Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[AMDGPU][MC][True16] Support VOP2 instructions with true16 format #115233

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

[AMDGPU][MC][True16] Support VOP2 instructions with true16 format #115233

wants to merge 1 commit into from
+794 −340

Conversation

broxigarchen
Copy link
Contributor

@broxigarchen broxigarchen commented Nov 6, 2024 •
edited
Loading

Support true16 format for VOP2 instructions in MC

This patch updates the true16 and fake16 vop_profile for the following instructions and update the asm/dasm tests:
v_fmac_f16
v_fmamk_f16
v_fmaak_f16

It seems vop2_t16_promote.s files are not yet updated with true16 flag in the previous batch update. It will be updated seperately

@broxigarchen broxigarchen force-pushed the main-merge-true16-vop2-mc-true16 branch from 595efc8 to 902962a Compare November 6, 2024 23:17
@broxigarchen broxigarchen marked this pull request as ready for review November 6, 2024 23:19
@llvmbot llvmbot added backend:AMDGPU mc Machine (object) code labels Nov 6, 2024
@llvmbot
Copy link
Collaborator

llvmbot commented Nov 6, 2024

@llvm/pr-subscribers-mc

Author: Brox Chen (broxigarchen)

Changes

Support true16 format for VOP2 instructions in MC

This patch updates the true16 and fake16 vop_profile for the following instructions and update the asm/dasm tests:
v_fmac_f16
v_fmamk_f16
v_fmaak_f16

Patch is 101.14 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/115233.diff

17 Files Affected:

  • (modified) llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp (+19)
  • (modified) llvm/lib/Target/AMDGPU/SIRegisterInfo.td (+9-1)
  • (modified) llvm/lib/Target/AMDGPU/VOP2Instructions.td (+70-6)
  • (modified) llvm/test/MC/AMDGPU/gfx11_asm_vop2.s (+72-54)
  • (modified) llvm/test/MC/AMDGPU/gfx11_asm_vop2_dpp16.s (+17-17)
  • (modified) llvm/test/MC/AMDGPU/gfx11_asm_vop2_dpp8.s (+8-5)
  • (modified) llvm/test/MC/AMDGPU/gfx11_asm_vop2_t16_err.s (+75-30)
  • (modified) llvm/test/MC/AMDGPU/gfx12_asm_vop2.s (+81-45)
  • (modified) llvm/test/MC/AMDGPU/gfx12_asm_vop2_dpp16.s (+20-14)
  • (modified) llvm/test/MC/AMDGPU/gfx12_asm_vop2_dpp8.s (+9-3)
  • (modified) llvm/test/MC/AMDGPU/gfx12_asm_vop2_t16_err.s (+75-30)
  • (modified) llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop2.txt (+117-54)
  • (modified) llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop2_dpp16.txt (+31-17)
  • (modified) llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop2_dpp8.txt (+12-3)
  • (modified) llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop2.txt (+132-45)
  • (modified) llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop2_dpp16.txt (+36-14)
  • (modified) llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop2_dpp8.txt (+11-2)
diff --git a/llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp b/llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp
index fdef9865b82c06..f4eec66fbb6157 100644
--- a/llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp
+++ b/llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp
@@ -345,6 +345,25 @@ static DecodeStatus decodeOperand_VSrcT16_Lo128(MCInst &Inst, unsigned Imm,
                               (AMDGPU::OperandSemantics)OperandSemantics));
 }
 
+template <AMDGPUDisassembler::OpWidthTy OpWidth, unsigned ImmWidth,
+          unsigned OperandSemantics>
+static DecodeStatus
+decodeOperand_VSrcT16_Lo128_Deferred(MCInst &Inst, unsigned Imm,
+                                     uint64_t /*Addr*/,
+                                     const MCDisassembler *Decoder) {
+  const auto *DAsm = static_cast<const AMDGPUDisassembler *>(Decoder);
+  assert(isUInt<9>(Imm) && "9-bit encoding expected");
+
+  if (Imm & AMDGPU::EncValues::IS_VGPR) {
+    bool IsHi = Imm & (1 << 7);
+    unsigned RegIdx = Imm & 0x7f;
+    return addOperand(Inst, DAsm->createVGPR16Operand(RegIdx, IsHi));
+  }
+  return addOperand(Inst, DAsm->decodeNonVGPRSrcOp(
+                              OpWidth, Imm & 0xFF, true, ImmWidth,
+                              (AMDGPU::OperandSemantics)OperandSemantics));
+}
+
 template <AMDGPUDisassembler::OpWidthTy OpWidth, unsigned ImmWidth,
           unsigned OperandSemantics>
 static DecodeStatus decodeOperand_VSrcT16(MCInst &Inst, unsigned Imm,
diff --git a/llvm/lib/Target/AMDGPU/SIRegisterInfo.td b/llvm/lib/Target/AMDGPU/SIRegisterInfo.td
index 852430129251c6..49be750acab960 100644
--- a/llvm/lib/Target/AMDGPU/SIRegisterInfo.td
+++ b/llvm/lib/Target/AMDGPU/SIRegisterInfo.td
@@ -1161,7 +1161,8 @@ def SSrcOrLds_b32 : SrcRegOrImm9 <SRegOrLds_32, "OPW32", "OPERAND_REG_IMM_INT32"
 class SrcRegOrImmDeferred9<RegisterClass regClass, string opWidth,
                            string operandType, int immWidth, int OperandSemantics>
     : RegOrImmOperand<regClass, operandType> {
-  let DecoderMethod = "decodeSrcRegOrImmDeferred9<AMDGPUDisassembler::" #
+  string DecoderMethodName = "decodeSrcRegOrImmDeferred9";
+  let DecoderMethod = DecoderMethodName # "<AMDGPUDisassembler::" #
                       opWidth # ", " # immWidth # ", " # OperandSemantics # ">";
 }
 
@@ -1222,6 +1223,13 @@ def VSrc_bf16_Deferred : SrcRegOrImmDeferred9<VS_32, "OPW16", "OPERAND_REG_IMM_B
 def VSrc_f16_Deferred : SrcRegOrImmDeferred9<VS_32, "OPW16", "OPERAND_REG_IMM_FP16_DEFERRED", 16, OperandSemantics.FP16>;
 def VSrc_f32_Deferred : SrcRegOrImmDeferred9<VS_32, "OPW32", "OPERAND_REG_IMM_FP32_DEFERRED", 32, OperandSemantics.FP32>;
 
+// True 16 Operands
+def VSrcT_f16_Lo128_Deferred : SrcRegOrImmDeferred9<VS_16_Lo128, "OPW16",
+                                                   "OPERAND_REG_IMM_FP16_DEFERRED", 16, OperandSemantics.FP16> {
+  let DecoderMethodName = "decodeOperand_VSrcT16_Lo128_Deferred";
+  let EncoderMethod = "getMachineOpValueT16Lo128";
+}
+
 def VSrcFake16_bf16_Lo128_Deferred
   : SrcRegOrImmDeferred9<VS_32_Lo128, "OPW16", "OPERAND_REG_IMM_BF16_DEFERRED", 16, OperandSemantics.BF16>;
 def VSrcFake16_f16_Lo128_Deferred
diff --git a/llvm/lib/Target/AMDGPU/VOP2Instructions.td b/llvm/lib/Target/AMDGPU/VOP2Instructions.td
index 925b60561c9d68..df41e3ceefe9e1 100644
--- a/llvm/lib/Target/AMDGPU/VOP2Instructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP2Instructions.td
@@ -374,6 +374,12 @@ class VOP_MADAK <ValueType vt> : VOP_MADK_Base<vt> {
 }
 
 def VOP_MADAK_F16 : VOP_MADAK <f16>;
+def VOP_MADAK_F16_t16 : VOP_MADAK <f16> {
+  let IsTrue16 = 1;
+  let IsRealTrue16 = 1;
+  let DstRC = getVALUDstForVT<DstVT, 1/*IsTrue16*/, 0/*IsVOP3Encoding*/>.ret;
+  let Ins32 = (ins VSrcT_f16_Lo128_Deferred:$src0, VGPRSrc_16_Lo128:$src1, ImmOpType:$imm);
+}
 def VOP_MADAK_F16_fake16 : VOP_MADAK <f16> {
   let IsTrue16 = 1;
   let DstRC = getVALUDstForVT_fake16<DstVT>.ret;
@@ -399,6 +405,12 @@ class VOP_MADMK <ValueType vt> : VOP_MADK_Base<vt> {
 }
 
