[PATCH] D37325: [AMDGPU] Use v_pm_max_f16 for fcanonicalize

[Stanislav Mekhanoshin via Phabricator via llvm-commits]

rampitec added inline comments.


================
Comment at: lib/Target/AMDGPU/SIInstructions.td:1289
+  (fcanonicalize (v2f16 (VOP3PMods v2f16:$src, i32:$src_mods))),
+  (V_PK_MUL_F16 SRCMODS.OP_SEL_1, (i32 CONST.V2FP16_ONE), $src_mods, $src, DSTCLAMP.NONE)
+>;
----------------
arsenm wrote:
> This won't work. For now it's probably easier to just throw an S_MOV_B32 of the constant. This won't encode correctly as a direct immediate because you need to manipulate op_sel
It was here before, this is the old code and the fix does not belong to the current change.


https://reviews.llvm.org/D37325