[AMDGPU][MC][GFX9] Added op_sel support for v_mad_*16, v_fma_f16, v_div_fixup_f16

[dp]

Summary

This change implements features postponed in D35424 because of a dependency on D36322

[dp]

Needs refactoring after integration of D36674

[dp]

Updated to reflect D36674 changes. Refactored to improve readability.

[SamWot]

LGTM with small issue.

lib/Target/AMDGPU/VOP3Instructions.td

108

Better use HasModifiers or HasSrc0FloatMods from VOPProfile class.

lib/Target/AMDGPU/VOP3Instructions.td

class getVOP3Pat<VOPProfile P, SDPatternOperator node> {
   list<dag> ret = !if(!eq(P.NumSrcArgs, 3), ret3,
                   !if(!eq(P.NumSrcArgs, 2), ret2,
                   ret1));
 }
 
 class VOP3Inst<string OpName, VOPProfile P, SDPatternOperator node = null_frag, bit VOP3Only = 0> :
   VOP3_Pseudo<OpName, P,
     !if(P.HasModifiers, getVOP3ModPat<P, node>.ret, getVOP3Pat<P, node>.ret),
     VOP3Only>;
 
 class VOP3OpSelInst<string OpName, VOPProfile P, SDPatternOperator node = null_frag> :
   VOP3_Pseudo<OpName, P,
     !if(isFloatType<P.Src0VT>.ret,
         getVOP3OpSelModPat<P, node>.ret,
         getVOP3OpSelPat<P, node>.ret),
     1, 0, 1> {
 

 
 
 let SubtargetPredicate = Has16BitInsts in {
 
 let F16_ZFILL = 1 in {
 def V_DIV_FIXUP_F16 : VOP3Inst <"v_div_fixup_f16", VOP3_Profile<VOP_F16_F16_F16_F16>, AMDGPUdiv_fixup>;
 }
 let SubtargetPredicate = isGFX9 in {
-def V_DIV_FIXUP_F16_gfx9   : VOP3Inst <"v_div_fixup_f16_gfx9", VOP3_Profile<VOP_F16_F16_F16_F16>>;
+def V_DIV_FIXUP_F16_gfx9 : VOP3OpSelInst <"v_div_fixup_f16_gfx9", VOP3OpSel_Profile<VOP_F16_F16_F16_F16>>;
 }
 
 let isCommutable = 1 in {
 
 let F16_ZFILL = 1 in {
 def V_MAD_F16 : VOP3Inst <"v_mad_f16", VOP3_Profile<VOP_F16_F16_F16_F16>, fmad>;
 def V_MAD_U16 : VOP3Inst <"v_mad_u16", VOP3_Profile<VOP_I16_I16_I16_I16>>;
 def V_MAD_I16 : VOP3Inst <"v_mad_i16", VOP3_Profile<VOP_I16_I16_I16_I16>>;
 def V_FMA_F16 : VOP3Inst <"v_fma_f16", VOP3_Profile<VOP_F16_F16_F16_F16>, fma>;
 }
 
 let SubtargetPredicate = isGFX9 in {
-def V_MAD_F16_gfx9   : VOP3Inst <"v_mad_f16_gfx9", VOP3_Profile<VOP_F16_F16_F16_F16>>;
-def V_MAD_U16_gfx9   : VOP3Inst <"v_mad_u16_gfx9", VOP3_Profile<VOP_I16_I16_I16_I16>>;
-def V_MAD_I16_gfx9   : VOP3Inst <"v_mad_i16_gfx9", VOP3_Profile<VOP_I16_I16_I16_I16>>;
-def V_FMA_F16_gfx9   : VOP3Inst <"v_fma_f16_gfx9", VOP3_Profile<VOP_F16_F16_F16_F16>>;
+def V_MAD_F16_gfx9   : VOP3OpSelInst <"v_mad_f16_gfx9", VOP3OpSel_Profile<VOP_F16_F16_F16_F16>>;
+def V_MAD_U16_gfx9   : VOP3OpSelInst <"v_mad_u16_gfx9", VOP3OpSel_Profile<VOP_I16_I16_I16_I16>>;
+def V_MAD_I16_gfx9   : VOP3OpSelInst <"v_mad_i16_gfx9", VOP3OpSel_Profile<VOP_I16_I16_I16_I16>>;
+def V_FMA_F16_gfx9   : VOP3OpSelInst <"v_fma_f16_gfx9", VOP3OpSel_Profile<VOP_F16_F16_F16_F16>>;
 } // End SubtargetPredicate = isGFX9
 
