[llvm] [AMDGPU][True16][CodeGen] support for true16 for vinterp 16bit instructions (PR #116702)

[Brox Chen via llvm-commits]

================
@@ -181,9 +181,43 @@ multiclass VInterpF16Pat <SDPatternOperator op, Instruction inst,
   def : VInterpF16Pat<op, inst, dst_type, 1, high_pat>;
 }
 
+class VInterpF16Pat_t16 <SDPatternOperator op, Instruction inst,
+                     ValueType dstVT, bit high, bit isP2> : GCNPat <
+   (dstVT (op
+      (VINTERPMods f32:$src0, i32:$src0_modifiers),
+      (VINTERPMods f32:$src1, i32:$src1_modifiers),
+      (VINTERPMods f32:$src2, i32:$src2_modifiers),
+      !if(high, (i1 -1), (i1 0)))),
+    (inst $src0_modifiers,
+          (f16 (EXTRACT_SUBREG VGPR_32:$src0, !if(high, hi16, lo16))),
----------------
broxigarchen wrote:

Hi Matt. I think the input types are depending on the intrinsic. From this test https://github.com/llvm/llvm-project/pull/116702/files#diff-67c39b2811e166871c5e79f74f33221b90c06a20645efe1745e294f99cb3bec8R123 it seems we are passing a 32 bits float here as the input

https://github.com/llvm/llvm-project/pull/116702