[AMDGPU][True16][MC] true16 for v_minmax/maxmin_f16 and v_minmax/maxmin_num_f16 (#120617)

True16 support for v_minmax/maxmin_f16(GFX11) and
v_minmax/maxmin_num_f16(GFX12).

These insts are updated at the same time since we are replacing the
`v_minmax/maxmin_f16` to `v_minmax/maxmin_fake16_f16` while
`v_minmax/maxmin_num_f16` are alias insts and share the same CodeGen
pattern.

Added a GFX12 runline in minmax.ll in fake16 flow

Author: broxigarchen

llvm/lib/Target/AMDGPU/SIInstructions.td

@@ -3710,12 +3710,15 @@ def : IntMinMaxPat<V_MAXMIN_U32_e64, umin, umax_oneuse>;
 def : IntMinMaxPat<V_MINMAX_U32_e64, umax, umin_oneuse>;
 def : FPMinMaxPat<V_MINMAX_F32_e64, f32, fmaxnum_like, fminnum_like_oneuse>;
 def : FPMinMaxPat<V_MAXMIN_F32_e64, f32, fminnum_like, fmaxnum_like_oneuse>;
-def : FPMinMaxPat<V_MINMAX_F16_e64, f16, fmaxnum_like, fminnum_like_oneuse>;
-def : FPMinMaxPat<V_MAXMIN_F16_e64, f16, fminnum_like, fmaxnum_like_oneuse>;
 def : FPMinCanonMaxPat<V_MINMAX_F32_e64, f32, fmaxnum_like, fminnum_like_oneuse>;
 def : FPMinCanonMaxPat<V_MAXMIN_F32_e64, f32, fminnum_like, fmaxnum_like_oneuse>;
-def : FPMinCanonMaxPat<V_MINMAX_F16_e64, f16, fmaxnum_like, fminnum_like_oneuse>;
-def : FPMinCanonMaxPat<V_MAXMIN_F16_e64, f16, fminnum_like, fmaxnum_like_oneuse>;
+}
+
+let True16Predicate = UseFakeTrue16Insts in {
+def : FPMinMaxPat<V_MINMAX_F16_fake16_e64, f16, fmaxnum_like, fminnum_like_oneuse>;
+def : FPMinMaxPat<V_MAXMIN_F16_fake16_e64, f16, fminnum_like, fmaxnum_like_oneuse>;
+def : FPMinCanonMaxPat<V_MINMAX_F16_fake16_e64, f16, fmaxnum_like, fminnum_like_oneuse>;
+def : FPMinCanonMaxPat<V_MAXMIN_F16_fake16_e64, f16, fminnum_like, fmaxnum_like_oneuse>;
 }
 
