[llvm] [AMDGPU][True16][MC] VOPC profile fake16 pseudo update (PR #113175)

Brox Chen via llvm-commits llvm-commits at lists.llvm.org
Mon Nov 4 10:36:01 PST 2024


================
@@ -3463,7 +3463,7 @@ def : GCNPat <
              SRCMODS.NONE,
              (V_MOV_B64_PSEUDO (i64 0x3fefffffffffffff))),
          $x,
-         (V_CMP_CLASS_F64_e64 SRCMODS.NONE, $x, (i32 3 /*NaN*/))))
+         (V_CMP_CLASS_F64_e64 SRCMODS.NONE, $x, SRCMODS.NONE, (i32 3 /*NaN*/))))
----------------
broxigarchen wrote:

Hi Matt. The context of this change is that the `v_cmp_class_f16` has f16 in the src1 type. We change this to `f16` to enable inline of both int and floating points immediates. However it seems we could not find a way to suppress the additonal src modifier. also @Sisyph to help clarify if I misunderstood anything

https://github.com/llvm/llvm-project/pull/113175


More information about the llvm-commits mailing list