 def VOP_MADMK_F16 : VOP_MADMK <f16>;
+def VOP_MADMK_F16_t16 : VOP_MADMK <f16> {
+  let IsTrue16 = 1;
+  let IsRealTrue16 = 1;
+  let DstRC = getVALUDstForVT<DstVT, 1/*IsTrue16*/, 0/*IsVOP3Encoding*/>.ret;
+  let Ins32 = (ins VSrcT_f16_Lo128_Deferred:$src0, ImmOpType:$imm, VGPRSrc_16_Lo128:$src1);
+}
 def VOP_MADMK_F16_fake16 : VOP_MADMK <f16> {
   let IsTrue16 = 1;
   let DstRC = getVALUDstForVT_fake16<DstVT>.ret;
@@ -467,6 +479,42 @@ class VOP_MAC <ValueType vt0, ValueType vt1=vt0> : VOPProfile <[vt0, vt1, vt1, v
 }
 
 def VOP_MAC_F16 : VOP_MAC <f16>;
+def VOP_MAC_F16_t16 : VOP_MAC <f16> {
+  let IsTrue16 = 1;
+  let IsRealTrue16 = 1;
+  let HasOpSel = 1;
+  let DstRC = VOPDstOperand_t16Lo128;
+  let Src0RC32 = getVOPSrc0ForVT<Src0VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret;
+  let Src1RC32 = getVregSrcForVT<Src1VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret;
+  let Ins32 = (ins Src0RC32:$src0, Src1RC32:$src1, getVregSrcForVT<Src2VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret:$src2);
+  let Src0DPP = getVregSrcForVT<Src0VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret;
+  let Src1DPP = getVregSrcForVT<Src1VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret;
+  let Src2DPP = getVregSrcForVT<Src2VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret;
+  let Src0ModDPP = getSrcModDPP_t16<Src0VT, 0/*IsFake16*/>.ret;
+  let Src1ModDPP = getSrcModDPP_t16<Src1VT, 0/*IsFake16*/>.ret;
+  let Src2ModDPP = getSrcModDPP_t16<Src2VT, 0/*IsFake16*/>.ret;
+  let InsDPP = (ins Src0ModDPP:$src0_modifiers, Src0DPP:$src0,
+                    Src1ModDPP:$src1_modifiers, Src1DPP:$src1,
+                    getVregSrcForVT<Src2VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret:$src2, // stub argument
+                    dpp_ctrl:$dpp_ctrl, DppRowMask:$row_mask,
+                    DppBankMask:$bank_mask, DppBoundCtrl:$bound_ctrl);
+  let InsDPP8 = (ins Src0ModDPP:$src0_modifiers, Src0DPP:$src0,
+                     Src1ModDPP:$src1_modifiers, Src1DPP:$src1,
+                     getVregSrcForVT<Src2VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret:$src2, // stub argument
+                     dpp8:$dpp8, Dpp8FI:$fi);
+  let DstRC64 = getVALUDstForVT<DstVT, 1/*IsTrue*/, 1/*IsVOP3Encoding*/>.ret;
+  let Src0RC64 = getVOP3SrcForVT<Src0VT, 1/*IsTrue16*/>.ret;
+  let Src1RC64 = getVOP3SrcForVT<Src1VT, 1/*IsTrue16*/>.ret;
+  let Src0VOP3DPP = VGPRSrc_16;
+  let Src1VOP3DPP = getVOP3DPPSrcForVT<Src1VT, 0/*IsFake16*/>.ret;
+  let Src2VOP3DPP = getVOP3DPPSrcForVT<Src2VT, 0/*IsFake16*/>.ret;
+  let Src0ModVOP3DPP = getSrc0ModVOP3DPP<Src0VT, DstVT, 0/*IsFake16*/>.ret;
+  let Src1ModVOP3DPP = getSrcModVOP3DPP<Src1VT, 0/*IsFake16*/>.ret;
+  let Src2ModVOP3DPP = getSrcModVOP3DPP<Src2VT, 0/*IsFake16*/>.ret;
+  let Src0Mod = getSrc0Mod<Src0VT, DstVT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret;
+  let Src1Mod = getSrcMod<Src1VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret;
+  let Src2Mod = getSrcMod<Src2VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret;
+}
 def VOP_MAC_F16_fake16 : VOP_MAC <f16> {
   let IsTrue16 = 1;
   let DstRC = getVALUDstForVT_fake16<DstVT>.ret;
@@ -998,6 +1046,9 @@ let FPDPRounding = 1, isReMaterializable = 1, FixedSize = 1 in {
 let SubtargetPredicate = isGFX10Plus, True16Predicate = NotHasTrue16BitInsts in {
   def V_FMAMK_F16 : VOP2_Pseudo <"v_fmamk_f16", VOP_MADMK_F16, [], "">;
 }
+let True16Predicate = UseRealTrue16Insts in {
+  def V_FMAMK_F16_t16 : VOP2_Pseudo <"v_fmamk_f16_t16", VOP_MADMK_F16_t16, [], "">;
+}
 let True16Predicate = UseFakeTrue16Insts in {
   def V_FMAMK_F16_fake16 : VOP2_Pseudo <"v_fmamk_f16_fake16", VOP_MADMK_F16_fake16, [], "">;
 }
@@ -1006,6 +1057,9 @@ let isCommutable = 1 in {
 let SubtargetPredicate = isGFX10Plus, True16Predicate = NotHasTrue16BitInsts in {
   def V_FMAAK_F16 : VOP2_Pseudo <"v_fmaak_f16", VOP_MADAK_F16, [], "">;
 }
+let True16Predicate = UseRealTrue16Insts in {
+  def V_FMAAK_F16_t16 : VOP2_Pseudo <"v_fmaak_f16_t16", VOP_MADAK_F16_t16, [], "">;
+}
 let True16Predicate = UseFakeTrue16Insts in {
   def V_FMAAK_F16_fake16 : VOP2_Pseudo <"v_fmaak_f16_fake16", VOP_MADAK_F16_fake16, [], "">;
 }
@@ -1020,6 +1074,9 @@ let SubtargetPredicate = isGFX10Plus in {
 let True16Predicate = NotHasTrue16BitInsts in {
   defm V_FMAC_F16 : VOP2Inst <"v_fmac_f16", VOP_MAC_F16>;
 }
+let True16Predicate = UseRealTrue16Insts in {
+  defm V_FMAC_F16_t16 : VOP2Inst <"v_fmac_f16_t16", VOP_MAC_F16_t16>;
+}
 let True16Predicate = UseFakeTrue16Insts in {
   defm V_FMAC_F16_fake16 : VOP2Inst <"v_fmac_f16_fake16", VOP_MAC_F16_fake16>;
 }
@@ -1692,8 +1749,8 @@ multiclass VOP3Only_Realtriple_t16_gfx11_gfx12<bits<10> op, string asmName, stri
   VOP3_Realtriple_t16_gfx12<op, asmName, OpName, "", /*IsSingle*/1>;
 
 multiclass VOP3Only_Realtriple_t16_and_fake16_gfx11_gfx12<bits<10> op, string asmName, string OpName = NAME> {
-  defm OpName#"_t16": VOP3Only_Realtriple_t16_gfx11_gfx12<op, asmName, OpName#"_t16">;
-  defm OpName#"_fake16": VOP3Only_Realtriple_t16_gfx11_gfx12<op, asmName, OpName#"_fake16">;
+  defm _t16: VOP3Only_Realtriple_t16_gfx11_gfx12<op, asmName, OpName#"_t16">;
+  defm _fake16: VOP3Only_Realtriple_t16_gfx11_gfx12<op, asmName, OpName#"_fake16">;
 }
 
 multiclass VOP3beOnly_Realtriple_gfx11_gfx12<bits<10> op> :
@@ -1712,7 +1769,14 @@ multiclass VOP2Only_Real_MADK_t16_and_fake16_gfx11_gfx12<bits<6> op, string asmN
 
