Commit 225fc4f
authored
[AMDGPU][SDAG] Try folding "lshr i64 + mad" to "mad_u64_u32" (#119218)
The intention is to use a "copy" instead of a "sub" to handle the high
parts of 64-bit multiply for this specific case.
This unlocks copy prop use cases where the copy can be reused by later
multiply+add sequences if possible.
Fixes: SWDEV-487672, SWDEV-4876691 parent f999b11 commit 225fc4f
File tree
2 files changed
+190
-115
lines changed- llvm
- lib/Target/AMDGPU
- test/CodeGen/AMDGPU
2 files changed
+190
-115
lines changed| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
13884 | 13884 | | |
13885 | 13885 | | |
13886 | 13886 | | |
| 13887 | + | |
| 13888 | + | |
| 13889 | + | |
| 13890 | + | |
| 13891 | + | |
| 13892 | + | |
| 13893 | + | |
| 13894 | + | |
| 13895 | + | |
| 13896 | + | |
| 13897 | + | |
| 13898 | + | |
| 13899 | + | |
| 13900 | + | |
| 13901 | + | |
| 13902 | + | |
| 13903 | + | |
| 13904 | + | |
| 13905 | + | |
| 13906 | + | |
| 13907 | + | |
| 13908 | + | |
| 13909 | + | |
| 13910 | + | |
| 13911 | + | |
| 13912 | + | |
| 13913 | + | |
| 13914 | + | |
| 13915 | + | |
| 13916 | + | |
| 13917 | + | |
13887 | 13918 | | |
13888 | 13919 | | |
13889 | 13920 | | |
| |||
13942 | 13973 | | |
13943 | 13974 | | |
13944 | 13975 | | |
| 13976 | + | |
| 13977 | + | |
| 13978 | + | |
13945 | 13979 | | |
13946 | 13980 | | |
13947 | 13981 | | |
| |||
0 commit comments