ggml : 2x faster scalar implementations

Author: ggerganov

ggml.c

@@ -615,7 +615,8 @@ static inline __m128i packNibbles( __m128i bytes1, __m128i bytes2 )
 
 #if __ARM_NEON
 
-static inline const uint8_t * bytes_from_nibbles_64(const int qk, const uint8_t * qs, uint64_t * qd) {
+// TODO: obosolete - will be removed
+static inline const uint8_t * b4_from_nibbles_64(const int qk, const uint8_t * qs, uint64_t * qd) {
     memcpy(qd, qs, qk/2);
 
     for (int l = 0; l < qk/16; ++l) {
@@ -875,14 +876,14 @@ static void quantize_row_q4_0_reference(const float * restrict x, block_q4_0 * r
         uint64_t qs[QK4_0 / 16] = {0};
 
         for (int l = 0; l < qk/2; ++l) {
-            const float v0 = x[i*qk + 0    + l]*id;
-            const float v1 = x[i*qk + qk/2 + l]*id;
+            const float x0 = x[i*qk + 0    + l]*id;
+            const float x1 = x[i*qk + qk/2 + l]*id;
 
-            const uint64_t vi0 = MIN(15, (int8_t)(v0 + 8.5f));
-            const uint64_t vi1 = MIN(15, (int8_t)(v1 + 8.5f));
+            const uint64_t xi0 = MIN(15, (int8_t)(x0 + 8.5f));
+            const uint64_t xi1 = MIN(15, (int8_t)(x1 + 8.5f));
 
-            qs[l/8] |= vi0 << (8*(l & 7));
-            qs[l/8] |= vi1 << (8*(l & 7) + 4);
+            qs[l/8] |= xi0 << (8*(l & 7));
+            qs[l/8] |= xi1 << (8*(l & 7) + 4);
         }
 
         memcpy(y[i].qs, qs, qk/2);
@@ -921,14 +922,14 @@ static void quantize_row_q4_1_reference(const float * restrict x, block_q4_1 * r
         uint64_t qs[QK4_1 / 16] = {0};
 
         for (int l = 0; l < qk/2; ++l) {
-            const float v0 = (x[0    + l] - min)*id;
-            const float v1 = (x[qk/2 + l] - min)*id;
+            const float x0 = (x[0    + l] - min)*id;
+            const float x1 = (x[qk/2 + l] - min)*id;
 
-            const uint64_t vi0 = MIN(15, (int8_t)(v0 + 0.5f));
-            const uint64_t vi1 = MIN(15, (int8_t)(v1 + 0.5f));
+            const uint64_t xi0 = MIN(15, (int8_t)(x0 + 0.5f));
+            const uint64_t xi1 = MIN(15, (int8_t)(x1 + 0.5f));
 
-            qs[l/8] |= vi0 << (8*(l & 7));
-            qs[l/8] |= vi1 << (8*(l & 7) + 4);
+            qs[l/8] |= xi0 << (8*(l & 7));
+            qs[l/8] |= xi1 << (8*(l & 7) + 4);
         }
 
         memcpy(y[i].qs, qs, qk/2);
@@ -968,14 +969,14 @@ static void quantize_row_q4_2_reference(const float * restrict x, block_q4_2 * r
         uint64_t qs[QK4_2 / 16] = {0};
 
         for (int l = 0; l < qk/2; ++l) {
-            const float v0 = x[i*qk + 0    + l]*id;
-            const float v1 = x[i*qk + qk/2 + l]*id;
+            const float x0 = x[i*qk + 0    + l]*id;
+            const float x1 = x[i*qk + qk/2 + l]*id;
 
-            const uint64_t vi0 = MIN(15, (int8_t)(v0 + 8.5f));
-            const uint64_t vi1 = MIN(15, (int8_t)(v1 + 8.5f));
+            const uint64_t xi0 = MIN(15, (int8_t)(x0 + 8.5f));
+            const uint64_t xi1 = MIN(15, (int8_t)(x1 + 8.5f));
 
-            qs[l/8] |= vi0 << (8*(l & 7));
-            qs[l/8] |= vi1 << (8*(l & 7) + 4);
+            qs[l/8] |= xi0 << (8*(l & 7));
+            qs[l/8] |= xi1 << (8*(l & 7) + 4);
         }
 
         memcpy(y[i].qs, qs, qk/2);
@@ -1015,18 +1016,18 @@ static void quantize_row_q5_0_reference(const float * restrict x, block_q5_0 * r
         uint64_t qs[QK5_0 / 16] = {0};
 
         for (int l = 0; l < qk/2; ++l) {
-            const float v0 = x[i*qk + 0    + l]*id;
-            const float v1 = x[i*qk + qk/2 + l]*id;
+            const float x0 = x[i*qk + 0    + l]*id;
+            const float x1 = x[i*qk + qk/2 + l]*id;
 
