c10::nullopt -> std::nullopt (#1032)

Differential Revision: D64835967

Pull Request resolved: #1151

torchao/csrc/cuda/tensor_core_tiled_layout/tensor_core_tiled_layout.cu

@@ -73,7 +73,7 @@ template <typename Out_t, int InnerKTiles, int groupSize, bool kDequant = true>
 __global__ void _dequantize_int4_kernel(
     const at::PackedTensorAccessor32<int32_t, 4, at::RestrictPtrTraits> in,
     at::PackedTensorAccessor32<Out_t, 2, at::RestrictPtrTraits> out,
-    at::optional<const at::PackedTensorAccessor32<c10::BFloat16, 3, at::RestrictPtrTraits>> scales_and_zeros = c10::nullopt) 
+    at::optional<const at::PackedTensorAccessor32<c10::BFloat16, 3, at::RestrictPtrTraits>> scales_and_zeros = std::nullopt)
 {
 
   constexpr int32_t kNTileSize = 8;
@@ -85,16 +85,16 @@ __global__ void _dequantize_int4_kernel(
 
   // n dimension that this lane loads from
   auto n0 = nTile * kNTileSize + (t / 4);
-  
+
   // 8 k-tile values, 4 per m16n8k16 mma.sync operand B
   // int32_t ks[8];
   //Only need 4 offsets since TC layout for single tile is 2x2 (2 pairs of 2 contiguous values)
   int32_t ks[4];
 
-  // Store address base offset  
+  // Store address base offset
   auto pOut = &out[n0][0];
-  
-// Unpack 2 k-tiles at a time since min pack size is InnerKTiles = 2    
+
+// Unpack 2 k-tiles at a time since min pack size is InnerKTiles = 2
 #pragma unroll
   for (int innerKTile = 0; innerKTile < InnerKTiles; innerKTile += 2) {
     //Tensor-core layout for m16n8k16 is such that each tile has 2 pairs of 2 contiguous values
@@ -111,15 +111,15 @@ __global__ void _dequantize_int4_kernel(
 
     // inner k-tiles unpack two at a time
     int32_t pack = in[nTile][kOuterTile][t][innerKTile / 2];
-   
+
     if constexpr(kDequant) {
       // static_assert(scales_and_zeros.has_value(), "scales_and_zeros must be set when dequantizing");
       static_assert(std::is_same<Out_t, c10::BFloat16>::value, "Out must be BFloat16 when dequantizing");
       // __nv_bfloat16 v[8];
 
       // // Extract u4, convert to s4 by subtracting by 2 ** nbits / 2, then convert to bfloat16
       bf16x2x4 v_bf16x2x4 = convert_i4x8_to_bf16x2x4(pack);
-      
+
       // All b values within a 16x16 tile should fall within the same q group
       // Hence we load 1 scale and zero per loop
       int qgroup = ks[0] /  groupSize;
@@ -132,7 +132,7 @@ __global__ void _dequantize_int4_kernel(
   #pragma unroll
       for (int i = 0; i < 4; i++) {
         reinterpret_cast<__nv_bfloat162*>(&pOut[ks[i]])[0] = __hfma2(v_bf16x2x4.vals[i], scale2, zero2);
-      }  
+      }
     }
     else {
       static_assert(std::is_same<Out_t, int32_t>::value, "Out must be int32_t when unpacking to int");
@@ -150,8 +150,8 @@ __global__ void _dequantize_int4_kernel(
 
     #pragma unroll
       for (int i = 0; i < 4; ++i) {
-        reinterpret_cast<int2 *>(&pOut[ks[i]])[0] = v_i32x2[i];      
-      }    
+        reinterpret_cast<int2 *>(&pOut[ks[i]])[0] = v_i32x2[i];
+      }
     }
   }
 }
@@ -164,7 +164,7 @@ at::Tensor _dequantize_tensor_core_tiled_layout(
     const at::Tensor& packed_w,
     const at::Tensor& scales_and_zeros,
     int64_t group_size,
-    int64_t innerKTiles) 
+    int64_t innerKTiles)
 {
 
   constexpr int32_t kNTileSize = 8;
@@ -201,7 +201,7 @@ at::Tensor _dequantize_tensor_core_tiled_layout(
 
   auto stream = at::cuda::getCurrentCUDAStream();
   dim3 grid(kSuperTiles, nTiles);
-  
+
 #define RUN_DEQUANT(QGROUPSIZE) \
   do { \
     switch(innerKTiles) { \
@@ -259,7 +259,7 @@ at::Tensor _dequantize_tensor_core_tiled_layout(
 // input is [n / 8][k / (InnerKTiles * 16)][32][innerKTiles / 2]
 at::Tensor _unpack_tensor_core_tiled_layout(
     const at::Tensor& packed_w,
-    int64_t innerKTiles) 
+    int64_t innerKTiles)
 {
 
   c10::cuda::CUDAGuard g(packed_w.device());
@@ -288,7 +288,7 @@ at::Tensor _unpack_tensor_core_tiled_layout(
 
   auto stream = at::cuda::getCurrentCUDAStream();
   dim3 grid(kSuperTiles, nTiles);
-  
+
   if (innerKTiles == 2) {
     _dequantize_int4_kernel<int32_t, 2, 0, false><<<grid, kWarpSize, 0, stream>>>(
         packed_w.packed_accessor32<int32_t, 4, at::RestrictPtrTraits>(),