Jokeren

import torch
import time
import sys


def run(nelems, iters):
    # Check if CUDA is available
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

    tensor_a = torch.randn(nelems, dtype=torch.float32, device=device)

Jokeren

#blocked = #triton_gpu.blocked<{sizePerThread = [1, 8], threadsPerWarp = [4, 8], warpsPerCTA = [4, 1], order = [1, 0]}>
#blocked1 = #triton_gpu.blocked<{sizePerThread = [1, 8], threadsPerWarp = [1, 32], warpsPerCTA = [4, 1], order = [1, 0]}>
#blocked2 = #triton_gpu.blocked<{sizePerThread = [1, 2], threadsPerWarp = [1, 32], warpsPerCTA = [1, 4], order = [1, 0]}>
#mma = #triton_gpu.nvidia_mma<{versionMajor = 2, versionMinor = 0, warpsPerCTA = [1, 4], instrShape = [16, 8]}>
#shared = #triton_gpu.shared<{vec = 1, perPhase = 1, maxPhase = 1, order = [0, 1], hasLeadingOffset = false}>
#shared1 = #triton_gpu.shared<{vec = 1, perPhase = 1, maxPhase = 1, order = [1, 0], hasLeadingOffset = false}>
module attributes {"triton_gpu.num-ctas" = 1 : i32, "triton_gpu.num-warps" = 4 : i32, triton_gpu.target = "cuda:80", "triton_gpu.threads-per-warp" = 32 : i32} {
  tt.func public @hoist_convert_above_extf_and_remat(%arg0: !tt.ptr<f16> {tt.divisibility = 16 : i32}, %arg1: !tt.ptr<f16> {tt.divisibility = 16 : i32}, %arg2: !tt.pt

Jokeren

; ModuleID = 'LLVMDialectModule'
source_filename = "LLVMDialectModule"
target triple = "amdgcn-amd-amdhsa"

@global_smem = external local_unnamed_addr addrspace(3) global [0 x i8], align 16

; Function Attrs: mustprogress nofree norecurse nounwind willreturn
define amdgpu_kernel void @flip_kernel(ptr addrspace(1) nocapture readonly %0, ptr addrspace(1) nocapture writeonly %1) local_unnamed_addr #0 !dbg !4 {
  %3 = tail call i32 @llvm.amdgcn.workitem.id.x(), !dbg !7
  %4 = shl i32 %3, 2, !dbg !8