upstream: add GEMM kernel references from 4 repos for BI-V100 porting

Sources (all CUDA 10.2 compatible, no CUTLASS/Triton dependency):
- leimao/CUDA-GEMM-Optimization: v00-v07, fp16 WMMA variant, double buffered
- siboehm/SGEMM_CUDA: kernel 1-12, warp tiling + double buffering
- wangzyon/NVIDIA_SGEMM_PRACTICE: kernel 1-7
- edtallison/sgemm-cuda: kernel 1-12 (reimplementation with notes)

Key porting issue: ALL kernels hardcode WARPSIZE=32.
BI-V100 has warp_size=64. Need to:
1. Replace all 32U / WARPSIZE constants with 64
2. Adjust warp subtile decomposition (WMITER, WNITER, WSUBM, WSUBN)
3. Adjust shared memory bank conflict avoidance (may have different bank count)
4. Test __shfl_down_sync with mask=0xFFFFFFFFFFFFFFFF (64-bit)

upstream_ref/nvidia_sgemm_practice/utils.cuh

@@ -0,0 +1,42 @@
+#pragma once
+#include <stdio.h>
+#include <stdlib.h>
+#include <time.h>
+#include <unistd.h>
+#include <sys/time.h>
+#include <cuda_runtime.h>
+#include <cublas_v2.h>
+
+/*
+=====================================
+CUDA操作
+=====================================
+*/
+void cudaCheck(cudaError_t error, const char *file, int line); //CUDA错误检查
+void CudaDeviceInfo();                                         // 打印CUDA信息
+
+/*
+=====================================
+矩阵操作
+=====================================
+*/
+void randomize_matrix(float *mat, int N);            // 随机初始化矩阵
+void copy_matrix(float *src, float *dest, int N);    // 复制矩阵
+void print_matrix(const float *A, int M, int N);     // 打印矩阵
+bool verify_matrix(float *mat1, float *mat2, int N); // 验证矩阵
+
+/*
+=====================================
+计时操作
+=====================================
+*/
+float get_current_sec();                        // 获取当前时刻
+float cpu_elapsed_time(float &beg, float &end); // 计算时间差
+
+/*
+=====================================
+kernel操作
+=====================================
+*/
+//调用指定核函数计算矩阵乘法
+void test_kernel(int kernel_num, int m, int n, int k, float alpha, float *A, float *B, float beta, float *C, cublasHandle_t handle);