[AMDGPU] SDWA: add disassembler support for GFX9

[SamWot]

Summary

Added decoder methods and tests

[artem.tamazov]

Good.

lib/Target/AMDGPU/SIDefines.h

302

Perhaps too much Sdwa's here - I would remove this prefix from everything defined in SDWA namespace. Just a recommendation.

[SamWot]

Fixed linux macros

lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.h

public:
 
   static MCOperand decodeIntImmed(unsigned Imm);
   static MCOperand decodeFPImmed(OpWidthTy Width, unsigned Imm);
   MCOperand decodeLiteralConstant() const;
 
   MCOperand decodeSrcOp(const OpWidthTy Width, unsigned Val) const;
   MCOperand decodeSpecialReg32(unsigned Val) const;
   MCOperand decodeSpecialReg64(unsigned Val) const;
+
+  MCOperand decodeSDWA9Src(const OpWidthTy Width, unsigned Val) const;
+  MCOperand decodeSDWA9Src16(unsigned Val) const;
+  MCOperand decodeSDWA9Src32(unsigned Val) const;
+  MCOperand decodeSDWA9VopcDst(unsigned Val) const;
 };
 
 //===----------------------------------------------------------------------===//
 // AMDGPUSymbolizer
 //===----------------------------------------------------------------------===//
 
 class AMDGPUSymbolizer : public MCSymbolizer {
 private:

lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp

 static DecodeStatus decodeOperand_VSrcV216(MCInst &Inst,
                                          unsigned Imm,
                                          uint64_t Addr,
                                          const void *Decoder) {
   auto DAsm = static_cast<const AMDGPUDisassembler*>(Decoder);
   return addOperand(Inst, DAsm->decodeOperand_VSrcV216(Imm));
 }
 
+#define DECODE_SDWA9(DecName) \
+static DecodeStatus decodeSDWA9##DecName##(MCInst &Inst, \
+                                           unsigned Imm, \
+                                           uint64_t /*Addr*/, \
+                                           const void *Decoder) { \
+  auto DAsm = static_cast<const AMDGPUDisassembler*>(Decoder); \
+  return addOperand(Inst, DAsm->decodeSDWA9##DecName(Imm)); \
+}
+
+DECODE_SDWA9(Src32)
+DECODE_SDWA9(Src16)
+DECODE_SDWA9(VopcDst)
+
 #include "AMDGPUGenDisassemblerTables.inc"
 
 //===----------------------------------------------------------------------===//
 //
 //===----------------------------------------------------------------------===//
 
 template <typename T> static inline T eatBytes(ArrayRef<uint8_t>& Bytes) {
   assert(Bytes.size() >= sizeof(T));
do {
     // encodings
     if (Bytes.size() >= 8) {
       const uint64_t QW = eatBytes<uint64_t>(Bytes);
       Res = tryDecodeInst(DecoderTableDPP64, MI, QW, Address);
       if (Res) break;
 
       Res = tryDecodeInst(DecoderTableSDWA64, MI, QW, Address);
       if (Res) break;
+
+      Res = tryDecodeInst(DecoderTableSDWA964, MI, QW, Address);
+      if (Res) break;
     }
 
     // Reinitialize Bytes as DPP64 could have eaten too much
     Bytes = Bytes_.slice(0, MaxInstBytesNum);
 
     // Try decode 32-bit instruction
     if (Bytes.size() < 4) break;
     const uint32_t DW = eatBytes<uint32_t>(Bytes);
MCOperand AMDGPUDisassembler::decodeSpecialReg64(unsigned Val) const {
   case 108: return createRegOperand(TBA);
   case 110: return createRegOperand(TMA);
   case 126: return createRegOperand(EXEC);
   default: break;
   }
   return errOperand(Val, "unknown operand encoding " + Twine(Val));
 }
 
+MCOperand AMDGPUDisassembler::decodeSDWA9Src(const OpWidthTy Width,
+                                             unsigned Val) const {
+  using namespace AMDGPU::SDWA;
+
+  if (SDWA9EncValues::SRC_VGPR_MIN <= Val &&
+      Val <= SDWA9EncValues::SRC_VGPR_MAX) {
+    return createRegOperand(getVgprClassId(Width),
+                            Val - SDWA9EncValues::SRC_VGPR_MIN);
+  } 
+  if (SDWA9EncValues::SRC_SGPR_MIN <= Val &&
+      Val <= SDWA9EncValues::SRC_SGPR_MAX) {
+    return createSRegOperand(getSgprClassId(Width),
+                             Val - SDWA9EncValues::SRC_SGPR_MIN);
+  }
+
+  return decodeSpecialReg32(Val - SDWA9EncValues::SRC_SGPR_MIN);
+}
+
+MCOperand AMDGPUDisassembler::decodeSDWA9Src16(unsigned Val) const {
+  return decodeSDWA9Src(OPW16, Val);
+}
+
+MCOperand AMDGPUDisassembler::decodeSDWA9Src32(unsigned Val) const {
+  return decodeSDWA9Src(OPW32, Val);
+}
+
+
+MCOperand AMDGPUDisassembler::decodeSDWA9VopcDst(unsigned Val) const {
+  using namespace AMDGPU::SDWA;
+
+  if (Val & SDWA9EncValues::VOPC_DST_VCC_MASK) {
+    Val &= SDWA9EncValues::VOPC_DST_SGPR_MASK;
+    if (Val > AMDGPU::EncValues::SGPR_MAX) {
+      return decodeSpecialReg64(Val);
+    } else {
+      return createSRegOperand(getSgprClassId(OPW64), Val);
+    }
+  } else {
+    return createRegOperand(AMDGPU::VCC);
+  }
+}
+
 //===----------------------------------------------------------------------===//
 // AMDGPUSymbolizer
 //===----------------------------------------------------------------------===//
 