 multiclass VOP2_Real_FULL_t16_gfx11_gfx12<bits<6> op, string asmName,
                                           string opName = NAME> :
-  VOP2_Real_FULL_with_name_gfx11_gfx12<op, opName, asmName>;
+  VOP2_Real_FULL_with_name<GFX11Gen, op, opName, asmName>,
+  VOP2_Real_FULL_with_name<GFX12Gen, op, opName, asmName>;
+
+multiclass VOP2_Real_FULL_t16_and_fake16_gfx11_gfx12<bits<6> op, string asmName,
+                                          string opName = NAME> {
+  defm _t16:    VOP2_Real_FULL_t16_gfx11_gfx12<op, asmName, opName#"_t16">;
+  defm _fake16: VOP2_Real_FULL_t16_gfx11_gfx12<op, asmName, opName#"_fake16">;
+}
 
 multiclass VOP2_Real_FULL_gfx11<bits<6> op> :
   VOP2_Real_FULL<GFX11Gen, op>;
@@ -1747,15 +1811,15 @@ defm V_SUBREV_F16_t16      : VOP2_Real_FULL_t16_gfx11_gfx12<0x034, "v_subrev_f16
 defm V_SUBREV_F16_fake16   : VOP2_Real_FULL_t16_gfx11_gfx12<0x034, "v_subrev_f16">;
 defm V_MUL_F16_t16         : VOP2_Real_FULL_t16_gfx11_gfx12<0x035, "v_mul_f16">;
 defm V_MUL_F16_fake16      : VOP2_Real_FULL_t16_gfx11_gfx12<0x035, "v_mul_f16">;
-defm V_FMAC_F16_fake16     : VOP2_Real_FULL_t16_gfx11_gfx12<0x036, "v_fmac_f16">;
+defm V_FMAC_F16            : VOP2_Real_FULL_t16_and_fake16_gfx11_gfx12<0x036, "v_fmac_f16">;
 defm V_LDEXP_F16_t16       : VOP2_Real_FULL_t16_gfx11_gfx12<0x03b, "v_ldexp_f16">;
 defm V_LDEXP_F16_fake16    : VOP2_Real_FULL_t16_gfx11_gfx12<0x03b, "v_ldexp_f16">;
 defm V_MAX_F16_t16         : VOP2_Real_FULL_t16_gfx11<0x039, "v_max_f16">;
 defm V_MAX_F16_fake16      : VOP2_Real_FULL_t16_gfx11<0x039, "v_max_f16">;
 defm V_MIN_F16_t16         : VOP2_Real_FULL_t16_gfx11<0x03a, "v_min_f16">;
 defm V_MIN_F16_fake16      : VOP2_Real_FULL_t16_gfx11<0x03a, "v_min_f16">;
-defm V_FMAMK_F16_fake16    : VOP2Only_Real_MADK_t16_gfx11_gfx12<0x037, "v_fmamk_f16">;
-defm V_FMAAK_F16_fake16    : VOP2Only_Real_MADK_t16_gfx11_gfx12<0x038, "v_fmaak_f16">;
+defm V_FMAMK_F16           : VOP2Only_Real_MADK_t16_and_fake16_gfx11_gfx12<0x037, "v_fmamk_f16">;
+defm V_FMAAK_F16           : VOP2Only_Real_MADK_t16_and_fake16_gfx11_gfx12<0x038, "v_fmaak_f16">;
 
 // VOP3 only.
 defm V_CNDMASK_B16         : VOP3Only_Realtriple_gfx11_gfx12<0x25d>;
diff --git a/llvm/test/MC/AMDGPU/gfx11_asm_vop2.s b/llvm/test/MC/AMDGPU/gfx11_asm_vop2.s
index 2a4b3ea2017013..0f2852fc531ed4 100644
--- a/llvm/test/MC/AMDGPU/gfx11_asm_vop2.s
+++ b/llvm/test/MC/AMDGPU/gfx11_asm_vop2.s
@@ -648,50 +648,56 @@ v_dot2c_f32_f16 v5, src_scc, v2
 v_dot2c_f32_f16 v255, 0xfe0b, v255
 // GFX11: encoding: [0xff,0xfe,0xff,0x05,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, v1, v2, 0xfe0b
+v_fmaak_f16 v5.l, v1.l, v2.l, 0xfe0b
 // GFX11: encoding: [0x01,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, v127, v2, 0xfe0b
+v_fmaak_f16 v5.l, v127.l, v2.l, 0xfe0b
 // GFX11: encoding: [0x7f,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, s1, v2, 0xfe0b
+v_fmaak_f16 v5.l, v1.h, v2.l, 0xfe0b
+// GFX11: encoding: [0x81,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00]
+
+v_fmaak_f16 v5.l, v127.h, v2.l, 0xfe0b
+// GFX11: encoding: [0xff,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00]
+
+v_fmaak_f16 v5.l, s1, v2.l, 0xfe0b
 // GFX11: encoding: [0x01,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, s105, v2, 0xfe0b
+v_fmaak_f16 v5.l, s105, v2.l, 0xfe0b
 // GFX11: encoding: [0x69,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, vcc_lo, v2, 0xfe0b
+v_fmaak_f16 v5.l, vcc_lo, v2.l, 0xfe0b
 // GFX11: encoding: [0x6a,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, vcc_hi, v2, 0xfe0b
+v_fmaak_f16 v5.l, vcc_hi, v2.l, 0xfe0b
 // GFX11: encoding: [0x6b,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, ttmp15, v2, 0xfe0b
+v_fmaak_f16 v5.l, ttmp15, v2.l, 0xfe0b
 // GFX11: encoding: [0x7b,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, m0, v2, 0xfe0b
+v_fmaak_f16 v5.l, m0, v2.l, 0xfe0b
 // GFX11: encoding: [0x7d,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, exec_lo, v2, 0xfe0b
+v_fmaak_f16 v5.l, exec_lo, v2.l, 0xfe0b
 // GFX11: encoding: [0x7e,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, exec_hi, v2, 0xfe0b
+v_fmaak_f16 v5.l, exec_hi, v2.l, 0xfe0b
 // GFX11: encoding: [0x7f,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, null, v2, 0xfe0b
+v_fmaak_f16 v5.l, null, v2.l, 0xfe0b
 // GFX11: encoding: [0x7c,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, -1, v2, 0xfe0b
+v_fmaak_f16 v5.l, -1, v2.l, 0xfe0b
 // GFX11: encoding: [0xc1,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, 0.5, v2, 0xfe0b
-// GFX11: encoding: [0xf0,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00]
+v_fmaak_f16 v127.l, 0.5, v127.l, 0xfe0b
+// GFX11: encoding: [0xf0,0xfe,0xfe,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, src_scc, v2, 0xfe0b
-// GFX11: encoding: [0xfd,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00]
+v_fmaak_f16 v5.h, src_scc, v2.h, 0xfe0b
+// GFX11: encoding: [0xfd,0x04,0x0b,0x71,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v127, 0xfe0b, v127, 0xfe0b
-// GFX11: encoding: [0xff,0xfe,0xfe,0x70,0x0b,0xfe,0x00,0x00]
+v_fmaak_f16 v127.h, 0xfe0b, v127.h, 0xfe0b
+// GFX11: encoding: [0xff,0xfe,0xff,0x71,0x0b,0xfe,0x00,0x00]
 
 v_fmaak_f32 v5, v1, v2, 0xaf123456
 // GFX11: encoding: [0x01,0x05,0x0a,0x5a,0x56,0x34,0x12,0xaf]
@@ -783,50 +789,56 @@ v_fmac_dx9_zero_f32 v5, src_scc, v2
 v_fmac_dx9_zero_f32 v255, 0xaf123456, v255
 // GFX11: encoding: [0xff,0xfe,0xff,0x0d,0x56,0x34,0x12,0xaf]
 
-v_fmac_f16 v5, v1, v2
+v_fmac_f16 v5.l, v1.l, v2.l
 // GFX11: encoding: [0x01,0x05,0x0a,0x6c]
 
-v_fmac_f16 v5, v127, v2
+v_fmac_f16 v5.l, v127.l, v2.l
 // GFX11: encoding: [0x7f,0x05,0x0a,0x6c]
 
-v_fmac_f16 v5, s1, v2
+v_fmac_f16 v5.l, v1.h, v2.l
+// GFX11: encoding: [0x81,0x05,0x0a,0x6c]
+
+v_fmac_f16 v5.l, v127.h, v2.l
+// GFX11: encoding: [0xff,0x05,0x0a,0x6c]
+
+v_fmac_f16 v5.l, s1, v2.l
 // GFX11: encoding: [0x01,0x04,0x0a,0x6c]
 
-v_fmac_f16 v5, s105, v2
+v_fmac_f16 v5.l, s105, v2.l
 // GFX11: encoding: [0x69,0x04,0x0a,0x6c]
 
