cuda : fix get_rows when ncols is odd

Author: slaren

Diff for: ggml-cuda.cu

@@ -1721,6 +1721,32 @@ static __global__ void k_get_rows(
     dst_row[iybs + iqs + y_offset] = v.y;
 }
 
+template<typename src0_t, typename dst_t>
+static __global__ void k_get_rows_float(
+            const src0_t * src0, const int32_t * src1, dst_t * dst,
+            int64_t ne00, /*int64_t ne01, int64_t ne02, int64_t ne03,*/
+            /*int64_t ne10, int64_t ne11,*/ int64_t ne12, /*int64_t ne13,*/
+            /*size_t s0,*/ size_t s1, size_t s2, size_t s3,
+            /*size_t nb00,*/ size_t nb01, size_t nb02, size_t nb03,
+            size_t s10, size_t s11, size_t s12/*, size_t s13*/) {
+
+    const int i00 = blockIdx.x*blockDim.x + threadIdx.x;
+    const int i10 = blockDim.y*blockIdx.y + threadIdx.y;
+    const int i11 = (blockIdx.z*blockDim.z + threadIdx.z)/ne12;
+    const int i12 = (blockIdx.z*blockDim.z + threadIdx.z)%ne12;
+
+    if (i00 >= ne00) {
+        return;
+    }
+
+    const int i01 = src1[i10*s10 + i11*s11 + i12*s12];
+
+    dst_t * dst_row = dst + i10*s1 + i11*s2 + i12*s3;
+    const src0_t * src0_row = (const src0_t *)((const char *)src0 + i01*nb01 + i11*nb02 + i12*nb03);
+
+    dst_row[i00] = src0_row[i00];
+}
+
 template <int qk, int qr, dequantize_kernel_t dequantize_kernel, typename dst_t>
 static __global__ void dequantize_block(const void * __restrict__ vx, dst_t * __restrict__ y, const int k) {
     const int i = blockDim.x*blockIdx.x + 2*threadIdx.x;
@@ -5083,6 +5109,8 @@ static void get_rows_cuda(const ggml_tensor * src0, const ggml_tensor * src1, gg
     const size_t s12 = nb12 / ggml_element_size(src1);
     //const size_t s13 = nb13 / ggml_element_size(src1);
 
+    GGML_ASSERT(ne00 % 2 == 0);
+
     k_get_rows<qk, qr, dq><<<block_nums, block_dims, 0, stream>>>(
             src0_dd, src1_dd, dst_dd,
             ne00, /*ne01, ne02, ne03,*/
@@ -5094,6 +5122,38 @@ static void get_rows_cuda(const ggml_tensor * src0, const ggml_tensor * src1, gg
     (void) dst;
 }
 
+template<typename src0_t>
+static void get_rows_cuda_float(const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst,
+                                const src0_t * src0_dd, const int32_t * src1_dd, float * dst_dd, cudaStream_t stream) {
+
+    GGML_TENSOR_BINARY_OP_LOCALS
+
+    const dim3 block_dims(CUDA_GET_ROWS_BLOCK_SIZE, 1, 1);
+    const int block_num_x = (ne00 + CUDA_GET_ROWS_BLOCK_SIZE - 1) / CUDA_GET_ROWS_BLOCK_SIZE;
+    const dim3 block_nums(block_num_x, ne10, ne11*ne12);
+
+    // strides in elements
+    //const size_t s0 = nb0 / ggml_element_size(dst);
+    const size_t s1 = nb1 / ggml_element_size(dst);
+    const size_t s2 = nb2 / ggml_element_size(dst);
+    const size_t s3 = nb3 / ggml_element_size(dst);
+
+    const size_t s10 = nb10 / ggml_element_size(src1);
+    const size_t s11 = nb11 / ggml_element_size(src1);
+    const size_t s12 = nb12 / ggml_element_size(src1);
+    //const size_t s13 = nb13 / ggml_element_size(src1);
+
+    k_get_rows_float<<<block_nums, block_dims, 0, stream>>>(
+            src0_dd, src1_dd, dst_dd,
+            ne00, /*ne01, ne02, ne03,*/
+            /*ne10, ne11,*/ ne12, /*ne13,*/
+            /* s0,*/ s1, s2, s3,
+            /* nb00,*/ nb01, nb02, nb03,
+            s10, s11, s12/*, s13*/);
+
+    (void) dst;
+}
+
 template<float (*bin_op)(const float, const float)>
 struct bin_bcast_cuda {
     template<typename src0_t, typename src1_t, typename dst_t>
@@ -6491,10 +6551,10 @@ static void ggml_cuda_op_get_rows(
 
     switch (src0->type) {
         case GGML_TYPE_F16:
-            get_rows_cuda<1, 1, convert_f16>(src0, src1, dst, src0_d, src1_i32, dst_d, stream);
+            get_rows_cuda_float(src0, src1, dst, (const half *)src0_d, src1_i32, dst_d, stream);
             break;
         case GGML_TYPE_F32:
-            get_rows_cuda<1, 1, convert_f32>(src0, src1, dst, src0_d, src1_i32, dst_d, stream);
+            get_rows_cuda_float(src0, src1, dst, src0_d, src1_i32, dst_d, stream);
             break;
         case GGML_TYPE_Q4_0:
             get_rows_cuda<QK4_0, QR4_0, dequantize_q4_0>(src0, src1, dst, src0_d, src1_i32, dst_d, stream);