 let SubtargetPredicate = isGFX9Plus in {

llvm/lib/Target/AMDGPU/VOP3Instructions.td

@@ -1374,8 +1374,8 @@ class VOP3_DOT_Profile_fake16<VOPProfile P, VOP3Features Features = VOP3_REGULAR
 let SubtargetPredicate = isGFX11Plus in {
   defm V_MAXMIN_F32     : VOP3Inst<"v_maxmin_f32", VOP3_Profile<VOP_F32_F32_F32_F32>>;
   defm V_MINMAX_F32     : VOP3Inst<"v_minmax_f32", VOP3_Profile<VOP_F32_F32_F32_F32>>;
-  defm V_MAXMIN_F16     : VOP3Inst<"v_maxmin_f16", VOP3_Profile<VOP_F16_F16_F16_F16>>;
-  defm V_MINMAX_F16     : VOP3Inst<"v_minmax_f16", VOP3_Profile<VOP_F16_F16_F16_F16>>;
+  defm V_MAXMIN_F16     : VOP3Inst_t16<"v_maxmin_f16", VOP_F16_F16_F16_F16>;
+  defm V_MINMAX_F16     : VOP3Inst_t16<"v_minmax_f16", VOP_F16_F16_F16_F16>;
   defm V_MAXMIN_U32     : VOP3Inst<"v_maxmin_u32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
   defm V_MINMAX_U32     : VOP3Inst<"v_minmax_u32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
   defm V_MAXMIN_I32     : VOP3Inst<"v_maxmin_i32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
@@ -1588,8 +1588,8 @@ defm V_MED3_NUM_F32       : VOP3_Realtriple_with_name_gfx12<0x231, "V_MED3_F32",
 defm V_MED3_NUM_F16       : VOP3_Realtriple_t16_and_fake16_gfx12<0x232, "v_med3_num_f16", "V_MED3_F16", "v_med3_f16">;
 defm V_MINMAX_NUM_F32     : VOP3_Realtriple_with_name_gfx12<0x268, "V_MINMAX_F32", "v_minmax_num_f32">;
 defm V_MAXMIN_NUM_F32     : VOP3_Realtriple_with_name_gfx12<0x269, "V_MAXMIN_F32", "v_maxmin_num_f32">;
-defm V_MINMAX_NUM_F16     : VOP3_Realtriple_with_name_gfx12<0x26a, "V_MINMAX_F16", "v_minmax_num_f16">;
-defm V_MAXMIN_NUM_F16     : VOP3_Realtriple_with_name_gfx12<0x26b, "V_MAXMIN_F16", "v_maxmin_num_f16">;
+defm V_MINMAX_NUM_F16     : VOP3_Realtriple_t16_and_fake16_gfx12<0x26a, "v_minmax_num_f16", "V_MINMAX_F16", "v_minmax_f16">;
+defm V_MAXMIN_NUM_F16     : VOP3_Realtriple_t16_and_fake16_gfx12<0x26b, "v_maxmin_num_f16", "V_MAXMIN_F16", "v_maxmin_f16">;
 defm V_MINIMUMMAXIMUM_F32 : VOP3Only_Realtriple_gfx12<0x26c>;
 defm V_MAXIMUMMINIMUM_F32 : VOP3Only_Realtriple_gfx12<0x26d>;
 defm V_MINIMUMMAXIMUM_F16 : VOP3Only_Realtriple_t16_gfx12<0x26e>;
@@ -1730,8 +1730,8 @@ defm V_PERMLANE16_B32      : VOP3_Real_Base_gfx11_gfx12<0x25b>;
 defm V_PERMLANEX16_B32     : VOP3_Real_Base_gfx11_gfx12<0x25c>;
 defm V_MAXMIN_F32          : VOP3_Realtriple_gfx11<0x25e>;
 defm V_MINMAX_F32          : VOP3_Realtriple_gfx11<0x25f>;
-defm V_MAXMIN_F16          : VOP3_Realtriple_gfx11<0x260>;
-defm V_MINMAX_F16          : VOP3_Realtriple_gfx11<0x261>;
+defm V_MAXMIN_F16          : VOP3_Realtriple_t16_and_fake16_gfx11<0x260, "v_maxmin_f16">;
+defm V_MINMAX_F16          : VOP3_Realtriple_t16_and_fake16_gfx11<0x261, "v_minmax_f16">;
 defm V_MAXMIN_U32          : VOP3_Realtriple_gfx11_gfx12<0x262>;
 defm V_MINMAX_U32          : VOP3_Realtriple_gfx11_gfx12<0x263>;
 defm V_MAXMIN_I32          : VOP3_Realtriple_gfx11_gfx12<0x264>;

llvm/lib/Target/AMDGPU/VOPInstructions.td

@@ -1909,8 +1909,8 @@ multiclass VOP3_Realtriple_t16_gfx11<bits<10> op, string asmName, string opName
 
 multiclass VOP3_Realtriple_t16_and_fake16_gfx11<bits<10> op, string asmName, string opName = NAME,
                                                 string pseudo_mnemonic = "", bit isSingle = 0> {
-  defm _t16: VOP3_Realtriple_t16_gfx11<op, opName#"_t16", asmName, pseudo_mnemonic, isSingle>;
-  defm _fake16: VOP3_Realtriple_t16_gfx11<op, opName#"_fake16", asmName, pseudo_mnemonic, isSingle>;
+  defm _t16: VOP3_Realtriple_t16_gfx11<op, asmName, opName#"_t16", pseudo_mnemonic, isSingle>;
+  defm _fake16: VOP3_Realtriple_t16_gfx11<op, asmName, opName#"_fake16", pseudo_mnemonic, isSingle>;
 }
 
 multiclass VOP3Only_Realtriple_t16_gfx11<bits<10> op, string asmName,