp2), mul_sum_i8_pairs_int(lhs_mat_01_2_sp2, rhs_mat_0145_2_sp2)), mul_sum_i8_pairs_int(lhs_mat_01_1_sp2, rhs_mat_0145_1_sp2)), mul_sum_i8_pairs_int(lhs_mat_01_0_sp2, rhs_mat_0145_0_sp2));
+                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_01_3_sp2, rhs_mat_0145_3_sp2), mul_sum_i8_pairs_int32x8(lhs_mat_01_2_sp2, rhs_mat_0145_2_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_01_1_sp2, rhs_mat_0145_1_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_01_0_sp2, rhs_mat_0145_0_sp2));
                 __m256i iacc_mat_01_sp2 =
-                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_01_3_sp2, rhs_mat_2367_3_sp2), mul_sum_i8_pairs_int(lhs_mat_01_2_sp2, rhs_mat_2367_2_sp2)), mul_sum_i8_pairs_int(lhs_mat_01_1_sp2, rhs_mat_2367_1_sp2)), mul_sum_i8_pairs_int(lhs_mat_01_0_sp2, rhs_mat_2367_0_sp2));
+                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_01_3_sp2, rhs_mat_2367_3_sp2), mul_sum_i8_pairs_int32x8(lhs_mat_01_2_sp2, rhs_mat_2367_2_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_01_1_sp2, rhs_mat_2367_1_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_01_0_sp2, rhs_mat_2367_0_sp2));
                 __m256i iacc_mat_10_sp2 =
-                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_23_3_sp2, rhs_mat_0145_3_sp2), mul_sum_i8_pairs_int(lhs_mat_23_2_sp2, rhs_mat_0145_2_sp2)), mul_sum_i8_pairs_int(lhs_mat_23_1_sp2, rhs_mat_0145_1_sp2)), mul_sum_i8_pairs_int(lhs_mat_23_0_sp2, rhs_mat_0145_0_sp2));
+                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_23_3_sp2, rhs_mat_0145_3_sp2), mul_sum_i8_pairs_int32x8(lhs_mat_23_2_sp2, rhs_mat_0145_2_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_23_1_sp2, rhs_mat_0145_1_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_23_0_sp2, rhs_mat_0145_0_sp2));
                 __m256i iacc_mat_11_sp2 =
-                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int(lhs_mat_23_3_sp2, rhs_mat_2367_3_sp2), mul_sum_i8_pairs_int(lhs_mat_23_2_sp2, rhs_mat_2367_2_sp2)), mul_sum_i8_pairs_int(lhs_mat_23_1_sp2, rhs_mat_2367_1_sp2)), mul_sum_i8_pairs_int(lhs_mat_23_0_sp2, rhs_mat_2367_0_sp2));
+                    _mm256_add_epi32(_mm256_add_epi32(_mm256_add_epi32(mul_sum_i8_pairs_int32x8(lhs_mat_23_3_sp2, rhs_mat_2367_3_sp2), mul_sum_i8_pairs_int32x8(lhs_mat_23_2_sp2, rhs_mat_2367_2_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_23_1_sp2, rhs_mat_2367_1_sp2)), mul_sum_i8_pairs_int32x8(lhs_mat_23_0_sp2, rhs_mat_2367_0_sp2));
 
                 // Output of both shuffle patterns are added in order to sum dot product outputs of all 32 values in block
                 __m256i iacc_mat_00 = _mm256_add_epi32(iacc_mat_00_sp1, iacc_mat_00_sp2);