 // Try to find symbol name for specified label
 bool AMDGPUSymbolizer::tryAddingSymbolicOperand(MCInst &Inst,
                                 raw_ostream &/*cStream*/, int64_t Value,
                                 uint64_t /*Address*/, bool IsBranch,

lib/Target/AMDGPU/MCTargetDesc/SIMCCodeEmitter.cpp

unsigned SIMCCodeEmitter::getSOPPBrEncoding(const MCInst &MI, unsigned OpNo,
 
   return getMachineOpValue(MI, MO, Fixups, STI);
 }
 
 unsigned
 SIMCCodeEmitter::getSDWA9SrcEncoding(const MCInst &MI, unsigned OpNo,
                                      SmallVectorImpl<MCFixup> &Fixups,
                                      const MCSubtargetInfo &STI) const {
+  using namespace AMDGPU::SDWA;
+  
   uint64_t RegEnc = 0;
 
   const MCOperand &MO = MI.getOperand(OpNo);
 
   unsigned Reg = MO.getReg();
   RegEnc |= MRI.getEncodingValue(Reg);
-  RegEnc &= SDWA9_SRC_REG_MASK;
+  RegEnc &= SDWA9EncValues::SRC_VGPR_MASK;
   if (AMDGPU::isSGPR(Reg, &MRI)) {
-    RegEnc |= SDWA9_SRC_SGPR_MASK;
+    RegEnc |= SDWA9EncValues::SRC_SGPR_MASK;
   }
   return RegEnc;
 }
 
 unsigned
 SIMCCodeEmitter::getSDWA9VopcDstEncoding(const MCInst &MI, unsigned OpNo,
                                          SmallVectorImpl<MCFixup> &Fixups,
                                          const MCSubtargetInfo &STI) const {
+  using namespace AMDGPU::SDWA;
+
   uint64_t RegEnc = 0;
 
   const MCOperand &MO = MI.getOperand(OpNo);
 
   unsigned Reg = MO.getReg();
   if (Reg != AMDGPU::VCC) {
     RegEnc |= MRI.getEncodingValue(Reg);
-    RegEnc &= SDWA9_VOPC_DST_REG_MASK;
-    RegEnc |= SDWA9_VOPC_DST_VCC_MASK;
+    RegEnc &= SDWA9EncValues::VOPC_DST_SGPR_MASK;
+    RegEnc |= SDWA9EncValues::VOPC_DST_VCC_MASK;
   }
   return RegEnc;
 }
 
 uint64_t SIMCCodeEmitter::getMachineOpValue(const MCInst &MI,
                                             const MCOperand &MO,
                                        SmallVectorImpl<MCFixup> &Fixups,
                                        const MCSubtargetInfo &STI) const {

lib/Target/AMDGPU/SIDefines.h

 };
 
 enum DstUnused {
   UNUSED_PAD = 0,
   UNUSED_SEXT = 1,
   UNUSED_PRESERVE = 2,
 };
 
-#define SDWA9_SRC_SGPR_MASK 0x100
-#define SDWA9_SRC_REG_MASK 0xFF
-#define SDWA9_VOPC_DST_VCC_MASK 0x80
-#define SDWA9_VOPC_DST_REG_MASK 0x7F
+enum SDWA9EncValues{
+  SRC_SGPR_MASK = 0x100,
+  SRC_VGPR_MASK = 0xFF,
+  VOPC_DST_VCC_MASK = 0x80,
+  VOPC_DST_SGPR_MASK = 0x7F,
+
+  SRC_VGPR_MIN = 0,
+  SRC_VGPR_MAX = 255,
+  SRC_SGPR_MIN = 256,
+  SRC_SGPR_MAX = 357,
+};
 