-v_fmac_f16 v5, vcc_lo, v2
+v_fmac_f16 v5.l, vcc_lo, v2.l
 // GFX11: encoding: [0x6a,0x04,0x0a,0x6c]
 
-v_fmac_f16 v5, vcc_hi, v2
+v_fmac_f16 v5.l, vcc_hi, v2.l
 // GFX11: encoding: [0x6b,0x04,0x0a,0x6c]
 
-v_fmac_f16 v5, ttmp15, v2
+v_fmac_f16 v5.l, ttmp15, v2.l
 // GFX11: encoding: [0x7b,0x04,0x0a,0x6c]
 
-v_fmac_f16 v5, m0, v2
+v_fmac_f16 v5.l, m0, v2.l
 // GFX11: encoding: [0x7d,0x04,0x0a,0x6c]
 
-v_fmac_f16 v5, exec_lo, v2
+v_fmac_f16 v5.l, exec_lo, v2.l
 // GFX11: encoding: [0x7e,0x04,0x0a,0x6c]
 
-v_fmac_f16 v5, exec_hi, v2
+v_fmac_f16 v5.l, exec_hi, v2.l
 // GFX11: encoding: [0x7f,0x04,0x0a,0x6c]
 
-v_fmac_f16 v5, null, v2
+v_fmac_f16 v5.l, null, v2.l
 // GFX11: encoding: [0x7c,0x04,0x0a,0x6c]
 
-v_fmac_f16 v5, -1, v2
+v_fmac_f16 v5.l, -1, v2.l
 // GFX11: encoding: [0xc1,0x04,0x0a,0x6c]
 
-v_fmac_f16 v5, 0.5, v2
-// GFX11: encoding: [0xf0,0x04,0x0a,0x6c]
+v_fmac_f16 v127.l, 0.5, v127.l
+// GFX11: encoding: [0xf0,0xfe,0xfe,0x6c]
 
-v_fmac_f16 v5, src_scc, v2
-// GFX11: encoding: [0xfd,0x04,0x0a,0x6c]
+v_fmac_f16 v5.h, src_scc, v2.h
+// GFX11: encoding: [0xfd,0x04,0x0b,0x6d]
 
-v_fmac_f16 v127, 0xfe0b, v127
-// GFX11: encoding: [0xff,0xfe,0xfe,0x6c,0x0b,0xfe,0x00,0x00]
+v_fmac_f16 v127.h, 0xfe0b, v127.h
+// GFX11: encoding: [0xff,0xfe,0xff,0x6d,0x0b,0xfe,0x00,0x00]
 
 v_fmac_f32 v5, v1, v2
 // GFX11: encoding: [0x01,0x05,0x0a,0x56]
@@ -918,50 +930,56 @@ v_fmac_legacy_f32 v5, src_scc, v2
 v_fmac_legacy_f32 v255, 0xaf123456, v255
 // GFX11: encoding: [0xff,0xfe,0xff,0x0d,0x56,0x34,0x12,0xaf]
 
-v_fmamk_f16 v5, v1, 0xfe0b, v3
+v_fmamk_f16 v5.l, v1.l, 0xfe0b, v3.l
 // GFX11: encoding: [0x01,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, v127, 0xfe0b, v3
+v_fmamk_f16 v5.l, v127.l, 0xfe0b, v3.l
 // GFX11: encoding: [0x7f,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, s1, 0xfe0b, v3
+v_fmamk_f16 v5.l, v1.h, 0xfe0b, v3.l
+// GFX11: encoding: [0x81,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
+
+v_fmamk_f16 v5.l, v127.h, 0xfe0b, v3.l
+// GFX11: encoding: [0xff,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
+
+v_fmamk_f16 v5.l, s1, 0xfe0b, v3.l
 // GFX11: encoding: [0x01,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, s105, 0xfe0b, v3
+v_fmamk_f16 v5.l, s105, 0xfe0b, v3.l
 // GFX11: encoding: [0x69,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, vcc_lo, 0xfe0b, v3
+v_fmamk_f16 v5.l, vcc_lo, 0xfe0b, v3.l
 // GFX11: encoding: [0x6a,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, vcc_hi, 0xfe0b, v3
+v_fmamk_f16 v5.l, vcc_hi, 0xfe0b, v3.l
 // GFX11: encoding: [0x6b,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, ttmp15, 0xfe0b, v3
+v_fmamk_f16 v5.l, ttmp15, 0xfe0b, v3.l
 // GFX11: encoding: [0x7b,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, m0, 0xfe0b, v3
+v_fmamk_f16 v5.l, m0, 0xfe0b, v3.l
 // GFX11: encoding: [0x7d,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, exec_lo, 0xfe0b, v3
+v_fmamk_f16 v5.l, exec_lo, 0xfe0b, v3.l
 // GFX11: encoding: [0x7e,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, exec_hi, 0xfe0b, v3
+v_fmamk_f16 v5.l, exec_hi, 0xfe0b, v3.l
 // GFX11: encoding: [0x7f,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, null, 0xfe0b, v3
+v_fmamk_f16 v5.l, null, 0xfe0b, v3.l
 // GFX11: encoding: [0x7c,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, -1, 0xfe0b, v3
+v_fmamk_f16 v5.l, -1, 0xfe0b, v3.l
 // GFX11: encoding: [0xc1,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, 0.5, 0xfe0b, v3
-// GFX11: encoding: [0xf0,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
+v_fmamk_f16 v127.l, 0.5, 0xfe0b, v127.l
+// GFX11: encoding: [0xf0,0xfe,0xfe,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, src_scc, 0xfe0b, v3
-// GFX11: encoding: [0xfd,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
+v_fmamk_f16 v5.h, src_scc, 0xfe0b, v3.h
+// GFX11: encoding: [0xfd,0x06,0x0b,0x6f,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v127, 0xfe0b, 0xfe0b, v127
-// GFX11: encoding: [0xff,0xfe,0xfe,0x6e,0x0b,0xfe,0x00,0x00]
+v_fmamk_f16 v127.h, 0xfe0b, 0xfe0b, v127.h
+// GFX11: encoding: [0xff,0xfe,0xff,0x6f,0x0b,0xfe,0x00,0x00]
 
 v_fmamk_f32 v5, v1, 0xaf123456, v3
 // GFX11: encoding: [0x01,0x07,0x0a,0x58,0x56,0x34,0x12,0xaf]
diff --git a/llvm/test/MC/AMDGPU/gfx11_asm_vop2_dpp16.s b/llvm/test/MC/AMDGPU/gfx11_asm_vop2_dpp16.s
index 3eff00bb96e475..f40278cb9c42e7 100644
--- a/llvm/test/MC/AMDGPU/gfx11_asm_vop2_dpp16.s
+++ b/llvm/test/MC/AMDGPU/gfx11_asm_vop2_dpp16.s
@@ -629,47 +629,47 @@ v_dot2c_f32_f16 v5, v1, v2 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 f
 v_dot2c_f32_f16 v255, -|v255|, -|v255| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
 // GFX11: encoding: [0xfa,0xfe,0xff,0x05,0xff,0x6f,0xf5,0x30]
 
-v_fmac_f16 v5, v1, v2 quad_perm:[3,2,1,0]
+v_fmac_f16 v5.l, v1.l, v2.l quad_perm:[3,2,1,0]
 // GFX11: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x1b,0x00,0xff]
 
-v_fmac_f16 v5, v1, v2 quad_perm:[0,1,2,3]
+v_fmac_f16 v5.l, v1.l, v2.l quad_perm:[0,1,2,3]
 // GFX11: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0xe4,0x00,0xff]
 
-v_fmac_f16 v5, v1, v2 row_mirror
+v_fmac_f16 v5.l, v1.l, v2.l row_mirror
 // GFX11: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x40,0x01,0xff]
 
-v_fmac_f16 v5, v1, v2 row_half_mirror
+v_fmac_f16 v5.l, v1.l, v2.l row_half_mirror
 // GFX11: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x41,0x01,0xff]
 
-v_fmac_f16 v5, v1, v2 row_shl:1
+v_fmac_f16 v5.l, v1.l, v2.l row_shl:1
 // GFX11: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x01,0x01,0xff]
 
-v_fmac_f16 v5, v1, v2 row_shl:15
+v_fmac_f16 v5.l, v...
[truncated]

