Switch to Triton FP8 Quantization in EMU1.6

Summary:
For some reason, the cuda `quantize_fp8_per_row` kernel is very slow in EMU. Switching it to the functionally equivalent triton kernel yields excellent speedups from FP8. For eager mode, I'm seeing a 20% e2e speedup and still getting proper outputs.

Eager:
BF16: 19702.10ms
FP8 Triton Quant: 16466.97ms

Compiled:
FP8 Native Quant: 14605.18ms
FP8 Triton Quant: 16043.92ms
BF16: 18030.98ms

We see that quantizing in native pytorch helps quite a bit when torch.compile is used. I added the option to choose which quantization function is used and default to triton when torch.compile is off and native torch when torch.compile is on. This gives us the best performance in either case.

Reviewed By: jiawenliu64

Differential Revision: D58167756

Author: jwfromm

fbgemm_gpu/experimental/gemm/test/fp8_gemm_test.py

@@ -50,7 +50,7 @@ def _test_quantize_fp8_row(
             )
 
             # Undo scaling.
-            a_torch = a_fp8.base.to(torch.bfloat16)
+            a_torch = a_fp8.to(torch.bfloat16)
             a_torch *= a_scale[:, None]
 
             self.assertTrue(
@@ -110,7 +110,7 @@ def _test_quantize_fp8_block(
 
             a_fp8, a_scale = quantize_fp8_block(a, BLOCK_M, BLOCK_K, scale_ub=scale_ub)
 
-            a_torch = a_fp8.base.to(torch.bfloat16)
+            a_torch = a_fp8.to(torch.bfloat16)
 
             # Undo scaling.
             for i in range(0, M, BLOCK_M):