 } // namespace SDWA
 } // namespace AMDGPU
 
 #define R_00B028_SPI_SHADER_PGM_RSRC1_PS                                0x00B028
 #define R_00B02C_SPI_SHADER_PGM_RSRC2_PS                                0x00B02C
 #define   S_00B02C_EXTRA_LDS_SIZE(x)                                  (((x) & 0xFF) << 8)
 #define R_00B128_SPI_SHADER_PGM_RSRC1_VS                                0x00B128

lib/Target/AMDGPU/SIInstrInfo.td

def ExpSrc2 : RegisterOperand<VGPR_32> {
   let ParserMatchClass = VReg32OrOffClass;
 }
 
 def ExpSrc3 : RegisterOperand<VGPR_32> {
   let PrintMethod = "printExpSrc3";
   let ParserMatchClass = VReg32OrOffClass;
 }
 
-def SDWA9Src : RegisterOperand<VS_32> {
+class SDWA9Src : RegisterOperand<VS_32> {
   let OperandNamespace = "AMDGPU";
   let OperandType = "OPERAND_SDWA9_SRC";
   let EncoderMethod = "getSDWA9SrcEncoding";
 }
 
+def SDWA9Src32 : SDWA9Src {
+  let DecoderMethod = "decodeSDWA9Src32";
+}
+
+def SDWA9Src16 : SDWA9Src {
+  let DecoderMethod = "decodeSDWA9Src16";
+}
+
 def SDWA9VopcDst : VOPDstOperand<SReg_64> {
   let OperandNamespace = "AMDGPU";
   let OperandType = "OPERAND_SDWA9_VOPC_DST";
   let EncoderMethod = "getSDWA9VopcDstEncoding";
+  let DecoderMethod = "decodeSDWA9VopcDst";
 }
 
 class NamedMatchClass<string CName, bit Optional = 1> : AsmOperandClass {
   let Name = "Imm"#CName;
   let PredicateMethod = "is"#CName;
   let ParserMethod = !if(Optional, "parseOptionalOperand", "parse"#CName);
   let RenderMethod = "addImmOperands";
   let IsOptional = Optional;

 }
 
 // Returns the vreg register class to use for source operand given VT
 class getVregSrcForVT<ValueType VT> {
   RegisterClass ret = !if(!eq(VT.Size, 128), VReg_128,
                         !if(!eq(VT.Size, 64), VReg_64, VGPR_32));
 }
 
+class getSDWA9SrcForVT <ValueType VT> {
+  RegisterOperand ret = !if(!eq(VT.Size, 16), SDWA9Src16, SDWA9Src32);
+}
+
 // Returns the register class to use for sources of VOP3 instructions for the
 // given VT.
 class getVOP3SrcForVT<ValueType VT> {
   bit isFP = !if(!eq(VT.Value, f16.Value), 1,
              !if(!eq(VT.Value, v2f16.Value), 1,
              !if(!eq(VT.Value, f32.Value), 1,
              !if(!eq(VT.Value, f64.Value), 1,
              0))));
class VOPProfile <list<ValueType> _ArgVT> {
   field RegisterClass Src1RC32 = getVregSrcForVT<Src1VT>.ret;
   field RegisterOperand Src0RC64 = getVOP3SrcForVT<Src0VT>.ret;
   field RegisterOperand Src1RC64 = getVOP3SrcForVT<Src1VT>.ret;
   field RegisterOperand Src2RC64 = getVOP3SrcForVT<Src2VT>.ret;
   field RegisterClass Src0DPP = getVregSrcForVT<Src0VT>.ret;
   field RegisterClass Src1DPP = getVregSrcForVT<Src1VT>.ret;
   field RegisterClass Src0SDWA = getVregSrcForVT<Src0VT>.ret;
   field RegisterClass Src1SDWA = getVregSrcForVT<Src1VT>.ret;
-  field RegisterOperand Src0SDWA9 = SDWA9Src;
-  field RegisterOperand Src1SDWA9 = SDWA9Src;
+  field RegisterOperand Src0SDWA9 = getSDWA9SrcForVT<Src0VT>.ret;
+  field RegisterOperand Src1SDWA9 = getSDWA9SrcForVT<Src0VT>.ret;
   field Operand Src0Mod = getSrcMod<Src0VT>.ret;
   field Operand Src1Mod = getSrcMod<Src1VT>.ret;
   field Operand Src2Mod = getSrcMod<Src2VT>.ret;
   field Operand Src0ModDPP = getSrcModExt<Src0VT>.ret;
   field Operand Src1ModDPP = getSrcModExt<Src1VT>.ret;
   field Operand Src0ModSDWA = getSrcModExt<Src0VT>.ret;
   field Operand Src1ModSDWA = getSrcModExt<Src1VT>.ret;
   field Operand Src0ModSDWA9 = getSrcModSDWA9<Src0VT>.ret;

