Enhance dtype support for the get_value operation. #57272
Conversation
|
你的PR提交成功，感谢你对开源项目的贡献! |
003d578 to
1e8ee36
Compare
00a7739 to
104cfe5
Compare
| self.ndtype = np.float64 | ||
| self.dtype = 'complex128' | ||
|
|
||
|
|
There was a problem hiding this comment.
辛苦补充下简单的case，测试补充的case在反向op下的可行性
5c30706 to
1b4aad3
Compare
|
Sorry to inform you that 1b4aad3's CIs have passed for more than 7 days. To prevent PR conflicts, you need to re-run all CIs manually. |
333a979 to
c2003e5
Compare
test/indexing/test_getitem.py
Outdated
| self.dtype = 'complex128' | ||
|
|
||
|
|
||
| class TestGetitemGard(unittest.TestCase): |
There was a problem hiding this comment.
Gard -> Grad, 等到下个PR一起改过来吧
a1b61b5 to
a4fb546
Compare
|
Sorry to inform you that a4fb546's CIs have passed for more than 7 days. To prevent PR conflicts, you need to re-run all CIs manually. |
a4fb546 to
4b9a08f
Compare
4b9a08f to
cd6cbf8
Compare
| @@ -938,6 +938,8 @@ | |||
| func : gather_nd_grad | |||
| composite : gather_nd_grad(x, index, out_grad, x_grad) | |||
| no_need_buffer : x | |||
| data_transform : | |||
| skip_transform : index | |||
There was a problem hiding this comment.
这里不对index做data_transform，是因为在kernel内部对这个参数做了限制吗？
There was a problem hiding this comment.
这里不对
index做data_transform，是因为在kernel内部对这个参数做了限制吗？
是的，index要求int32/int64类型，但会在x是complex类型时被自动转换为complex导致报错，因此希望加上skip_transform保留原始类型
* add more type&test for getitem * fix uint8 * add atomicAdd some int types * fix cpplint error * fix some type error * add some miss type * add return for cudaAdd
* add more type&test for getitem * fix uint8 * add atomicAdd some int types * fix cpplint error * fix some type error * add some miss type * add return for cudaAdd
PR types
Function optimization
PR changes
APIs
Description
完善get_value相关OP的dtype支持