@llvmbot
Copy link
Collaborator

llvmbot commented Nov 6, 2024

@llvm/pr-subscribers-backend-amdgpu

Author: Brox Chen (broxigarchen)

Changes

Support true16 format for VOP2 instructions in MC

This patch updates the true16 and fake16 vop_profile for the following instructions and update the asm/dasm tests:
v_fmac_f16
v_fmamk_f16
v_fmaak_f16

Patch is 101.14 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/115233.diff

17 Files Affected:

  • (modified) llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp (+19)
  • (modified) llvm/lib/Target/AMDGPU/SIRegisterInfo.td (+9-1)
  • (modified) llvm/lib/Target/AMDGPU/VOP2Instructions.td (+70-6)
  • (modified) llvm/test/MC/AMDGPU/gfx11_asm_vop2.s (+72-54)
  • (modified) llvm/test/MC/AMDGPU/gfx11_asm_vop2_dpp16.s (+17-17)
  • (modified) llvm/test/MC/AMDGPU/gfx11_asm_vop2_dpp8.s (+8-5)
  • (modified) llvm/test/MC/AMDGPU/gfx11_asm_vop2_t16_err.s (+75-30)
  • (modified) llvm/test/MC/AMDGPU/gfx12_asm_vop2.s (+81-45)
  • (modified) llvm/test/MC/AMDGPU/gfx12_asm_vop2_dpp16.s (+20-14)
  • (modified) llvm/test/MC/AMDGPU/gfx12_asm_vop2_dpp8.s (+9-3)
  • (modified) llvm/test/MC/AMDGPU/gfx12_asm_vop2_t16_err.s (+75-30)
  • (modified) llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop2.txt (+117-54)
  • (modified) llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop2_dpp16.txt (+31-17)
  • (modified) llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop2_dpp8.txt (+12-3)
  • (modified) llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop2.txt (+132-45)
  • (modified) llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop2_dpp16.txt (+36-14)
  • (modified) llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop2_dpp8.txt (+11-2)
diff --git a/llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp b/llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp
index fdef9865b82c06..f4eec66fbb6157 100644
--- a/llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp
+++ b/llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp
@@ -345,6 +345,25 @@ static DecodeStatus decodeOperand_VSrcT16_Lo128(MCInst &Inst, unsigned Imm,
                               (AMDGPU::OperandSemantics)OperandSemantics));
 }
 
+template <AMDGPUDisassembler::OpWidthTy OpWidth, unsigned ImmWidth,
+          unsigned OperandSemantics>
+static DecodeStatus
+decodeOperand_VSrcT16_Lo128_Deferred(MCInst &Inst, unsigned Imm,
+                                     uint64_t /*Addr*/,
+                                     const MCDisassembler *Decoder) {
+  const auto *DAsm = static_cast<const AMDGPUDisassembler *>(Decoder);
+  assert(isUInt<9>(Imm) && "9-bit encoding expected");
+
+  if (Imm & AMDGPU::EncValues::IS_VGPR) {
+    bool IsHi = Imm & (1 << 7);
+    unsigned RegIdx = Imm & 0x7f;
+    return addOperand(Inst, DAsm->createVGPR16Operand(RegIdx, IsHi));
+  }
+  return addOperand(Inst, DAsm->decodeNonVGPRSrcOp(
+                              OpWidth, Imm & 0xFF, true, ImmWidth,
+                              (AMDGPU::OperandSemantics)OperandSemantics));
+}
+
 template <AMDGPUDisassembler::OpWidthTy OpWidth, unsigned ImmWidth,
           unsigned OperandSemantics>
 static DecodeStatus decodeOperand_VSrcT16(MCInst &Inst, unsigned Imm,
diff --git a/llvm/lib/Target/AMDGPU/SIRegisterInfo.td b/llvm/lib/Target/AMDGPU/SIRegisterInfo.td
index 852430129251c6..49be750acab960 100644
--- a/llvm/lib/Target/AMDGPU/SIRegisterInfo.td
+++ b/llvm/lib/Target/AMDGPU/SIRegisterInfo.td
@@ -1161,7 +1161,8 @@ def SSrcOrLds_b32 : SrcRegOrImm9 <SRegOrLds_32, "OPW32", "OPERAND_REG_IMM_INT32"
 class SrcRegOrImmDeferred9<RegisterClass regClass, string opWidth,
                            string operandType, int immWidth, int OperandSemantics>
     : RegOrImmOperand<regClass, operandType> {
-  let DecoderMethod = "decodeSrcRegOrImmDeferred9<AMDGPUDisassembler::" #
+  string DecoderMethodName = "decodeSrcRegOrImmDeferred9";
+  let DecoderMethod = DecoderMethodName # "<AMDGPUDisassembler::" #
                       opWidth # ", " # immWidth # ", " # OperandSemantics # ">";
 }
 
@@ -1222,6 +1223,13 @@ def VSrc_bf16_Deferred : SrcRegOrImmDeferred9<VS_32, "OPW16", "OPERAND_REG_IMM_B
 def VSrc_f16_Deferred : SrcRegOrImmDeferred9<VS_32, "OPW16", "OPERAND_REG_IMM_FP16_DEFERRED", 16, OperandSemantics.FP16>;
 def VSrc_f32_Deferred : SrcRegOrImmDeferred9<VS_32, "OPW32", "OPERAND_REG_IMM_FP32_DEFERRED", 32, OperandSemantics.FP32>;
 
+// True 16 Operands
+def VSrcT_f16_Lo128_Deferred : SrcRegOrImmDeferred9<VS_16_Lo128, "OPW16",
+                                                   "OPERAND_REG_IMM_FP16_DEFERRED", 16, OperandSemantics.FP16> {
+  let DecoderMethodName = "decodeOperand_VSrcT16_Lo128_Deferred";
+  let EncoderMethod = "getMachineOpValueT16Lo128";
+}
+
 def VSrcFake16_bf16_Lo128_Deferred
   : SrcRegOrImmDeferred9<VS_32_Lo128, "OPW16", "OPERAND_REG_IMM_BF16_DEFERRED", 16, OperandSemantics.BF16>;
 def VSrcFake16_f16_Lo128_Deferred
diff --git a/llvm/lib/Target/AMDGPU/VOP2Instructions.td b/llvm/lib/Target/AMDGPU/VOP2Instructions.td
index 925b60561c9d68..df41e3ceefe9e1 100644
--- a/llvm/lib/Target/AMDGPU/VOP2Instructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP2Instructions.td
@@ -374,6 +374,12 @@ class VOP_MADAK <ValueType vt> : VOP_MADK_Base<vt> {
 }
 
 def VOP_MADAK_F16 : VOP_MADAK <f16>;
+def VOP_MADAK_F16_t16 : VOP_MADAK <f16> {
+  let IsTrue16 = 1;
+  let IsRealTrue16 = 1;
+  let DstRC = getVALUDstForVT<DstVT, 1/*IsTrue16*/, 0/*IsVOP3Encoding*/>.ret;
+  let Ins32 = (ins VSrcT_f16_Lo128_Deferred:$src0, VGPRSrc_16_Lo128:$src1, ImmOpType:$imm);
+}
 def VOP_MADAK_F16_fake16 : VOP_MADAK <f16> {
   let IsTrue16 = 1;
   let DstRC = getVALUDstForVT_fake16<DstVT>.ret;
@@ -399,6 +405,12 @@ class VOP_MADMK <ValueType vt> : VOP_MADK_Base<vt> {
 }
 
 def VOP_MADMK_F16 : VOP_MADMK <f16>;
+def VOP_MADMK_F16_t16 : VOP_MADMK <f16> {
+  let IsTrue16 = 1;
+  let IsRealTrue16 = 1;
+  let DstRC = getVALUDstForVT<DstVT, 1/*IsTrue16*/, 0/*IsVOP3Encoding*/>.ret;
+  let Ins32 = (ins VSrcT_f16_Lo128_Deferred:$src0, ImmOpType:$imm, VGPRSrc_16_Lo128:$src1);
+}
 def VOP_MADMK_F16_fake16 : VOP_MADMK <f16> {
   let IsTrue16 = 1;
   let DstRC = getVALUDstForVT_fake16<DstVT>.ret;
@@ -467,6 +479,42 @@ class VOP_MAC <ValueType vt0, ValueType vt1=vt0> : VOPProfile <[vt0, vt1, vt1, v
 }
 