-            const uint64_t vi0 = MIN(31, (int8_t)(v0 + 16.5f));
-            const uint64_t vi1 = MIN(31, (int8_t)(v1 + 16.5f));
+            const uint64_t xi0 = MIN(31, (int8_t)(x0 + 16.5f));
+            const uint64_t xi1 = MIN(31, (int8_t)(x1 + 16.5f));
 
-            qs[l/8] |= vi0 << (8*(l & 7));
-            qs[l/8] |= vi1 << (8*(l & 7) + 4);
+            qs[l/8] |= xi0 << (8*(l & 7));
+            qs[l/8] |= xi1 << (8*(l & 7) + 4);
 
             // get the 5-th bit and store it in qh at the right position
-            qh |= ((vi0 & 0x10) >> 4) << (l + 0);
-            qh |= ((vi1 & 0x10) >> 4) << (l + qk/2);
+            qh |= ((xi0 & 0x10) >> 4) << (l + 0);
+            qh |= ((xi1 & 0x10) >> 4) << (l + qk/2);
         }
 
         memcpy( y[i].qs,  qs, qk/2);
@@ -1447,15 +1448,15 @@ static void dequantize_row_q4_0(const block_q4_0 * restrict x, float * restrict
 
     const int nb = k / qk;
 
-    uint64_t qs[QK4_0 / 8];
-
     for (int i = 0; i < nb; i++) {
         const float d = x[i].d;
 
-        const uint8_t * qsp = bytes_from_nibbles_64(qk, x[i].qs, qs);
+        for (int j = 0; j < qk/2; ++j) {
+            const int x0 = (x[i].qs[j] & 0xf) - 8;
+            const int x1 = (x[i].qs[j] >>  4) - 8;
 
-        for (int l = 0; l < qk; ++l) {
-            y[i*qk + l] = (qsp[l] - 8)*d;
+            y[i*qk + j + 0   ] = x0*d;
+            y[i*qk + j + qk/2] = x1*d;
         }
     }
 }
@@ -1468,21 +1469,22 @@ static void dequantize_row_q4_1(const block_q4_1 * restrict x, float * restrict
 
     const int nb = k / qk;
 
-    uint64_t qs[QK4_0 / 8];
-
     for (int i = 0; i < nb; i++) {
         const float d = x[i].d;
         const float m = x[i].m;
 
-        const uint8_t * qsp = bytes_from_nibbles_64(qk, x[i].qs, qs);
+        for (int j = 0; j < qk/2; ++j) {
+            const int x0 = (x[i].qs[j] & 0xf);
+            const int x1 = (x[i].qs[j] >>  4);
 
-        for (int l = 0; l < qk; ++l) {
-            y[i*qk + l] = qsp[l]*d + m;
+            y[i*qk + j + 0   ] = x0*d + m;
+            y[i*qk + j + qk/2] = x1*d + m;
         }
     }
 }
 
 static void dequantize_row_q4_2(const block_q4_2 * restrict x, float * restrict y, int k) {
+    // BORKEN !!!
     static const int qk = QK4_2;
 
     assert(qk / 16 == 0);
@@ -1495,7 +1497,7 @@ static void dequantize_row_q4_2(const block_q4_2 * restrict x, float * restrict
     for (int i = 0; i < nb; i++) {
         const float d = GGML_FP16_TO_FP32(x[i].d);
 
-        const uint8_t * qsp = bytes_from_nibbles_64(qk, x[i].qs, qs);
+        const uint8_t * qsp = b4_from_nibbles_64(qk, x[i].qs, qs);
 
         for (int l = 0; l < qk; ++l) {
             y[i*qk + l] = (qsp[l] - 8)*d;
@@ -1511,20 +1513,21 @@ static void dequantize_row_q5_0(const block_q5_0 * restrict x, float * restrict
 
     const int nb = k / qk;
 
-    uint64_t qs[QK5_0 / 8];
-
     for (int i = 0; i < nb; i++) {
         const float d = GGML_FP16_TO_FP32(x[i].d);
 
         uint32_t qh;
         memcpy(&qh, x[i].qh, sizeof(qh));
 
