submission 67723

- #!POPCORN leaderboard vectorsum_v2
- #!POPCORN gpus B200
- 
- # This is a submission template for popcorn leaderboard 'vectorsum_v2'.
- # Your task is as follows:
- # > Implement a vector sum reduction kernel. This kernel computes the sum of all elements in the input tensor.
- # >
- # > Input: A tensor of shape `(N,)` with values from a normal distribution with mean 0 and variance 1.
- # > Output: A scalar value equal to the sum of all elements in the input tensor.
- # The deadline for this leaderboard is 2025-12-30 00:00:00+00:00
- 
- # You can automatically route this file to specific GPUs by adding a line
- # `#!POPCORN gpus <GPUs>` to the header of this file.
- # Happy hacking!
- 
- from task import input_t, output_t
- from utils import DeterministicContext
- import torch
- import triton
- import triton.language as tl
- import time
- 
- 
- @triton.jit
- def sum_kernel(
-     i_ptr, # input ptr
-     o_ptr, # output ptr
-     n_elem, # total number of elements in input pointer
-     BLOCK_SIZE: tl.constexpr
- ):
-     pid = tl.program_id(0)
- 
-     offsets = (pid * BLOCK_SIZE) + tl.arange(0, BLOCK_SIZE)
-     mask = offsets < n_elem
- 
-     in_v = tl.load(i_ptr + offsets, mask, other=0.0)
-     # sum
-     sum_result = tl.sum(in_v)
-     o_offset = pid * 1
-     tl.store(o_ptr + o_offset, sum_result)
- 
- 
- 
- def custom_kernel(data: input_t) -> output_t:
-     data, output = data
-     # Let's be on the safe side here, and do the reduction in 64 bit
-     # output = data.to(torch.float64).sum().to(torch.float32)
- 
-     n_elem = data.numel()
-     C_BLOCK_SIZE = 1024
-     c_grid_size = triton.cdiv(n_elem, C_BLOCK_SIZE)
-     BLOCK_SIZE = triton.cdiv(C_BLOCK_SIZE, 1)
-     grid_size = int(c_grid_size)
-     output = torch.zeros([grid_size]).cuda()
-     output_result = torch.zeros([1]).cuda()
-     grid = lambda meta: (triton.cdiv(n_elem, meta["BLOCK_SIZE"]), )
-     # start = time.time_ns()
-     sum_kernel[grid](data, output, n_elem, BLOCK_SIZE=BLOCK_SIZE)
-     # end = time.time_ns()
-     # total1 = (end - start)
-     # start = time.time_ns()
-     # result = output.sum()
-     new_block_size = triton.next_power_of_2(grid_size)
-     grid = lambda meta: (triton.cdiv(new_block_size, meta["BLOCK_SIZE"]), )
-     sum_kernel[grid](output, output_result, grid_size, BLOCK_SIZE=new_block_size)
-     # end = time.time_ns()
-     # total2 = (end - start)
- 
-     # per = total2/(total1 + total2)
-     # print(n_elem," -> ", per)
-     # calculate sum of output
+ #!POPCORN leaderboard vectorsum_v2
+ #!POPCORN gpus A100
+  
+ # This is a submission template for popcorn leaderboard 'vectorsum_v2'.
+ # Your task is as follows:
+ # > Implement a vector sum reduction kernel. This kernel computes the sum of all elements in the input tensor.
+ # >
+ # > Input: A tensor of shape `(N,)` with values from a normal distribution with mean 0 and variance 1.
+ # > Output: A scalar value equal to the sum of all elements in the input tensor.
+ # The deadline for this leaderboard is 2025-12-30 00:00:00+00:00
+  
+ # You can automatically route this file to specific GPUs by adding a line
+ # `#!POPCORN gpus <GPUs>` to the header of this file.
+ # Happy hacking!
+  
+ from task import input_t, output_t
+ from utils import DeterministicContext
+ import torch
+ import triton
+ import triton.language as tl
+ import time
+  
+  
+ @triton.jit
+ def sum_kernel(
+     i_ptr, # input ptr
+     o_ptr, # output ptr
+     n_elem, # total number of elements in input pointer
+     BLOCK_SIZE: tl.constexpr
+ ):
+     pid = tl.program_id(0)
+  
+     offsets = (pid * BLOCK_SIZE) + tl.arange(0, BLOCK_SIZE)
+     mask = offsets < n_elem
+  
+     in_v = tl.load(i_ptr + offsets, mask, other=0.0)
+     # sum
+     sum_result = tl.sum(in_v)
+     o_offset = pid * 1
+     tl.store(o_ptr + o_offset, sum_result)
+  
+  
+  
+ def custom_kernel(data: input_t) -> output_t:
+     data, output = data
+     # Let's be on the safe side here, and do the reduction in 64 bit
+     # output = data.to(torch.float64).sum().to(torch.float32)
+  
+     n_elem = data.numel()
+     C_BLOCK_SIZE = 1024
+     c_grid_size = triton.cdiv(n_elem, C_BLOCK_SIZE)
+     BLOCK_SIZE = triton.cdiv(C_BLOCK_SIZE, 1)
+     grid_size = int(c_grid_size)
+     output = torch.zeros([grid_size]).cuda()
+     output_result = torch.zeros([1]).cuda()
+     grid = lambda meta: (triton.cdiv(n_elem, meta["BLOCK_SIZE"]), )
+     # start = time.time_ns()
+     sum_kernel[grid](data, output, n_elem, BLOCK_SIZE=BLOCK_SIZE)
+     # end = time.time_ns()
+     # total1 = (end - start)
+     # start = time.time_ns()
+     # result = output.sum()
+     new_block_size = triton.next_power_of_2(grid_size)
+     grid = lambda meta: (triton.cdiv(new_block_size, meta["BLOCK_SIZE"]), )
+     sum_kernel[grid](output, output_result, grid_size, BLOCK_SIZE=new_block_size)
+     # end = time.time_ns()
+     # total2 = (end - start)
+  
+     # per = total2/(total1 + total2)
+     # print(n_elem," -> ", per)
+     # calculate sum of output
      return output_result[0]
   No newline at end of file