 def V_INTERP_P1LL_F16 : VOP3Interp <"v_interp_p1ll_f16", VOP3_INTERP16<[f32, f32, i32, untyped]>>;
 def V_INTERP_P1LV_F16 : VOP3Interp <"v_interp_p1lv_f16", VOP3_INTERP16<[f32, f32, i32, f16]>>;
 def V_INTERP_P2_F16   : VOP3Interp <"v_interp_p2_f16",   VOP3_INTERP16<[f16, f32, i32, f32]>>;
 
 }  // End isCommutable = 1
 } // End SubtargetPredicate = Has16BitInsts

 multiclass VOP3_F16_Real_gfx9<bits<10> op, string OpName, string AsmName> {
   def _vi : VOP3_Real<!cast<VOP3_Pseudo>(OpName), SIEncodingFamily.GFX9>,
             VOP3e_vi <op, !cast<VOP3_Pseudo>(OpName).Pfl> {
               VOP3_Pseudo ps = !cast<VOP3_Pseudo>(OpName);
               let AsmString = AsmName # ps.AsmOperands;
             }
 }
 
+multiclass VOP3OpSel_F16_Real_gfx9<bits<10> op, string AsmName> {
+  def _vi : VOP3_Real<!cast<VOP3_Pseudo>(NAME), SIEncodingFamily.GFX9>,
+            VOP3OpSel_gfx9 <op, !cast<VOP3_Pseudo>(NAME).Pfl> {
+              VOP3_Pseudo ps = !cast<VOP3_Pseudo>(NAME);
+              let AsmString = AsmName # ps.AsmOperands;
+            }
+}
+
 } // End AssemblerPredicates = [isGFX9], DecoderNamespace = "GFX9"
 
 defm V_MAD_U64_U32      : VOP3be_Real_vi <0x1E8>;
 defm V_MAD_I64_I32      : VOP3be_Real_vi <0x1E9>;
 
 defm V_MAD_LEGACY_F32   : VOP3_Real_vi <0x1c0>;
 defm V_MAD_F32          : VOP3_Real_vi <0x1c1>;
 defm V_MAD_I32_I24      : VOP3_Real_vi <0x1c2>;

 defm V_DIV_FIXUP_F16    : VOP3_F16_Real_vi <0x1ef>;
 
 defm V_MAD_LEGACY_F16       : VOP3_F16_Real_gfx9 <0x1ea, "V_MAD_F16",       "v_mad_legacy_f16">;
 defm V_MAD_LEGACY_U16       : VOP3_F16_Real_gfx9 <0x1eb, "V_MAD_U16",       "v_mad_legacy_u16">;
 defm V_MAD_LEGACY_I16       : VOP3_F16_Real_gfx9 <0x1ec, "V_MAD_I16",       "v_mad_legacy_i16">;
 defm V_FMA_LEGACY_F16       : VOP3_F16_Real_gfx9 <0x1ee, "V_FMA_F16",       "v_fma_legacy_f16">;
 defm V_DIV_FIXUP_LEGACY_F16 : VOP3_F16_Real_gfx9 <0x1ef, "V_DIV_FIXUP_F16", "v_div_fixup_legacy_f16">;
 