-        const uint8_t * qsp = bytes_from_nibbles_64(qk, x[i].qs, qs);
+        for (int j = 0; j < qk/2; ++j) {
+            const uint8_t xh_0 = ((qh & (1u << (j + 0 ))) >> (j + 0 )) << 4;
+            const uint8_t xh_1 = ((qh & (1u << (j + 16))) >> (j + 12));
 
-        for (int l = 0; l < qk; ++l) {
-            const uint8_t vh = ((qh & (1u << l)) >> l) << 4;
+            const int32_t x0 = ((x[i].qs[j] & 0xf) | xh_0) - 16;
+            const int32_t x1 = ((x[i].qs[j] >>  4) | xh_1) - 16;
 
-            y[i*qk + l] = ((qsp[l] | vh) - 16)*d;
+            y[i*qk + j + 0   ] = x0*d;
+            y[i*qk + j + qk/2] = x1*d;
         }
     }
 }
@@ -2388,17 +2391,16 @@ static void ggml_vec_dot_q4_0_q8_0(const int n, float * restrict s, const void *
     // scalar
     float sumf = 0.0;
 
-    uint64_t qs[QK8_0 / 8];
-
     for (int i = 0; i < nb; i++) {
-        // unpack nibbles into bytes
-        const uint8_t * px = bytes_from_nibbles_64(qk, x[i].qs, qs);
-        const  int8_t * py = y[i].qs;
+        const int8_t * py = y[i].qs;
 
         int sumi = 0;
 
-        for (int j = 0; j < qk; ++j) {
-            sumi += (px[j] - 8) * py[j];
+        for (int j = 0; j < qk/2; ++j) {
+            const int v0 = (x[i].qs[j] & 0xf) - 8;
+            const int v1 = (x[i].qs[j] >>  4) - 8;
+
+            sumi += (v0 * py[j]) + (v1 * py[j + qk/2]);
         }
 
         sumf += (x[i].d*y[i].d)*sumi;
@@ -2513,16 +2515,16 @@ static void ggml_vec_dot_q4_1_q8_1(const int n, float * restrict s, const void *
     // scalar
     float sumf = 0.0;
 
-    uint64_t qs[QK8_1 / 8];
-
     for (int i = 0; i < nb; i++) {
-        const uint8_t * px = bytes_from_nibbles_64(qk, x[i].qs, qs);
-        const  int8_t * py = y[i].qs;
+        const int8_t * py = y[i].qs;
 
         int sumi = 0;
 
-        for (int j = 0; j < qk; ++j) {
-            sumi += px[j]*py[j];
+        for (int j = 0; j < qk/2; ++j) {
+            const int v0 = (x[i].qs[j] & 0xf);
+            const int v1 = (x[i].qs[j] >>  4);
+
+            sumi += (v0 * py[j]) + (v1 * py[j + qk/2]);
         }
 
         sumf += (x[i].d*y[i].d)*sumi + x[i].m*(y[i].s0 + y[i].s1);
@@ -2847,22 +2849,22 @@ static void ggml_vec_dot_q5_0_q8_0(const int n, float * restrict s, const void *
     // scalar
     float sumf = 0.0;
 
-    uint64_t qs[QK8_0 / 8];
-
     for (int i = 0; i < nb; i++) {
-        // unpack nibbles into bytes
-        const uint8_t * px = bytes_from_nibbles_64(qk, x[i].qs, qs);
-        const  int8_t * py = y[i].qs;
+        const int8_t * py = y[i].qs;
 
         uint32_t qh;
         memcpy(&qh, x[i].qh, sizeof(qh));
 
         int sumi = 0;
 
-        for (int j = 0; j < qk; ++j) {
-            const int xh = ((qh & (1u << j)) >> j) << 4;
+        for (int j = 0; j < qk/2; ++j) {
+            const uint8_t xh_0 = ((qh & (1u << (j + 0 ))) >> (j + 0 )) << 4;
+            const uint8_t xh_1 = ((qh & (1u << (j + 16))) >> (j + 12));
+
+            const int32_t x0 = ((x[i].qs[j] & 0xf) | xh_0) - 16;
+            const int32_t x1 = ((x[i].qs[j] >>  4) | xh_1) - 16;
 
-            sumi += ((px[j] | xh) - 16)*py[j];
+            sumi += (x0 * py[j]) + (x1 * py[j + qk/2]);
         }
 
         sumf += (GGML_FP16_TO_FP32(x[i].d)*y[i].d)*sumi;