MCPcopy Create free account
hub / github.com/NVIDIA/cuda-samples / run_sample

Function run_sample

python/2_CoreConcepts/blockwiseSum/blockwiseSum.py:112–245  ·  view source on GitHub ↗

Run block-wise sum demonstration. Parameters ---------- num_elements : int Number of array elements device_id : int CUDA device ID Returns ------- bool True if all tests passed

(num_elements: int = 1024 * 1024, device_id: int = 0)

Source from the content-addressed store, hash-verified

110
111
112def run_sample(num_elements: int = 1024 * 1024, device_id: int = 0) -> bool:
113 """
114 Run block-wise sum demonstration.
115
116 Parameters
117 ----------
118 num_elements : int
119 Number of array elements
120 device_id : int
121 CUDA device ID
122
123 Returns
124 -------
125 bool
126 True if all tests passed
127 """
128 threads_per_block = 256
129 num_blocks = 64
130
131 device = Device(device_id)
132 device.set_current()
133 stream = device.create_stream()
134
135 arch = f"sm_{device.arch}"
136 print(f"Device: {device.name}")
137 print(f"Compute Capability: {arch}")
138 print(f"Array size: {num_elements:,} elements\n")
139
140 try:
141 # Make CuPy use our stream
142 cp.cuda.Stream.from_external(stream).use()
143
144 # Compile kernels
145 program = Program(
146 KERNELS_CODE, code_type="c++", options=ProgramOptions(arch=arch)
147 )
148 module = program.compile(target_type="cubin")
149 kernel_simple = module.get_kernel("simple_indexing")
150 kernel_strided = module.get_kernel("strided_loop")
151 kernel_sum = module.get_kernel("block_partial_sum")
152
153 # Test data
154 h_input = np.arange(num_elements, dtype=np.float32)
155 d_input = cp.asarray(h_input)
156 d_output = cp.zeros_like(d_input)
157 expected = cp.asarray(h_input * 2.0)
158
159 # Demo 1: Simple indexing (1 thread = 1 element)
160 full_blocks = (num_elements + threads_per_block - 1) // threads_per_block
161 config = LaunchConfig(grid=full_blocks, block=threads_per_block)
162 launch(
163 stream,
164 config,
165 kernel_simple,
166 d_input.data.ptr,
167 d_output.data.ptr,
168 cp.uint64(num_elements),
169 )

Callers 1

mainFunction · 0.85

Calls 2

verify_array_resultFunction · 0.90
fullMethod · 0.45

Tested by

no test coverage detected