even more configs

(cherry picked from commit f97c7a9)
(cherry picked from commit db49466)
(cherry picked from commit 6e9b4ee)
(cherry picked from commit c36d85f)

Author: AmdSampsa

torch/_inductor/runtime/triton_heuristics.py

@@ -1913,6 +1913,9 @@ def pointwise(
                 triton_config_with_settings(
                     size_hints, TRITON_MAX_BLOCK["X"], waves_per_eu=2
                 ),
+                triton_config_with_settings(
+                    size_hints, 4096 # wrt: better than the max_block for some kernel
+                ),
                 *hinted_configs,
             ]
             # Additional reduction configs appended for ROCm builds
@@ -1935,10 +1938,12 @@ def pointwise(
         else:
             configs = [
                 triton_config_with_settings(size_hints, 32, 32),
+                triton_config_with_settings(size_hints, 64, 32), # wrt: better for some kernels
                 triton_config_with_settings(size_hints, 64, 64),  # ~8% better for fp16
                 triton_config_with_settings(size_hints, 256, 16),
                 triton_config_with_settings(size_hints, 16, 256),
                 triton_config_with_settings(size_hints, 128, 16), # wrt: +10% for some kernels
+                triton_config_with_settings(size_hints, 128, 32), # wrt: ..additional 10% more
                 triton_config_with_settings(size_hints, 32, 512), # wrt: +30% for some kernels
                 triton_config_with_settings(size_hints, bs, 1),
                 triton_config_with_settings(size_hints, 1, bs),