test/MC/Disassembler/AMDGPU/sdwa_gfx9.txt

+# RUN: llvm-mc -arch=amdgcn -mcpu=gfx900 -disassemble -show-encoding < %s | FileCheck %s -check-prefix=GFX9
+
+#-----------------------------------------------------------------------------#
+# Input modifiers
+#-----------------------------------------------------------------------------#
+
+# GFX9: v_fract_f32_sdwa v0, |v0| dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x36,0x00,0x7e,0x00,0x06,0x25,0x06]
+0xf9 0x36 0x00 0x7e 0x00 0x06 0x25 0x06
+
+# GFX9: v_sin_f32_sdwa v0, -|v0| dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x52,0x00,0x7e,0x00,0x06,0x35,0x06]
+0xf9 0x52 0x00 0x7e 0x00 0x06 0x35 0x06
+
+# GFX9: v_add_f32_sdwa v0, -|v0|, -v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x00,0x00,0x02,0x00,0x06,0x35,0x12]
+0xf9 0x00 0x00 0x02 0x00 0x06 0x35 0x12
+
+# GFX9: v_min_f32_sdwa v0, |v0|, -v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x00,0x00,0x14,0x00,0x06,0x25,0x12]
+0xf9 0x00 0x00 0x14 0x00 0x06 0x25 0x12
+
+#-----------------------------------------------------------------------------#
+# VOP1
+#-----------------------------------------------------------------------------#
+
+# GFX9: v_mov_b32_sdwa v1, v2 dst_sel:BYTE_0 dst_unused:UNUSED_PRESERVE src0_sel:DWORD ; encoding: [0xf9,0x02,0x02,0x7e,0x02,0x10,0x06,0x06]
+0xf9 0x02 0x02 0x7e 0x02 0x10 0x06 0x06
+
+# GFX9: v_mov_b32_sdwa v3, v4 dst_sel:BYTE_1 dst_unused:UNUSED_PRESERVE src0_sel:WORD_1 ; encoding: [0xf9,0x02,0x06,0x7e,0x04,0x11,0x05,0x06]
+0xf9 0x02 0x06 0x7e 0x04 0x11 0x05 0x06
+
+# GFX9: v_mov_b32_sdwa v15, v99 dst_sel:BYTE_2 dst_unused:UNUSED_SEXT src0_sel:WORD_0 ; encoding: [0xf9,0x02,0x1e,0x7e,0x63,0x0a,0x04,0x06]
+0xf9 0x02 0x1e 0x7e 0x63 0x0a 0x04 0x06
+
+# GFX9: v_min_u32_sdwa v194, v13, v1 dst_sel:BYTE_3 dst_unused:UNUSED_SEXT src0_sel:BYTE_3 src1_sel:BYTE_2 ; encoding: [0xf9,0x02,0x84,0x1d,0x0d,0x0b,0x03,0x02]
+0xf9 0x02 0x84 0x1d 0x0d 0x0b 0x03 0x02
+
+# GFX9: v_min_u32_sdwa v255, v4, v1 dst_sel:WORD_0 dst_unused:UNUSED_PAD src0_sel:BYTE_2 src1_sel:WORD_1 ; encoding: [0xf9,0x02,0xfe,0x1d,0x04,0x04,0x02,0x05]
+0xf9 0x02 0xfe 0x1d 0x04 0x04 0x02 0x05
+
+# GFX9: v_min_u32_sdwa v200, v200, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD ; encoding: [0xf9,0x02,0x90,0x1d,0xc8,0x05,0x01,0x06]
+0xf9 0x02 0x90 0x1d 0xc8 0x05 0x01 0x06
+
+# GFX9: v_min_u32_sdwa v1, v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD ; encoding: [0xf9,0x02,0x02,0x1c,0x01,0x06,0x00,0x06]
+0xf9 0x02 0x02 0x1c 0x01 0x06 0x00 0x06
+
+# GFX9: v_cvt_u32_f32_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PRESERVE src0_sel:DWORD ; encoding: [0xf9,0x0e,0x00,0x7e,0x00,0x16,0x06,0x06]
+0xf9 0x0e 0x00 0x7e 0x00 0x16 0x06 0x06
+
+# GFX9: v_fract_f32_sdwa v0, v0 clamp dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD ; encoding: [0xf9,0x36,0x00,0x7e,0x00,0x26,0x06,0x06]
+0xf9 0x36 0x00 0x7e 0x00 0x26 0x06 0x06
+
+# GFX9: v_sin_f32_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x52,0x00,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x52 0x00 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_mov_b32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PRESERVE src0_sel:WORD_1 ; encoding: [0xf9,0x02,0x02,0x7e,0x00,0x16,0x05,0x06]
+0xf9 0x02 0x02 0x7e 0x00 0x16 0x05 0x06
+
+# GFX9: v_trunc_f32_sdwa v1, v0 clamp dst_sel:DWORD dst_unused:UNUSED_PRESERVE src0_sel:WORD_1 ; encoding: [0xf9,0x38,0x02,0x7e,0x00,0x36,0x05,0x06]
+0xf9 0x38 0x02 0x7e 0x00 0x36 0x05 0x06
+
+# GFX9: v_mov_b32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PRESERVE src0_sel:DWORD ; encoding: [0xf9,0x02,0x02,0x7e,0x00,0x16,0x06,0x06]
+0xf9 0x02 0x02 0x7e 0x00 0x16 0x06 0x06
+
+# GFX9: v_nop ; encoding: [0xf9,0x00,0x00,0x7e,0x00,0x16,0x06,0x06]
+0xf9 0x00 0x00 0x7e 0x00 0x16 0x06 0x06
+
+# GFX9: v_cvt_u32_f32_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x0e,0x00,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x0e 0x00 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_fract_f32_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x36,0x00,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x36 0x00 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_sin_f32_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x52,0x00,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x52 0x00 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_mov_b32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x02,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x02 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_cvt_f32_i32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x0a,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x0a 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_cvt_f32_u32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x0c,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x0c 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_cvt_i32_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x10,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x10 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_cvt_f16_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x14,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x14 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_cvt_f32_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x16,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x16 