已合并
round_size pad 32 byte #28329
zhaoyu65创建于 2025年12月19日
round_size pad 32 byte #28329
已合并
共 2 个文件变更+1-26
| @@ -73,16 +73,6 @@ class Test_roundup_power2_divisions(TestCase): | |||
| 73 | self.assertEqual(allocated_bytes, active_bytes) | 73 | self.assertEqual(allocated_bytes, active_bytes) |
| 74 | self.assertEqual(allocated_bytes, 1024) | 74 | self.assertEqual(allocated_bytes, 1024) |
| 75 | 75 | ||
| 76 | - @SupportedDevices(['Ascend910B', 'Ascend910C']) | ||
| 77 | - def test_without_add_32(self): | ||
| 78 | - torch.npu.memory.empty_cache() | ||
| 79 | - | ||
| 80 | - x = torch.from_numpy(np.random.randn(512).astype(np.uint8)).to("npu") | ||
| 81 | - allocated_bytes = torch.npu.memory_stats()["allocated_bytes.all.current"] | ||
| 82 | - active_bytes = torch.npu.memory_stats()["active_bytes.all.current"] | ||
| 83 | - self.assertEqual(allocated_bytes, active_bytes) | ||
| 84 | - self.assertEqual(allocated_bytes, 512) | ||
| 85 | - | ||
| 86 | 76 | ||
| 87 | if __name__ == '__main__': | 77 | if __name__ == '__main__': |
| 88 | run_tests() | 78 | run_tests() |
| @@ -1992,19 +1992,6 @@ public: | |||
| 1992 | return result; | 1992 | return result; |
| 1993 | } | 1993 | } |
| 1994 | 1994 | ||
| 1995 | - static bool should_apply_legacy_round_padding() | ||
| 1996 | - { | ||
| 1997 | - static bool should_pad = []() -> bool { | ||
| 1998 | - const auto &soc_version = c10_npu::GetSocVersion(); | ||
| 1999 | - const bool disable_padding = | ||
| 2000 | - (soc_version >= c10_npu::SocVersion::Ascend910B1 && | ||
| 2001 | - soc_version <= c10_npu::SocVersion::Ascend910B4_1) || | ||
| 2002 | - (soc_version >= c10_npu::SocVersion::Ascend910_9391); | ||
| 2003 | - return !disable_padding; | ||
| 2004 | - }(); | ||
| 2005 | - return should_pad; | ||
| 2006 | - } | ||
| 2007 | - | ||
| 2008 | // This function takes the size and number of divisions argument and rounds | 1995 | // This function takes the size and number of divisions argument and rounds |
| 2009 | // up the size argument for the nearest power-of-2 division. | 1996 | // up the size argument for the nearest power-of-2 division. |
| 2010 | // For example, if we need to round-up 1200 and number of divisions is 4, | 1997 | // For example, if we need to round-up 1200 and number of divisions is 4, |
| @@ -2035,9 +2022,7 @@ public: | |||
| 2035 | static size_t round_size(size_t size) | 2022 | static size_t round_size(size_t size) |
| 2036 | { | 2023 | { |
| 2037 | constexpr size_t kPadSize = 32; | 2024 | constexpr size_t kPadSize = 32; |
| 2038 | - if (should_apply_legacy_round_padding()) { | 2025 | + size += kPadSize; |
| 2039 | - size += kPadSize; | ||
| 2040 | - } | ||
| 2041 | 2026 | ||
| 2042 | if (size < kMinBlockSize) { | 2027 | if (size < kMinBlockSize) { |
| 2043 | return kMinBlockSize; | 2028 | return kMinBlockSize; |