-defm V_MAD_F16_gfx9         : VOP3_F16_Real_gfx9 <0x203, "V_MAD_F16_gfx9",       "v_mad_f16">;
-defm V_MAD_U16_gfx9         : VOP3_F16_Real_gfx9 <0x204, "V_MAD_U16_gfx9",       "v_mad_u16">;
-defm V_MAD_I16_gfx9         : VOP3_F16_Real_gfx9 <0x205, "V_MAD_I16_gfx9",       "v_mad_i16">;
-defm V_FMA_F16_gfx9         : VOP3_F16_Real_gfx9 <0x206, "V_FMA_F16_gfx9",       "v_fma_f16">;
-defm V_DIV_FIXUP_F16_gfx9   : VOP3_F16_Real_gfx9 <0x207, "V_DIV_FIXUP_F16_gfx9", "v_div_fixup_f16">;
+defm V_MAD_F16_gfx9         : VOP3OpSel_F16_Real_gfx9 <0x203, "v_mad_f16">;
+defm V_MAD_U16_gfx9         : VOP3OpSel_F16_Real_gfx9 <0x204, "v_mad_u16">;
+defm V_MAD_I16_gfx9         : VOP3OpSel_F16_Real_gfx9 <0x205, "v_mad_i16">;
+defm V_FMA_F16_gfx9         : VOP3OpSel_F16_Real_gfx9 <0x206, "v_fma_f16">;
+defm V_DIV_FIXUP_F16_gfx9   : VOP3OpSel_F16_Real_gfx9 <0x207, "v_div_fixup_f16">;
 
 defm V_INTERP_P1_F32_e64  : VOP3Interp_Real_vi <0x270>;
 defm V_INTERP_P2_F32_e64  : VOP3Interp_Real_vi <0x271>;
 defm V_INTERP_MOV_F32_e64 : VOP3Interp_Real_vi <0x272>;
 
 defm V_INTERP_P1LL_F16  : VOP3Interp_Real_vi <0x274>;
 defm V_INTERP_P1LV_F16  : VOP3Interp_Real_vi <0x275>;
 defm V_INTERP_P2_F16    : VOP3Interp_Real_vi <0x276>;

test/MC/AMDGPU/vop3-gfx9.s

 // GFX9: v_fma_f16 v5, v1, -v2, v3 ; encoding: [0x05,0x00,0x06,0xd2,0x01,0x05,0x0e,0x44]
 
 v_fma_f16 v5, v1, v2, |v3|
 // GFX9: v_fma_f16 v5, v1, v2, |v3| ; encoding: [0x05,0x04,0x06,0xd2,0x01,0x05,0x0e,0x04]
 
 v_fma_f16 v5, v1, v2, v3 clamp
 // GFX9: v_fma_f16 v5, v1, v2, v3 clamp ; encoding: [0x05,0x80,0x06,0xd2,0x01,0x05,0x0e,0x04]
 
+v_fma_f16 v5, v1, v2, v3 op_sel:[1,0,0,0]
+// GFX9: v_fma_f16 v5, v1, v2, v3 op_sel:[1,0,0,0] ; encoding: [0x05,0x08,0x06,0xd2,0x01,0x05,0x0e,0x04]
+
+v_fma_f16 v5, v1, v2, v3 op_sel:[0,1,0,0]
+// GFX9: v_fma_f16 v5, v1, v2, v3 op_sel:[0,1,0,0] ; encoding: [0x05,0x10,0x06,0xd2,0x01,0x05,0x0e,0x04]
+
+v_fma_f16 v5, v1, v2, v3 op_sel:[1,1,1,1]
+// GFX9: v_fma_f16 v5, v1, v2, v3 op_sel:[1,1,1,1] ; encoding: [0x05,0x78,0x06,0xd2,0x01,0x05,0x0e,0x04]
+
 v_fma_legacy_f16_e64 v5, v1, v2, v3
 // GFX9: v_fma_legacy_f16 v5, v1, v2, v3 ; encoding:  [0x05,0x00,0xee,0xd1,0x01,0x05,0x0e,0x04]
 
 v_fma_legacy_f16 v5, -v1, v2, v3
 // GFX9: v_fma_legacy_f16 v5, -v1, v2, v3 ; encoding:  [0x05,0x00,0xee,0xd1,0x01,0x05,0x0e,0x24]
 
 v_fma_legacy_f16 v5, v1, |v2|, v3
 // GFX9: v_fma_legacy_f16 v5, v1, |v2|, v3 ; encoding:  [0x05,0x02,0xee,0xd1,0x01,0x05,0x0e,0x04]