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_cvt_rpi_i32_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x18,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x18 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_cvt_flr_i32_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x1a,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x1a 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_cvt_off_f32_i4_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x1c,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x1c 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_cvt_f32_ubyte0_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x22,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x22 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_cvt_f32_ubyte1_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x24,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x24 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_cvt_f32_ubyte2_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x26,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x26 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_cvt_f32_ubyte3_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x28,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x28 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_trunc_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x38,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x38 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_ceil_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x3a,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x3a 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_rndne_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x3c,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x3c 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_floor_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x3e,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x3e 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_exp_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x40,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x40 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_log_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x42,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x42 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_rcp_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x44,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x44 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_rcp_iflag_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x46,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x46 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_rsq_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x48,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x48 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_sqrt_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x4e,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x4e 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_cos_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x54,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x54 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_not_b32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x56,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x56 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_bfrev_b32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x58,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x58 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_ffbh_u32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x5a,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x5a 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_ffbl_b32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x5c,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x5c 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_ffbh_i32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x5e,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x5e 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_frexp_exp_i32_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x66,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x66 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_frexp_mant_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x68,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x68 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_log_legacy_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x98,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x98 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_exp_legacy_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x96,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x96 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_cvt_f16_u16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x72,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x72 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_cvt_f16_i16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x74,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x74 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_cvt_u16_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x76,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x76 