[llvm] [AMDGPU][GISel] Add RegBankLegalize support for G_STRICT_{FADD|FMUL} (PR #169406)

[Chinmay Deshpande via llvm-commits]

================
@@ -953,7 +962,9 @@ RegBankLegalizeRules::RegBankLegalizeRules(const GCNSubtarget &_ST,
            hasSALUFloat)
       .Div(V2S16, {{VgprV2S16}, {VgprV2S16, VgprV2S16}})
       .Any({{UniV2S32}, {{UniInVgprV2S32}, {VgprV2S32, VgprV2S32}}})
-      .Any({{DivV2S32}, {{VgprV2S32}, {VgprV2S32, VgprV2S32}}});
+      .Any({{DivV2S32}, {{VgprV2S32}, {VgprV2S32, VgprV2S32}}})
+      .Any({{UniV2S64}, {{UniInVgprV2S64}, {VgprV2S64, VgprV2S64}}})
+      .Any({{DivV2S64}, {{VgprV2S64}, {VgprV2S64, VgprV2S64}}});
----------------
chinmaydd wrote:

Removed v2s64.
Added tests for gfx12.

For packed operands, the appropriate file is `packed-fp32.ll` which requires support for `STRICT_FMA`. That needs to be handled in another PR.

https://github.com/llvm/llvm-project/pull/169406