 // GFX9: v_div_fixup_f16 v5, v1, v2, 0.5 ; encoding: [0x05,0x00,0x07,0xd2,0x01,0x05,0xc2,0x03]
 
 v_div_fixup_f16 v5, -v1, v2, v3
 // GFX9: v_div_fixup_f16 v5, -v1, v2, v3 ; encoding: [0x05,0x00,0x07,0xd2,0x01,0x05,0x0e,0x24]
 
 v_div_fixup_f16 v5, |v1|, v2, v3
 // GFX9: v_div_fixup_f16 v5, |v1|, v2, v3 ; encoding: [0x05,0x01,0x07,0xd2,0x01,0x05,0x0e,0x04]
 
+v_div_fixup_f16 v5, v1, v2, v3 op_sel:[1,0,0,0]
+// GFX9: v_div_fixup_f16 v5, v1, v2, v3 op_sel:[1,0,0,0] ; encoding: [0x05,0x08,0x07,0xd2,0x01,0x05,0x0e,0x04]
+
+v_div_fixup_f16 v5, v1, v2, v3 op_sel:[0,0,1,0]
+// GFX9: v_div_fixup_f16 v5, v1, v2, v3 op_sel:[0,0,1,0] ; encoding: [0x05,0x20,0x07,0xd2,0x01,0x05,0x0e,0x04]
+
+v_div_fixup_f16 v5, v1, v2, v3 op_sel:[0,0,0,1]
+// GFX9: v_div_fixup_f16 v5, v1, v2, v3 op_sel:[0,0,0,1] ; encoding: [0x05,0x40,0x07,0xd2,0x01,0x05,0x0e,0x04]
+
 v_div_fixup_f16 v5, v1, v2, v3 clamp
 // GFX9: v_div_fixup_f16 v5, v1, v2, v3 clamp ; encoding: [0x05,0x80,0x07,0xd2,0x01,0x05,0x0e,0x04]
 
 v_div_fixup_legacy_f16_e64 v5, 0.5, v2, v3
 // GFX9: v_div_fixup_legacy_f16 v5, 0.5, v2, v3 ; encoding: [0x05,0x00,0xef,0xd1,0xf0,0x04,0x0e,0x04]
 
 v_div_fixup_legacy_f16 v5, v1, 0.5, v3
 // GFX9: v_div_fixup_legacy_f16 v5, v1, 0.5, v3 ; encoding: [0x05,0x00,0xef,0xd1,0x01,0xe1,0x0d,0x04]

 // GFX9: v_mad_f16 v5, v1, v2, 0.5 ; encoding: [0x05,0x00,0x03,0xd2,0x01,0x05,0xc2,0x03]
 
 v_mad_f16 v5, v1, v2, -v3
 // GFX9: v_mad_f16 v5, v1, v2, -v3 ; encoding: [0x05,0x00,0x03,0xd2,0x01,0x05,0x0e,0x84]
 
 v_mad_f16 v5, v1, v2, |v3|
 // GFX9: v_mad_f16 v5, v1, v2, |v3| ; encoding: [0x05,0x04,0x03,0xd2,0x01,0x05,0x0e,0x04]
 
