cuda : alternative q4_q8 kernel

ggml-cuda.cu

@@ -274,6 +274,92 @@ template <int block_size> static __global__ void dequantize_mul_mat_q4_0(const v
     }
 }
 
+template <int NT, int NR> static __global__ void dequantize_mul_mat_q4_0_test(const void * vx, const void * vy, float * dst, const int ncols, const int nrows) {
+    const block_q4_0 * x = (const block_q4_0 *) vx;
+    const block_q8_0 * y = (const block_q8_0 *) vy;
+
+    const int bid = blockIdx.x;
+    const int tid = threadIdx.x;
+
+    __shared__ float tmp[NR][NT];
+    for (int i = 0; i < NR; ++i) {
+        tmp[i][tid] = 0.0f;
+    }
+
+    const int nbc = (ncols + 16*NT - 1)/(16*NT);
+    const int nbm = ncols/QK8_0;
+
+    uint64_t xa0;
+    uint64_t xa1;
+
+    const int8_t * xb0 = (const int8_t *) &xa0;
+    const int8_t * xb1 = (const int8_t *) &xa1;
+
+    for (int ibc = 0; ibc < nbc; ++ibc) {
+        const int iyb = (ibc*(16*NT) + 16*tid)/QK8_0;
+        const int iyq = (ibc*(16*NT) + 16*tid)%QK8_0;
+
+        if (iyb >= nbm) {
+            continue;
+        }
+
+        const int8_t * yb = (const int8_t *) &y[iyb].qs[iyq];
+
+        const float dy = y[iyb].d;
+
+        for (int ibr = 0; ibr < NR; ++ibr) {
+            const int ir = bid*NR + ibr;
+            if (ir >= nrows) {
+                continue;
+            }
+
+            // block offset
+            const int ixo = (ir*ncols)/QK4_0 + iyb;
+
+            memcpy(&xa0, &x[ixo].qs[iyq/2 + 0], sizeof(uint64_t));
+            xa1 = xa0;
+
+            xa0 = (xa0     ) & 0x0F0F0F0F0F0F0F0F;
+            xa1 = (xa1 >> 4) & 0x0F0F0F0F0F0F0F0F;
+
+            const float dx = x[ixo].d;
+
+            // the (int) cast is probably unnecessary, but just to make sure the result is accumulated in 32 bits
+            tmp[ibr][tid] += (
+                    ((int)(xb0[0] - 8))*yb[0]  + ((int)(xb1[0] - 8))*yb[1]  +
+                    ((int)(xb0[1] - 8))*yb[2]  + ((int)(xb1[1] - 8))*yb[3]  +
+                    ((int)(xb0[2] - 8))*yb[4]  + ((int)(xb1[2] - 8))*yb[5]  +
+                    ((int)(xb0[3] - 8))*yb[6]  + ((int)(xb1[3] - 8))*yb[7]  +
+                    ((int)(xb0[4] - 8))*yb[8]  + ((int)(xb1[4] - 8))*yb[9]  +
+                    ((int)(xb0[5] - 8))*yb[10] + ((int)(xb1[5] - 8))*yb[11] +
+                    ((int)(xb0[6] - 8))*yb[12] + ((int)(xb1[6] - 8))*yb[13] +
+                    ((int)(xb0[7] - 8))*yb[14] + ((int)(xb1[7] - 8))*yb[15]
+                    )*dx*dy;
+        }
+    }
+
+    // reduce
+    __syncthreads();
+
+    for (int s = NT/2; s > 0; s >>= 1) {
+        if (tid < s) {
+            for (int ibr = 0; ibr < NR; ++ibr) {
+                tmp[ibr][tid] += tmp[ibr][tid + s];
+            }
+        }
+        __syncthreads();
+    }
+
+    if (tid == 0) {
+        for (int ibr = 0; ibr < NR; ++ibr) {
+            const int ir = bid*NR + ibr;
+            if (ir < nrows) {
+                dst[ir] = tmp[ibr][0];
+            }
+        }
+    }
+}
+
 static void dequantize_row_q4_0_cuda(const void * vx, float * y, int k, cudaStream_t stream) {
     const int nb = k / QK4_0;
     dequantize_block_q4_0<<<nb, 1, 0, stream>>>(vx, y);
@@ -316,9 +402,14 @@ static void dequantize_mul_mat_q4_0_cuda(const void * vx, const void * y, float
     //     }
     // }
     // dequantize_mul_mat_q4_0<<<nrows, block_size, 0, stream>>>(vx, y, dst, ncols);
-    const int block_size = 32;
-    GGML_ASSERT(ncols % block_size == 0);
-    dequantize_mul_mat_q4_0<block_size><<<nrows, block_size, 0, stream>>>(vx, y, dst, ncols);
+    //const int block_size = 32;
+    //GGML_ASSERT(ncols % block_size == 0);
+    //dequantize_mul_mat_q4_0<block_size><<<nrows, block_size, 0, stream>>>(vx, y, dst, ncols);
+
+    const int NR = 1;  // unroll rows (seems to not help)
+    const int NT = 64; // number of thrads per row
+
+    dequantize_mul_mat_q4_0_test<NT, NR><<<(nrows + NR - 1)/NR, NT, 0, stream>>>(vx, y, dst, ncols, nrows);
 }
 
 // TODO: optimize