 def VOP_MAC_F16 : VOP_MAC <f16>;
+def VOP_MAC_F16_t16 : VOP_MAC <f16> {
+  let IsTrue16 = 1;
+  let IsRealTrue16 = 1;
+  let HasOpSel = 1;
+  let DstRC = VOPDstOperand_t16Lo128;
+  let Src0RC32 = getVOPSrc0ForVT<Src0VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret;
+  let Src1RC32 = getVregSrcForVT<Src1VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret;
+  let Ins32 = (ins Src0RC32:$src0, Src1RC32:$src1, getVregSrcForVT<Src2VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret:$src2);
+  let Src0DPP = getVregSrcForVT<Src0VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret;
+  let Src1DPP = getVregSrcForVT<Src1VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret;
+  let Src2DPP = getVregSrcForVT<Src2VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret;
+  let Src0ModDPP = getSrcModDPP_t16<Src0VT, 0/*IsFake16*/>.ret;
+  let Src1ModDPP = getSrcModDPP_t16<Src1VT, 0/*IsFake16*/>.ret;
+  let Src2ModDPP = getSrcModDPP_t16<Src2VT, 0/*IsFake16*/>.ret;
+  let InsDPP = (ins Src0ModDPP:$src0_modifiers, Src0DPP:$src0,
+                    Src1ModDPP:$src1_modifiers, Src1DPP:$src1,
+                    getVregSrcForVT<Src2VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret:$src2, // stub argument
+                    dpp_ctrl:$dpp_ctrl, DppRowMask:$row_mask,
+                    DppBankMask:$bank_mask, DppBoundCtrl:$bound_ctrl);
+  let InsDPP8 = (ins Src0ModDPP:$src0_modifiers, Src0DPP:$src0,
+                     Src1ModDPP:$src1_modifiers, Src1DPP:$src1,
+                     getVregSrcForVT<Src2VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret:$src2, // stub argument
+                     dpp8:$dpp8, Dpp8FI:$fi);
+  let DstRC64 = getVALUDstForVT<DstVT, 1/*IsTrue*/, 1/*IsVOP3Encoding*/>.ret;
+  let Src0RC64 = getVOP3SrcForVT<Src0VT, 1/*IsTrue16*/>.ret;
+  let Src1RC64 = getVOP3SrcForVT<Src1VT, 1/*IsTrue16*/>.ret;
+  let Src0VOP3DPP = VGPRSrc_16;
+  let Src1VOP3DPP = getVOP3DPPSrcForVT<Src1VT, 0/*IsFake16*/>.ret;
+  let Src2VOP3DPP = getVOP3DPPSrcForVT<Src2VT, 0/*IsFake16*/>.ret;
+  let Src0ModVOP3DPP = getSrc0ModVOP3DPP<Src0VT, DstVT, 0/*IsFake16*/>.ret;
+  let Src1ModVOP3DPP = getSrcModVOP3DPP<Src1VT, 0/*IsFake16*/>.ret;
+  let Src2ModVOP3DPP = getSrcModVOP3DPP<Src2VT, 0/*IsFake16*/>.ret;
+  let Src0Mod = getSrc0Mod<Src0VT, DstVT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret;
+  let Src1Mod = getSrcMod<Src1VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret;
+  let Src2Mod = getSrcMod<Src2VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret;
+}
 def VOP_MAC_F16_fake16 : VOP_MAC <f16> {
   let IsTrue16 = 1;
   let DstRC = getVALUDstForVT_fake16<DstVT>.ret;
@@ -998,6 +1046,9 @@ let FPDPRounding = 1, isReMaterializable = 1, FixedSize = 1 in {
 let SubtargetPredicate = isGFX10Plus, True16Predicate = NotHasTrue16BitInsts in {
   def V_FMAMK_F16 : VOP2_Pseudo <"v_fmamk_f16", VOP_MADMK_F16, [], "">;
 }
+let True16Predicate = UseRealTrue16Insts in {
+  def V_FMAMK_F16_t16 : VOP2_Pseudo <"v_fmamk_f16_t16", VOP_MADMK_F16_t16, [], "">;
+}
 let True16Predicate = UseFakeTrue16Insts in {
   def V_FMAMK_F16_fake16 : VOP2_Pseudo <"v_fmamk_f16_fake16", VOP_MADMK_F16_fake16, [], "">;
 }
@@ -1006,6 +1057,9 @@ let isCommutable = 1 in {
 let SubtargetPredicate = isGFX10Plus, True16Predicate = NotHasTrue16BitInsts in {
   def V_FMAAK_F16 : VOP2_Pseudo <"v_fmaak_f16", VOP_MADAK_F16, [], "">;
 }
+let True16Predicate = UseRealTrue16Insts in {
+  def V_FMAAK_F16_t16 : VOP2_Pseudo <"v_fmaak_f16_t16", VOP_MADAK_F16_t16, [], "">;
+}
 let True16Predicate = UseFakeTrue16Insts in {
   def V_FMAAK_F16_fake16 : VOP2_Pseudo <"v_fmaak_f16_fake16", VOP_MADAK_F16_fake16, [], "">;
 }
@@ -1020,6 +1074,9 @@ let SubtargetPredicate = isGFX10Plus in {
 let True16Predicate = NotHasTrue16BitInsts in {
   defm V_FMAC_F16 : VOP2Inst <"v_fmac_f16", VOP_MAC_F16>;
 }
+let True16Predicate = UseRealTrue16Insts in {
+  defm V_FMAC_F16_t16 : VOP2Inst <"v_fmac_f16_t16", VOP_MAC_F16_t16>;
+}
 let True16Predicate = UseFakeTrue16Insts in {
   defm V_FMAC_F16_fake16 : VOP2Inst <"v_fmac_f16_fake16", VOP_MAC_F16_fake16>;
 }
@@ -1692,8 +1749,8 @@ multiclass VOP3Only_Realtriple_t16_gfx11_gfx12<bits<10> op, string asmName, stri
   VOP3_Realtriple_t16_gfx12<op, asmName, OpName, "", /*IsSingle*/1>;
 
 multiclass VOP3Only_Realtriple_t16_and_fake16_gfx11_gfx12<bits<10> op, string asmName, string OpName = NAME> {
-  defm OpName#"_t16": VOP3Only_Realtriple_t16_gfx11_gfx12<op, asmName, OpName#"_t16">;
-  defm OpName#"_fake16": VOP3Only_Realtriple_t16_gfx11_gfx12<op, asmName, OpName#"_fake16">;
+  defm _t16: VOP3Only_Realtriple_t16_gfx11_gfx12<op, asmName, OpName#"_t16">;
+  defm _fake16: VOP3Only_Realtriple_t16_gfx11_gfx12<op, asmName, OpName#"_fake16">;
 }
 
 multiclass VOP3beOnly_Realtriple_gfx11_gfx12<bits<10> op> :
@@ -1712,7 +1769,14 @@ multiclass VOP2Only_Real_MADK_t16_and_fake16_gfx11_gfx12<bits<6> op, string asmN
 
 multiclass VOP2_Real_FULL_t16_gfx11_gfx12<bits<6> op, string asmName,
                                           string opName = NAME> :
-  VOP2_Real_FULL_with_name_gfx11_gfx12<op, opName, asmName>;
+  VOP2_Real_FULL_with_name<GFX11Gen, op, opName, asmName>,
+  VOP2_Real_FULL_with_name<GFX12Gen, op, opName, asmName>;
+
+multiclass VOP2_Real_FULL_t16_and_fake16_gfx11_gfx12<bits<6> op, string asmName,
+                                          string opName = NAME> {
+  defm _t16:    VOP2_Real_FULL_t16_gfx11_gfx12<op, asmName, opName#"_t16">;
+  defm _fake16: VOP2_Real_FULL_t16_gfx11_gfx12<op, asmName, opName#"_fake16">;
+}
 