+v_mad_f16 v5, v1, v2, v3 op_sel:[0,0,0,0]
+// GFX9: v_mad_f16 v5, v1, v2, v3 ; encoding: [0x05,0x00,0x03,0xd2,0x01,0x05,0x0e,0x04]
+
+v_mad_f16 v5, v1, v2, v3 op_sel:[1,0,0,0]
+// GFX9: v_mad_f16 v5, v1, v2, v3 op_sel:[1,0,0,0] ; encoding: [0x05,0x08,0x03,0xd2,0x01,0x05,0x0e,0x04]
+
+v_mad_f16 v5, v1, v2, v3 op_sel:[0,1,0,0]
+// GFX9: v_mad_f16 v5, v1, v2, v3 op_sel:[0,1,0,0] ; encoding: [0x05,0x10,0x03,0xd2,0x01,0x05,0x0e,0x04]
+
+v_mad_f16 v5, v1, v2, v3 op_sel:[0,0,1,0]
+// GFX9: v_mad_f16 v5, v1, v2, v3 op_sel:[0,0,1,0] ; encoding: [0x05,0x20,0x03,0xd2,0x01,0x05,0x0e,0x04]
+
+v_mad_f16 v5, v1, v2, v3 op_sel:[0,0,0,1]
+// GFX9: v_mad_f16 v5, v1, v2, v3 op_sel:[0,0,0,1] ; encoding: [0x05,0x40,0x03,0xd2,0x01,0x05,0x0e,0x04]
+
+v_mad_f16 v5, v1, v2, v3 op_sel:[1,1,1,1]
+// GFX9: v_mad_f16 v5, v1, v2, v3 op_sel:[1,1,1,1] ; encoding: [0x05,0x78,0x03,0xd2,0x01,0x05,0x0e,0x04]
+
 v_mad_f16 v5, v1, v2, v3 clamp
 // GFX9: v_mad_f16 v5, v1, v2, v3 clamp ; encoding: [0x05,0x80,0x03,0xd2,0x01,0x05,0x0e,0x04]
 
 v_mad_i16_e64 v5, 0, v2, v3
 // GFX9: v_mad_i16 v5, 0, v2, v3 ; encoding: [0x05,0x00,0x05,0xd2,0x80,0x04,0x0e,0x04]
 
 v_mad_i16 v5, v1, -1, v3
 // GFX9: v_mad_i16 v5, v1, -1, v3 ; encoding: [0x05,0x00,0x05,0xd2,0x01,0x83,0x0d,0x04]
 
 v_mad_i16 v5, v1, v2, -4.0
 // GFX9: v_mad_i16 v5, v1, v2, -4.0 ; encoding: [0x05,0x00,0x05,0xd2,0x01,0x05,0xde,0x03]
 
+v_mad_i16 v5, v1, v2, v3 op_sel:[0,0,0,1]
+// GFX9: v_mad_i16 v5, v1, v2, v3 op_sel:[0,0,0,1] ; encoding: [0x05,0x40,0x05,0xd2,0x01,0x05,0x0e,0x04]
+
+v_mad_i16 v5, v1, v2, v3 op_sel:[1,1,1,1]
+// GFX9: v_mad_i16 v5, v1, v2, v3 op_sel:[1,1,1,1] ; encoding: [0x05,0x78,0x05,0xd2,0x01,0x05,0x0e,0x04]
+
 v_mad_legacy_f16_e64 v5, 0.5, v2, v3
 // GFX9: v_mad_legacy_f16 v5, 0.5, v2, v3 ; encoding: [0x05,0x00,0xea,0xd1,0xf0,0x04,0x0e,0x04]
 
 v_mad_legacy_f16 v5, v1, 0.5, v3
 // GFX9: v_mad_legacy_f16 v5, v1, 0.5, v3 ; encoding: [0x05,0x00,0xea,0xd1,0x01,0xe1,0x0d,0x04]
 
 v_mad_legacy_f16 v5, v1, v2, 0.5
 // GFX9: v_mad_legacy_f16 v5, v1, v2, 0.5 ; encoding: [0x05,0x00,0xea,0xd1,0x01,0x05,0xc2,0x03]

 v_mad_u16_e64 v5, 0, v2, v3
 // GFX9: v_mad_u16 v5, 0, v2, v3 ; encoding: [0x05,0x00,0x04,0xd2,0x80,0x04,0x0e,0x04]
 
