[CUDA][64-bit indexing] Support 64-bit indexing in distribution_elementwise_grid_stride_kernel (pytorch#141613)

For pytorch#141544
Overhead doesn't seem to be noticeable even on small sizes (e.g., 2**10 elements)

Pull Request resolved: pytorch#141613
Approved by: https://github.com/Skylion007, https://github.com/ngimel

Author: eqy

aten/src/ATen/native/cuda/DistributionTemplates.h

@@ -63,25 +63,25 @@ std::tuple<uint64_t, dim3, dim3> calc_execution_policy(const int64_t total_eleme
 // grid stride loop kernel for distributions
 template<typename accscalar_t, int unroll_factor, typename dist_t, typename transform_t>
 C10_LAUNCH_BOUNDS_2(block_size_bound, grid_size_bound)
-__global__ void distribution_elementwise_grid_stride_kernel(int numel,
+__global__ void distribution_elementwise_grid_stride_kernel(int64_t numel,
                                                             PhiloxCudaState philox_args,
                                                             const dist_t dist_func,
                                                             const transform_t transform_func) {
   auto seeds = at::cuda::philox::unpack(philox_args);
-  int idx = blockIdx.x * blockDim.x + threadIdx.x;
+  int64_t idx = blockIdx.x * blockDim.x + threadIdx.x;
   curandStatePhilox4_32_10_t state;
   curand_init(std::get<0>(seeds),
               idx,
               std::get<1>(seeds),
               &state);
 
-  int rounded_size = ((numel - 1)/(blockDim.x * gridDim.x * unroll_factor)+1) *
+  int64_t rounded_size = ((numel - 1)/(blockDim.x * gridDim.x * unroll_factor)+1) *
       blockDim.x * gridDim.x * unroll_factor;
-  for(int linear_index = idx; linear_index < rounded_size; linear_index += blockDim.x * gridDim.x * unroll_factor) {
+  for(int64_t linear_index = idx; linear_index < rounded_size; linear_index += blockDim.x * gridDim.x * unroll_factor) {
     auto rand = dist_func(&state);
     #pragma unroll
     for (int ii = 0; ii < unroll_factor; ii++) {
-      int li = linear_index + blockDim.x * gridDim.x * ii;
+      int64_t li = linear_index + blockDim.x * gridDim.x * ii;
       if (li < numel) {
         transform_func(li, static_cast<accscalar_t>((&rand.x)[ii]));
       }

test/test_cuda.py

@@ -36,6 +36,7 @@
 )
 from torch.testing._internal.common_device_type import (
     instantiate_device_type_tests,
+    largeTensorTest,
     onlyCUDA,
     onlyNativeDeviceTypes,
 )
@@ -1051,6 +1052,12 @@ def run(dev: torch.device) -> int:
             abs(run(torch.device("cuda")) - run(torch.device("cpu"))) < 10_000
         )
 
+    @largeTensorTest("20GB", "cuda")
+    def test_randint_generation_for_large_numel(self) -> None:
+        numel = 2**31 + 1
+        s = torch.randint(2, (numel,), device="cuda", dtype=torch.int8).sum()
+        self.assertTrue(s > 0, "expected randint in [0, 1] to generate nonzero values")
+
     @parametrize("dtype", [torch.float32, torch.double])
     def test_random_no_reused_random_states(self, dtype: torch.dtype) -> None:
         # Test if random states do not overlap between consecutive rand/randn calls.