Run example 07 GEMM all-scatter benchmark across 10 shapes on MI300X

[Copilot]

Benchmark results for examples/07_gemm_all_scatter across 10 GEMM shapes, run on 2x AMD Instinct MI300X (gfx942, 304 SMs/GPU), fp16, BLK_M=256/BLK_N=64/BLK_K=64, 2 ranks.

#	M	N	K	Time (ms)	TFLOPs
1	1024	2048	2048	0.249	34.5
2	2048	2048	2048	0.300	57.3
3	4096	2048	4096	0.469	146.4
4	4096	4096	4096	0.670	205.0
5	4096	8192	8192	1.565	351.3
6	8192	4608	36864	4.745	586.6
7	8192	8192	8192	2.605	422.0
8	16384	4096	4096	2.385	230.5
9	4096	16384	4096	2.726	201.7
10	8192	16384	8192	7.510	292.8

Peak 586.6 TFLOPs on the LLM-representative shape (M=8192, N=4608, K=36864). Larger square shapes plateau around 420 TFLOPs. Performance degrades for non-square layouts with high M or N imbalance.

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.