 v_mad_u16 v5, v1, -1, v3
 // GFX9: v_mad_u16 v5, v1, -1, v3 ; encoding: [0x05,0x00,0x04,0xd2,0x01,0x83,0x0d,0x04]
 
 v_mad_u16 v5, v1, v2, -4.0
 // GFX9: v_mad_u16 v5, v1, v2, -4.0 ; encoding: [0x05,0x00,0x04,0xd2,0x01,0x05,0xde,0x03]
+
+v_mad_u16 v5, v1, v2, v3 op_sel:[1,0,0,0]
+// GFX9: v_mad_u16 v5, v1, v2, v3 op_sel:[1,0,0,0] ; encoding: [0x05,0x08,0x04,0xd2,0x01,0x05,0x0e,0x04]
+
+v_mad_u16 v5, v1, v2, v3 op_sel:[0,0,0,1]
+// GFX9: v_mad_u16 v5, v1, v2, v3 op_sel:[0,0,0,1] ; encoding: [0x05,0x40,0x04,0xd2,0x01,0x05,0x0e,0x04]
+
+v_mad_u16 v5, v1, v2, v3 op_sel:[1,1,1,1]
+// GFX9: v_mad_u16 v5, v1, v2, v3 op_sel:[1,1,1,1] ; encoding: [0x05,0x78,0x04,0xd2,0x01,0x05,0x0e,0x04]

test/MC/Disassembler/AMDGPU/vop3_gfx9.txt

 # RUN: llvm-mc -arch=amdgcn -mcpu=gfx901 -disassemble -show-encoding < %s | FileCheck %s -check-prefix=GFX9
 
 # GFX9: v_fma_f16 v5, v1, v2, v3    ; encoding: [0x05,0x00,0x06,0xd2,0x01,0x05,0x0e,0x04]
 0x05,0x00,0x06,0xd2,0x01,0x05,0x0e,0x04
 
 # GFX9: v_fma_f16 v5, -v1, v2, v3    ; encoding: [0x05,0x00,0x06,0xd2,0x01,0x05,0x0e,0x24]
 0x05,0x00,0x06,0xd2,0x01,0x05,0x0e,0x24
 
 # GFX9: v_fma_f16 v5, v1, |v2|, v3    ; encoding: [0x05,0x02,0x06,0xd2,0x01,0x05,0x0e,0x04]
 0x05,0x02,0x06,0xd2,0x01,0x05,0x0e,0x04
 
+# GFX9: v_fma_f16 v5, v1, v2, v3 op_sel:[1,0,0,0] ; encoding: [0x05,0x08,0x06,0xd2,0x01,0x05,0x0e,0x04]
+0x05,0x08,0x06,0xd2,0x01,0x05,0x0e,0x04
+
+# GFX9: v_fma_f16 v5, v1, v2, v3 op_sel:[1,1,1,1] ; encoding: [0x05,0x78,0x06,0xd2,0x01,0x05,0x0e,0x04]
+0x05,0x78,0x06,0xd2,0x01,0x05,0x0e,0x04
+
 # GFX9: v_fma_f16 v5, v1, v2, v3 clamp    ; encoding: [0x05,0x80,0x06,0xd2,0x01,0x05,0x0e,0x04]
 0x05,0x80,0x06,0xd2,0x01,0x05,0x0e,0x04
 
 # CHECK: v_fma_legacy_f16 v5, v1, v2, v3    ; encoding: [0x05,0x00,0xee,0xd1,0x01,0x05,0x0e,0x04]
 0x05,0x00,0xee,0xd1,0x01,0x05,0x0e,0x04
 
 # CHECK: v_fma_legacy_f16 v5, v1, v2, -v3    ; encoding: [0x05,0x00,0xee,0xd1,0x01,0x05,0x0e,0x84]
 0x05,0x00,0xee,0xd1,0x01,0x05,0x0e,0x84