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_cvt_i16_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x78,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x78 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_rcp_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x7a,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x7a 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_sqrt_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x7c,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x7c 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_rsq_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x7e,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x7e 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_log_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x80,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x80 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_exp_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x82,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x82 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_frexp_mant_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x84,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x84 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_frexp_exp_i16_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x86,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x86 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_floor_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x88,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x88 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_ceil_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x8a,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x8a 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_trunc_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x8c,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x8c 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_rndne_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x8e,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x8e 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_fract_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x90,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x90 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_sin_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x92,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x92 0x02 0x7e 0x00 0x06 0x05 0x06
+
+# GFX9: v_cos_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x94,0x02,0x7e,0x00,0x06,0x05,0x06]
+0xf9 0x94 0x02 0x7e 0x00 0x06 0x05 0x06
+
+#-----------------------------------------------------------------------------#
+# VOP2
+#-----------------------------------------------------------------------------#
+
+# GFX9: v_add_f32_sdwa v0, v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD ; encoding: [0xf9,0x00,0x00,0x02,0x00,0x06,0x05,0x06]
+0xf9 0x00 0x00 0x02 0x00 0x06 0x05 0x06
+
+# GFX9: v_min_f32_sdwa v0, v0, v0 clamp dst_sel:DWORD dst_unused:UNUSED_PRESERVE src0_sel:DWORD src1_sel:BYTE_2 ; encoding: [0xf9,0x00,0x00,0x14,0x00,0x36,0x06,0x02]
+0xf9 0x00 0x00 0x14 0x00 0x36 0x06 0x02
+
+# GFX9: v_and_b32_sdwa v0, v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_2 ; encoding: [0xf9,0x00,0x00,0x26,0x00,0x06,0x06,0x02]
+0xf9 0x00 0x00 0x26 0x00 0x06 0x06 0x02
+
+# GFX9: v_mul_i32_i24_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PRESERVE src0_sel:DWORD src1_sel:DWORD ; encoding: [0xf9,0x06,0x02,0x0c,0x02,0x16,0x06,0x06]
+0xf9 0x06 0x02 0x0c 0x02 0x16 0x06 0x06
+
+# GFX9: v_add_f32_sdwa v0, v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x00,0x00,0x02,0x00,0x06,0x05,0x02]
+0xf9 0x00 0x00 0x02 0x00 0x06 0x05 0x02
+
+# GFX9: v_min_f32_sdwa v0, v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x00,0x00,0x14,0x00,0x06,0x05,0x02]
+0xf9 0x00 0x00 0x14 0x00 0x06 0x05 0x02
+
+# GFX9: v_and_b32_sdwa v0, v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x00,0x00,0x26,0x00,0x06,0x05,0x02]
+0xf9 0x00 0x00 0x26 0x00 0x06 0x05 0x02
+
+# GFX9: v_mul_i32_i24_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x0c,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x0c 0x02 0x06 0x05 0x02
+
+# GFX9: v_sub_f32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x04,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x04 0x02 0x06 0x05 0x02
+
+# GFX9: v_subrev_f32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x06,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x06 0x02 0x06 0x05 0x02
+
+# GFX9: v_mul_f32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x0a,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x0a 0x02 0x06 0x05 0x02
+
+# GFX9: v_mul_hi_i32_i24_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x0e,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x0e 0x02 0x06 0x05 0x02
+
+# GFX9: v_mul_u32_u24_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x10,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x10 0x02 0x06 0x05 0x02
+
+# GFX9: v_mul_hi_u32_u24_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x12,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x12 0x02 0x06 0x05 0x02
+
+# GFX9: v_max_f32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x16,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x16 0x02 0x06 0x05 0x02
+
+# GFX9: v_min_i32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x18,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x18 0x02 0x06 0x05 0x02
