1
0
mirror of https://github.com/fumiama/gozel.git synced 2026-06-05 00:10:24 +08:00
Files
gozel/examples/vadd_event/README.md
2026-03-29 17:12:56 +08:00

62 lines
2.6 KiB
Markdown

# Vector Addition — Immediate Command List with Events
> [!Note]
> **SYCL** is used to write this kernel, which is not a common practice.
> Please also have a look at the **OpenCL** kernel examples like [image_scale](../image_scale/).
The same vector addition workload as the `vadd` example, but driven by an **immediate command list** and **events** instead of explicit command queues. This demonstrates fine-grained dependency tracking: memory copies signal events, and the kernel launch waits on those events before executing.
## What It Does
1. Discovers a GPU device and prints its basic & compute properties
2. Allocates host and device memory for two float32 vectors (256 MiB each)
3. Fills both vectors with random values
4. Loads a SPIR-V kernel (`vector_add`) that computes `a[i] += b[i]` in parallel
5. Creates an **event pool** with 3 events to express data-flow dependencies
6. Submits all work through a single **immediate command list**:
- Two H→D copies, each signaling its own event
- Kernel launch that **waits** on both copy events before executing
- D→H copy that waits on the kernel event
7. Synchronizes via `HostSynchronize` on the immediate command list
8. Validates every element against the CPU reference
## Key Difference from `vadd`
| Aspect | `vadd` | `vadd_event` |
|--------|--------|-------------|
| Submission | 3 separate command lists executed on a command queue | 1 immediate command list |
| Synchronization | `zeCommandQueueSynchronize` | `zeCommandListHostSynchronize` |
| Dependencies | Implicit via command list ordering + barriers | Explicit via events (wait lists) |
## Run
```bash
go run main.go
```
## Sample Output
```
=============== Device Basic Properties ===============
Running on device: ID = 32103 , Name = Intel(R) Graphics @ 0.00 GHz.
=============== Device Compute Properties ===============
Max Group Size (X, Y, Z): (1024, 1024, 1024)
Max Group Count (X, Y, Z): (4294967295, 4294967295, 4294967295)
Max Total Group Size: 1024
Max Shared Local Memory: 65536
Num Subgroup Sizes: 3
Subgroup Sizes: [8 16 32 0 0 0 0 0]
=============== Computation Configuration ===============
Group Size (X, Y, Z): (1024, 1, 1)
Group Count: 65536
Total Elements (N): 67108864
Buffer Size: 256 MiB
=============== Calculation Results ===============
GPU Execution Time: 51.768500 ms
GPU Throughput: 5.19 GiB/s
=============== Validation Results ===============
CPU Execution Time: 38.237400 ms
CPU Throughput: 7.02 GiB/s
Test Passed!!!
```