 multiclass VOP2_Real_FULL_gfx11<bits<6> op> :
   VOP2_Real_FULL<GFX11Gen, op>;
@@ -1747,15 +1811,15 @@ defm V_SUBREV_F16_t16      : VOP2_Real_FULL_t16_gfx11_gfx12<0x034, "v_subrev_f16
 defm V_SUBREV_F16_fake16   : VOP2_Real_FULL_t16_gfx11_gfx12<0x034, "v_subrev_f16">;
 defm V_MUL_F16_t16         : VOP2_Real_FULL_t16_gfx11_gfx12<0x035, "v_mul_f16">;
 defm V_MUL_F16_fake16      : VOP2_Real_FULL_t16_gfx11_gfx12<0x035, "v_mul_f16">;
-defm V_FMAC_F16_fake16     : VOP2_Real_FULL_t16_gfx11_gfx12<0x036, "v_fmac_f16">;
+defm V_FMAC_F16            : VOP2_Real_FULL_t16_and_fake16_gfx11_gfx12<0x036, "v_fmac_f16">;
 defm V_LDEXP_F16_t16       : VOP2_Real_FULL_t16_gfx11_gfx12<0x03b, "v_ldexp_f16">;
 defm V_LDEXP_F16_fake16    : VOP2_Real_FULL_t16_gfx11_gfx12<0x03b, "v_ldexp_f16">;
 defm V_MAX_F16_t16         : VOP2_Real_FULL_t16_gfx11<0x039, "v_max_f16">;
 defm V_MAX_F16_fake16      : VOP2_Real_FULL_t16_gfx11<0x039, "v_max_f16">;
 defm V_MIN_F16_t16         : VOP2_Real_FULL_t16_gfx11<0x03a, "v_min_f16">;
 defm V_MIN_F16_fake16      : VOP2_Real_FULL_t16_gfx11<0x03a, "v_min_f16">;
-defm V_FMAMK_F16_fake16    : VOP2Only_Real_MADK_t16_gfx11_gfx12<0x037, "v_fmamk_f16">;
-defm V_FMAAK_F16_fake16    : VOP2Only_Real_MADK_t16_gfx11_gfx12<0x038, "v_fmaak_f16">;
+defm V_FMAMK_F16           : VOP2Only_Real_MADK_t16_and_fake16_gfx11_gfx12<0x037, "v_fmamk_f16">;
+defm V_FMAAK_F16           : VOP2Only_Real_MADK_t16_and_fake16_gfx11_gfx12<0x038, "v_fmaak_f16">;
 
 // VOP3 only.
 defm V_CNDMASK_B16         : VOP3Only_Realtriple_gfx11_gfx12<0x25d>;
diff --git a/llvm/test/MC/AMDGPU/gfx11_asm_vop2.s b/llvm/test/MC/AMDGPU/gfx11_asm_vop2.s
index 2a4b3ea2017013..0f2852fc531ed4 100644
--- a/llvm/test/MC/AMDGPU/gfx11_asm_vop2.s
+++ b/llvm/test/MC/AMDGPU/gfx11_asm_vop2.s
@@ -648,50 +648,56 @@ v_dot2c_f32_f16 v5, src_scc, v2
 v_dot2c_f32_f16 v255, 0xfe0b, v255
 // GFX11: encoding: [0xff,0xfe,0xff,0x05,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, v1, v2, 0xfe0b
+v_fmaak_f16 v5.l, v1.l, v2.l, 0xfe0b
 // GFX11: encoding: [0x01,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, v127, v2, 0xfe0b
+v_fmaak_f16 v5.l, v127.l, v2.l, 0xfe0b
 // GFX11: encoding: [0x7f,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, s1, v2, 0xfe0b
+v_fmaak_f16 v5.l, v1.h, v2.l, 0xfe0b
+// GFX11: encoding: [0x81,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00]
+
+v_fmaak_f16 v5.l, v127.h, v2.l, 0xfe0b
+// GFX11: encoding: [0xff,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00]
+
+v_fmaak_f16 v5.l, s1, v2.l, 0xfe0b
 // GFX11: encoding: [0x01,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, s105, v2, 0xfe0b
+v_fmaak_f16 v5.l, s105, v2.l, 0xfe0b
 // GFX11: encoding: [0x69,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, vcc_lo, v2, 0xfe0b
+v_fmaak_f16 v5.l, vcc_lo, v2.l, 0xfe0b
 // GFX11: encoding: [0x6a,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, vcc_hi, v2, 0xfe0b
+v_fmaak_f16 v5.l, vcc_hi, v2.l, 0xfe0b
 // GFX11: encoding: [0x6b,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, ttmp15, v2, 0xfe0b
+v_fmaak_f16 v5.l, ttmp15, v2.l, 0xfe0b
 // GFX11: encoding: [0x7b,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, m0, v2, 0xfe0b
+v_fmaak_f16 v5.l, m0, v2.l, 0xfe0b
 // GFX11: encoding: [0x7d,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, exec_lo, v2, 0xfe0b
+v_fmaak_f16 v5.l, exec_lo, v2.l, 0xfe0b
 // GFX11: encoding: [0x7e,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, exec_hi, v2, 0xfe0b
+v_fmaak_f16 v5.l, exec_hi, v2.l, 0xfe0b
 // GFX11: encoding: [0x7f,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, null, v2, 0xfe0b
+v_fmaak_f16 v5.l, null, v2.l, 0xfe0b
 // GFX11: encoding: [0x7c,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, -1, v2, 0xfe0b
+v_fmaak_f16 v5.l, -1, v2.l, 0xfe0b
 // GFX11: encoding: [0xc1,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, 0.5, v2, 0xfe0b
-// GFX11: encoding: [0xf0,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00]
+v_fmaak_f16 v127.l, 0.5, v127.l, 0xfe0b
+// GFX11: encoding: [0xf0,0xfe,0xfe,0x70,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v5, src_scc, v2, 0xfe0b
-// GFX11: encoding: [0xfd,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00]
+v_fmaak_f16 v5.h, src_scc, v2.h, 0xfe0b
+// GFX11: encoding: [0xfd,0x04,0x0b,0x71,0x0b,0xfe,0x00,0x00]
 
-v_fmaak_f16 v127, 0xfe0b, v127, 0xfe0b
-// GFX11: encoding: [0xff,0xfe,0xfe,0x70,0x0b,0xfe,0x00,0x00]
+v_fmaak_f16 v127.h, 0xfe0b, v127.h, 0xfe0b
+// GFX11: encoding: [0xff,0xfe,0xff,0x71,0x0b,0xfe,0x00,0x00]
 
 v_fmaak_f32 v5, v1, v2, 0xaf123456
 // GFX11: encoding: [0x01,0x05,0x0a,0x5a,0x56,0x34,0x12,0xaf]
@@ -783,50 +789,56 @@ v_fmac_dx9_zero_f32 v5, src_scc, v2
 v_fmac_dx9_zero_f32 v255, 0xaf123456, v255
 // GFX11: encoding: [0xff,0xfe,0xff,0x0d,0x56,0x34,0x12,0xaf]
 
-v_fmac_f16 v5, v1, v2
+v_fmac_f16 v5.l, v1.l, v2.l
 // GFX11: encoding: [0x01,0x05,0x0a,0x6c]
 
-v_fmac_f16 v5, v127, v2
+v_fmac_f16 v5.l, v127.l, v2.l
 // GFX11: encoding: [0x7f,0x05,0x0a,0x6c]
 
-v_fmac_f16 v5, s1, v2
+v_fmac_f16 v5.l, v1.h, v2.l
+// GFX11: encoding: [0x81,0x05,0x0a,0x6c]
+
+v_fmac_f16 v5.l, v127.h, v2.l
+// GFX11: encoding: [0xff,0x05,0x0a,0x6c]
+
+v_fmac_f16 v5.l, s1, v2.l
 // GFX11: encoding: [0x01,0x04,0x0a,0x6c]
 
-v_fmac_f16 v5, s105, v2
+v_fmac_f16 v5.l, s105, v2.l
 // GFX11: encoding: [0x69,0x04,0x0a,0x6c]
 
-v_fmac_f16 v5, vcc_lo, v2
+v_fmac_f16 v5.l, vcc_lo, v2.l
 // GFX11: encoding: [0x6a,0x04,0x0a,0x6c]
 
-v_fmac_f16 v5, vcc_hi, v2
+v_fmac_f16 v5.l, vcc_hi, v2.l
 // GFX11: encoding: [0x6b,0x04,0x0a,0x6c]
 
-v_fmac_f16 v5, ttmp15, v2
+v_fmac_f16 v5.l, ttmp15, v2.l
 // GFX11: encoding: [0x7b,0x04,0x0a,0x6c]
 
-v_fmac_f16 v5, m0, v2
+v_fmac_f16 v5.l, m0, v2.l
 // GFX11: encoding: [0x7d,0x04,0x0a,0x6c]
 
-v_fmac_f16 v5, exec_lo, v2
+v_fmac_f16 v5.l, exec_lo, v2.l
 // GFX11: encoding: [0x7e,0x04,0x0a,0x6c]
 