+
+# GFX9: v_max_i32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x1a,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x1a 0x02 0x06 0x05 0x02
+
+# GFX9: v_min_u32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x1c,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x1c 0x02 0x06 0x05 0x02
+
+# GFX9: v_max_u32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x1e,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x1e 0x02 0x06 0x05 0x02
+
+# GFX9: v_lshrrev_b32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x20,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x20 0x02 0x06 0x05 0x02
+
+# GFX9: v_ashrrev_i32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x22,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x22 0x02 0x06 0x05 0x02
+
+# GFX9: v_lshlrev_b32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x24,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x24 0x02 0x06 0x05 0x02
+
+# GFX9: v_or_b32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x28,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x28 0x02 0x06 0x05 0x02
+
+# GFX9: v_xor_b32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x2a,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x2a 0x02 0x06 0x05 0x02
+
+# GFX9: v_add_f16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x3e,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x3e 0x02 0x06 0x05 0x02
+
+# GFX9: v_sub_f16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x40,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x40 0x02 0x06 0x05 0x02
+
+# GFX9: v_subrev_f16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x42,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x42 0x02 0x06 0x05 0x02
+
+# GFX9: v_mul_f16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x44,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x44 0x02 0x06 0x05 0x02
+
+# GFX9: v_add_u16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x4c,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x4c 0x02 0x06 0x05 0x02
+
+# GFX9: v_sub_u16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x4e,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x4e 0x02 0x06 0x05 0x02
+
+# GFX9: v_subrev_u16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x50,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x50 0x02 0x06 0x05 0x02
+
+# GFX9: v_mul_lo_u16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x52,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x52 0x02 0x06 0x05 0x02
+
+# GFX9: v_lshlrev_b16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x54,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x54 0x02 0x06 0x05 0x02
+
+# GFX9: v_lshrrev_b16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x56,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x56 0x02 0x06 0x05 0x02
+
+# GFX9: v_ashrrev_i16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x58,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x58 0x02 0x06 0x05 0x02
+
+# GFX9: v_max_f16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x5a,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x5a 0x02 0x06 0x05 0x02
+
+# GFX9: v_min_f16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x5c,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x5c 0x02 0x06 0x05 0x02
+
+# GFX9: v_max_u16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x5e,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x5e 0x02 0x06 0x05 0x02
+
+# GFX9: v_max_i16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x60,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x60 0x02 0x06 0x05 0x02
+
+# GFX9: v_min_u16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x62,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x62 0x02 0x06 0x05 0x02
+
+# GFX9: v_min_i16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x64,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x64 0x02 0x06 0x05 0x02
+
+# GFX9: v_ldexp_f16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x66,0x02,0x06,0x05,0x02]
+0xf9 0x06 0x02 0x66 0x02 0x06 0x05 0x02
+
+#-----------------------------------------------------------------------------#
+# VOPC
+#-----------------------------------------------------------------------------#
+
+# GFX9: v_cmp_eq_f32_sdwa vcc, v1, v2 src0_sel:BYTE_2 src1_sel:WORD_0 ; encoding: [0xf9,0x04,0x84,0x7c,0x01,0x00,0x02,0x04]
+0xf9 0x04 0x84 0x7c 0x01 0x00 0x02 0x04
+
+# GFX9: v_cmp_nle_f32_sdwa vcc, v1, v2 src0_sel:BYTE_2 src1_sel:WORD_0 ; encoding: [0xf9,0x04,0x98,0x7c,0x01,0x00,0x02,0x04]
+0xf9 0x04 0x98 0x7c 0x01 0x00 0x02 0x04
+
+# GFX9: v_cmpx_gt_f32_sdwa vcc, v1, v2 src0_sel:BYTE_2 src1_sel:WORD_0 ; encoding: [0xf9,0x04,0xa8,0x7c,0x01,0x00,0x02,0x04]
+0xf9 0x04 0xa8 0x7c 0x01 0x00 0x02 0x04
+
+# GFX9: v_cmpx_nlt_f32_sdwa vcc, v1, v2 src0_sel:BYTE_2 src1_sel:WORD_0 ; encoding: [0xf9,0x04,0xbc,0x7c,0x01,0x00,0x02,0x04]
+0xf9 0x04 0xbc 0x7c 0x01 0x00 0x02 0x04
+
+# GFX9: v_cmp_lt_i32_sdwa vcc, v1, v2 src0_sel:BYTE_2 src1_sel:WORD_0 ; encoding: [0xf9,0x04,0x82,0x7d,0x01,0x00,0x02,0x04]
+0xf9 0x04 0x82 0x7d 0x01 0x00 0x02 0x04
+
+# GFX9: v_cmp_t_i32_sdwa vcc, v1, v2 src0_sel:BYTE_2 src1_sel:WORD_0 ; encoding: [0xf9,0x04,0x8e,0x7d,0x01,0x00,0x02,0x04]
+0xf9 0x04 0x8e 0x7d 0x01 0x00 0x02 0x04
+
+# GFX9: v_cmpx_eq_i32_sdwa vcc, v1, v2 src0_sel:BYTE_2 src1_sel:WORD_0 ; encoding: [0xf9,0x04,0xa4,0x7d,0x01,0x00,0x02,0x04]
