[release/2.6] Update SDPA skip logic for Navi (#2280)

Fixes this one: https://ontrack-internal.amd.com/browse/SWDEV-522391

Author: AmdSampsa

test/inductor/test_aot_inductor.py

@@ -30,6 +30,7 @@
 from torch.testing._internal import common_utils
 from torch.testing._internal.common_cuda import (
     PLATFORM_SUPPORTS_FLASH_ATTENTION,
+    PLATFORM_SUPPORTS_MEM_EFF_ATTENTION,
     SM80OrLater,
     SM90OrLater,
 )
@@ -929,7 +930,10 @@ def forward(self, q, k, v):
     @unittest.skipIf(IS_FBCODE, "Not yet runnable in fbcode")
     @unittest.skipIf(not SM80OrLater, "bfloat16 only supported in sm80+")
     @unittest.skipIf(
-        not PLATFORM_SUPPORTS_FLASH_ATTENTION, "Some archs don't support SDPA"
+        # for archs where this isn't lowered to flash attention, the math
+        # backend will be used and it doesn't work for bfloat16
+        not PLATFORM_SUPPORTS_FLASH_ATTENTION,
+        "Some archs don't support SDPA with bfloat16",
     )
     def test_sdpa_2(self):
         class Model(torch.nn.Module):
@@ -1039,7 +1043,7 @@ def forward(self, x, y):
 
     @skipIfXpu(msg="_scaled_dot_product_flash_attention is not supported on XPU yet")
     @unittest.skipIf(
-        not PLATFORM_SUPPORTS_FLASH_ATTENTION, "Some archs don't support SDPA"
+        not PLATFORM_SUPPORTS_FLASH_ATTENTION, "Some archs don't support flash SDPA"
     )
     def test_fallback_kernel_with_symexpr_output(self):
         if self.device != GPU_TYPE:
@@ -3036,7 +3040,7 @@ def grid(meta):
         )
 
     @unittest.skipIf(
-        not PLATFORM_SUPPORTS_FLASH_ATTENTION, "Some archs don't support SDPA"
+        not PLATFORM_SUPPORTS_MEM_EFF_ATTENTION, "Some archs don't support mem eff SDPA"
     )
     def test_scaled_dot_product_efficient_attention(self):
         if self.device != GPU_TYPE:

test/inductor/test_torchinductor.py

@@ -10611,6 +10611,9 @@ def fn(q, k, v):
         )
 
     @expectedFailureXPU
+    @unittest.skipIf(
+        not PLATFORM_SUPPORTS_MEM_EFF_ATTENTION, "Some archs don't support mem eff SDPA"
+    )
     def test_scaled_dot_product_efficient_attention(self):
         if self.device == "cpu":
             raise unittest.SkipTest(f"requires {GPU_TYPE}")