-v_fmac_f16 v5, exec_hi, v2
+v_fmac_f16 v5.l, exec_hi, v2.l
 // GFX11: encoding: [0x7f,0x04,0x0a,0x6c]
 
-v_fmac_f16 v5, null, v2
+v_fmac_f16 v5.l, null, v2.l
 // GFX11: encoding: [0x7c,0x04,0x0a,0x6c]
 
-v_fmac_f16 v5, -1, v2
+v_fmac_f16 v5.l, -1, v2.l
 // GFX11: encoding: [0xc1,0x04,0x0a,0x6c]
 
-v_fmac_f16 v5, 0.5, v2
-// GFX11: encoding: [0xf0,0x04,0x0a,0x6c]
+v_fmac_f16 v127.l, 0.5, v127.l
+// GFX11: encoding: [0xf0,0xfe,0xfe,0x6c]
 
-v_fmac_f16 v5, src_scc, v2
-// GFX11: encoding: [0xfd,0x04,0x0a,0x6c]
+v_fmac_f16 v5.h, src_scc, v2.h
+// GFX11: encoding: [0xfd,0x04,0x0b,0x6d]
 
-v_fmac_f16 v127, 0xfe0b, v127
-// GFX11: encoding: [0xff,0xfe,0xfe,0x6c,0x0b,0xfe,0x00,0x00]
+v_fmac_f16 v127.h, 0xfe0b, v127.h
+// GFX11: encoding: [0xff,0xfe,0xff,0x6d,0x0b,0xfe,0x00,0x00]
 
 v_fmac_f32 v5, v1, v2
 // GFX11: encoding: [0x01,0x05,0x0a,0x56]
@@ -918,50 +930,56 @@ v_fmac_legacy_f32 v5, src_scc, v2
 v_fmac_legacy_f32 v255, 0xaf123456, v255
 // GFX11: encoding: [0xff,0xfe,0xff,0x0d,0x56,0x34,0x12,0xaf]
 
-v_fmamk_f16 v5, v1, 0xfe0b, v3
+v_fmamk_f16 v5.l, v1.l, 0xfe0b, v3.l
 // GFX11: encoding: [0x01,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, v127, 0xfe0b, v3
+v_fmamk_f16 v5.l, v127.l, 0xfe0b, v3.l
 // GFX11: encoding: [0x7f,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, s1, 0xfe0b, v3
+v_fmamk_f16 v5.l, v1.h, 0xfe0b, v3.l
+// GFX11: encoding: [0x81,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
+
+v_fmamk_f16 v5.l, v127.h, 0xfe0b, v3.l
+// GFX11: encoding: [0xff,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
+
+v_fmamk_f16 v5.l, s1, 0xfe0b, v3.l
 // GFX11: encoding: [0x01,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, s105, 0xfe0b, v3
+v_fmamk_f16 v5.l, s105, 0xfe0b, v3.l
 // GFX11: encoding: [0x69,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, vcc_lo, 0xfe0b, v3
+v_fmamk_f16 v5.l, vcc_lo, 0xfe0b, v3.l
 // GFX11: encoding: [0x6a,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, vcc_hi, 0xfe0b, v3
+v_fmamk_f16 v5.l, vcc_hi, 0xfe0b, v3.l
 // GFX11: encoding: [0x6b,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, ttmp15, 0xfe0b, v3
+v_fmamk_f16 v5.l, ttmp15, 0xfe0b, v3.l
 // GFX11: encoding: [0x7b,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, m0, 0xfe0b, v3
+v_fmamk_f16 v5.l, m0, 0xfe0b, v3.l
 // GFX11: encoding: [0x7d,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, exec_lo, 0xfe0b, v3
+v_fmamk_f16 v5.l, exec_lo, 0xfe0b, v3.l
 // GFX11: encoding: [0x7e,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, exec_hi, 0xfe0b, v3
+v_fmamk_f16 v5.l, exec_hi, 0xfe0b, v3.l
 // GFX11: encoding: [0x7f,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, null, 0xfe0b, v3
+v_fmamk_f16 v5.l, null, 0xfe0b, v3.l
 // GFX11: encoding: [0x7c,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, -1, 0xfe0b, v3
+v_fmamk_f16 v5.l, -1, 0xfe0b, v3.l
 // GFX11: encoding: [0xc1,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, 0.5, 0xfe0b, v3
-// GFX11: encoding: [0xf0,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
+v_fmamk_f16 v127.l, 0.5, 0xfe0b, v127.l
+// GFX11: encoding: [0xf0,0xfe,0xfe,0x6e,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v5, src_scc, 0xfe0b, v3
-// GFX11: encoding: [0xfd,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00]
+v_fmamk_f16 v5.h, src_scc, 0xfe0b, v3.h
+// GFX11: encoding: [0xfd,0x06,0x0b,0x6f,0x0b,0xfe,0x00,0x00]
 
-v_fmamk_f16 v127, 0xfe0b, 0xfe0b, v127
-// GFX11: encoding: [0xff,0xfe,0xfe,0x6e,0x0b,0xfe,0x00,0x00]
+v_fmamk_f16 v127.h, 0xfe0b, 0xfe0b, v127.h
+// GFX11: encoding: [0xff,0xfe,0xff,0x6f,0x0b,0xfe,0x00,0x00]
 
 v_fmamk_f32 v5, v1, 0xaf123456, v3
 // GFX11: encoding: [0x01,0x07,0x0a,0x58,0x56,0x34,0x12,0xaf]
diff --git a/llvm/test/MC/AMDGPU/gfx11_asm_vop2_dpp16.s b/llvm/test/MC/AMDGPU/gfx11_asm_vop2_dpp16.s
index 3eff00bb96e475..f40278cb9c42e7 100644
--- a/llvm/test/MC/AMDGPU/gfx11_asm_vop2_dpp16.s
+++ b/llvm/test/MC/AMDGPU/gfx11_asm_vop2_dpp16.s
@@ -629,47 +629,47 @@ v_dot2c_f32_f16 v5, v1, v2 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 f
 v_dot2c_f32_f16 v255, -|v255|, -|v255| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
 // GFX11: encoding: [0xfa,0xfe,0xff,0x05,0xff,0x6f,0xf5,0x30]
 
-v_fmac_f16 v5, v1, v2 quad_perm:[3,2,1,0]
+v_fmac_f16 v5.l, v1.l, v2.l quad_perm:[3,2,1,0]
 // GFX11: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x1b,0x00,0xff]
 
-v_fmac_f16 v5, v1, v2 quad_perm:[0,1,2,3]
+v_fmac_f16 v5.l, v1.l, v2.l quad_perm:[0,1,2,3]
 // GFX11: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0xe4,0x00,0xff]
 
-v_fmac_f16 v5, v1, v2 row_mirror
+v_fmac_f16 v5.l, v1.l, v2.l row_mirror
 // GFX11: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x40,0x01,0xff]
 
-v_fmac_f16 v5, v1, v2 row_half_mirror
+v_fmac_f16 v5.l, v1.l, v2.l row_half_mirror
 // GFX11: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x41,0x01,0xff]
 
-v_fmac_f16 v5, v1, v2 row_shl:1
+v_fmac_f16 v5.l, v1.l, v2.l row_shl:1
 // GFX11: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x01,0x01,0xff]
 
-v_fmac_f16 v5, v1, v2 row_shl:15
+v_fmac_f16 v5.l, v...
[truncated]

arsenm
arsenm reviewed Nov 7, 2024
View reviewed changes
llvm/lib/Target/AMDGPU/VOP2Instructions.td
@@ -467,6 +479,42 @@ class VOP_MAC <ValueType vt0, ValueType vt1=vt0> : VOPProfile <[vt0, vt1, vt1, v
}

def VOP_MAC_F16 : VOP_MAC <f16>;
def VOP_MAC_F16_t16 : VOP_MAC <f16> {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is a lot of copy paste. Is there a way to avoid replicating so many lets?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@arsenm arsenm arsenm left review comments

@mariusz-sikora-at-amd mariusz-sikora-at-amd Awaiting requested review from mariusz-sikora-at-amd

@Sisyph Sisyph Awaiting requested review from Sisyph

@kosarev kosarev Awaiting requested review from kosarev

Assignees
No one assigned
Labels
backend:AMDGPU mc Machine (object) code
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@broxigarchen @llvmbot @arsenm