+0xf9 0x04 0xa4 0x7d 0x01 0x00 0x02 0x04
+
+# GFX9: v_cmpx_ne_i32_sdwa vcc, v1, v2 src0_sel:BYTE_2 src1_sel:WORD_0 ; encoding: [0xf9,0x04,0xaa,0x7d,0x01,0x00,0x02,0x04]
+0xf9 0x04 0xaa 0x7d 0x01 0x00 0x02 0x04
+
+# GFX9: v_cmp_f_u32_sdwa vcc, v1, v2 src0_sel:BYTE_2 src1_sel:WORD_0 ; encoding: [0xf9,0x04,0x90,0x7d,0x01,0x00,0x02,0x04]
+0xf9 0x04 0x90 0x7d 0x01 0x00 0x02 0x04
+
+# GFX9: v_cmp_gt_u32_sdwa vcc, v1, v2 src0_sel:BYTE_2 src1_sel:WORD_0 ; encoding: [0xf9,0x04,0x98,0x7d,0x01,0x00,0x02,0x04]
+0xf9 0x04 0x98 0x7d 0x01 0x00 0x02 0x04
+
+# GFX9: v_cmpx_le_u32_sdwa vcc, v1, v2 src0_sel:BYTE_2 src1_sel:WORD_0 ; encoding: [0xf9,0x04,0xb6,0x7d,0x01,0x00,0x02,0x04]
+0xf9 0x04 0xb6 0x7d 0x01 0x00 0x02 0x04
+
+# GFX9: v_cmpx_ne_u32_sdwa vcc, v1, v2 src0_sel:BYTE_2 src1_sel:WORD_0 ; encoding: [0xf9,0x04,0xba,0x7d,0x01,0x00,0x02,0x04]
+0xf9 0x04 0xba 0x7d 0x01 0x00 0x02 0x04
+
+# GFX9: v_cmp_class_f32_sdwa vcc, v1, v2 src0_sel:BYTE_2 src1_sel:WORD_0 ; encoding: [0xf9,0x04,0x20,0x7c,0x01,0x00,0x02,0x04]
+0xf9 0x04 0x20 0x7c 0x01 0x00 0x02 0x04
+
+# GFX9: v_cmpx_class_f32_sdwa vcc, v1, v2 src0_sel:BYTE_2 src1_sel:WORD_0 ; encoding: [0xf9,0x04,0x22,0x7c,0x01,0x00,0x02,0x04]
+0xf9 0x04 0x22 0x7c 0x01 0x00 0x02 0x04
+
+#-----------------------------------------------------------------------------#
+# Modifiers
+#-----------------------------------------------------------------------------#
+
+# GFX9: v_fract_f32_sdwa v0, |v0| dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x36,0x00,0x7e,0x00,0x06,0x25,0x06]
+0xf9 0x36 0x00 0x7e 0x00 0x06 0x25 0x06
+
+# GFX9: v_sin_f32_sdwa v0, -|v0| dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x52,0x00,0x7e,0x00,0x06,0x35,0x06]
+0xf9 0x52 0x00 0x7e 0x00 0x06 0x35 0x06
+
+# GFX9: v_add_f32_sdwa v0, -|v0|, -v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x00,0x00,0x02,0x00,0x06,0x35,0x12]
+0xf9 0x00 0x00 0x02 0x00 0x06 0x35 0x12
+
+# GFX9: v_min_f32_sdwa v0, |v0|, -v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x00,0x00,0x14,0x00,0x06,0x25,0x12]
+0xf9 0x00 0x00 0x14 0x00 0x06 0x25 0x12
+
+# GFX9: v_mov_b32_sdwa v1, sext(v0) dst_sel:DWORD dst_unused:UNUSED_PRESERVE src0_sel:DWORD ; encoding: [0xf9,0x02,0x02,0x7e,0x00,0x16,0x0e,0x06]
+0xf9 0x02 0x02 0x7e 0x00 0x16 0x0e 0x06
+
+# GFX9: v_and_b32_sdwa v0, sext(v0), sext(v0) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_2 ; encoding: [0xf9,0x00,0x00,0x26,0x00,0x06,0x0e,0x0a]
+0xf9 0x00 0x00 0x26 0x00 0x06 0x0e 0x0a
+
+# GFX9: v_cmp_class_f32_sdwa vcc, -v1, sext(v2) src0_sel:BYTE_2 src1_sel:WORD_0 ; encoding: [0xf9,0x04,0x20,0x7c,0x01,0x00,0x12,0x0c]
+0xf9 0x04 0x20 0x7c 0x01 0x00 0x12 0x0c
+
+#===------------------------------------------------------------------------===#
+# Scalar registers are allowed
+#===------------------------------------------------------------------------===#
+
+# GFX9: v_mov_b32_sdwa v1, s2 dst_sel:BYTE_0 dst_unused:UNUSED_PRESERVE src0_sel:DWORD ; encoding: [0xf9,0x02,0x02,0x7e,0x02,0x10,0x86,0x06]
+0xf9 0x02 0x02 0x7e 0x02 0x10 0x86 0x06
+
+# GFX9: v_mov_b32_sdwa v1, exec_lo dst_sel:BYTE_0 dst_unused:UNUSED_PRESERVE src0_sel:DWORD ; encoding: [0xf9,0x02,0x02,0x7e,0x7e,0x10,0x86,0x06]
+0xf9 0x02 0x02 0x7e 0x7e 0x10 0x86 0x06
+
+# GFX9: v_add_f32_sdwa v0, s0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x00,0x00,0x02,0x00,0x06,0x85,0x02]
+0xf9 0x00 0x00 0x02 0x00 0x06 0x85 0x02
+
+# GFX9: v_add_f32_sdwa v0, v0, s22 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x2c,0x00,0x02,0x00,0x06,0x05,0x82]
+0xf9 0x2c 0x00 0x02 0x00 0x06 0x05 0x82
+
+# GFX9: v_cmp_eq_f32_sdwa vcc, s1, v2 src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x04,0x84,0x7c,0x01,0x00,0x85,0x02]
+0xf9 0x04 0x84 0x7c 0x01 0x00 0x85 0x02
+
+# GFX9: v_cmp_eq_f32_sdwa vcc, v1, s22 src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x2c,0x84,0x7c,0x01,0x00,0x05,0x82]
+0xf9 0x2c 0x84 0x7c 0x01 0x00 0x05 0x82
+
+#===------------------------------------------------------------------------===#
+# VOPC with arbitrary SGPR destination
+#===------------------------------------------------------------------------===#
+
+# GFX9: v_cmp_eq_f32_sdwa s[2:3], v1, v2 src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x04,0x84,0x7c,0x01,0x82,0x05,0x02]
+0xf9 0x04 0x84 0x7c 0x01 0x82 0x05 0x02
+
+# GFX9: v_cmp_eq_f32_sdwa exec, v1, v2 src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x04,0x84,0x7c,0x01,0xfe,0x05,0x02]
+0xf9 0x04 0x84 0x7c 0x01 0xfe 0x05 0x02
+
+# GFX9: v_cmp_eq_f32_sdwa exec, s2, v2 src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x04,0x84,0x7c,0x02,0xfe,0x85,0x02]
+0xf9 0x04 0x84 0x7c 0x02 0xfe 0x85 0x02
+
+#===------------------------------------------------------------------------===#
+# OMod output modifier allowed
+#===------------------------------------------------------------------------===#
+
+# GFX9: v_trunc_f32_sdwa v1, v2 mul:2 dst_sel:BYTE_0 dst_unused:UNUSED_PRESERVE src0_sel:DWORD ; encoding: [0xf9,0x38,0x02,0x7e,0x02,0x50,0x06,0x06]
+0xf9 0x38 0x02 0x7e 0x02 0x50 0x06 0x06
+
+# GFX9: v_trunc_f32_sdwa v1, v2 clamp div:2 dst_sel:BYTE_0 dst_unused:UNUSED_PRESERVE src0_sel:DWORD ; encoding: [0xf9,0x38,0x02,0x7e,0x02,0xf0,0x06,0x06]
+0xf9 0x38 0x02 0x7e 0x02 0xf0 0x06 0x06
+
+# GFX9: v_add_f32_sdwa v0, v0, v0 mul:2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x00,0x00,0x02,0x00,0x46,0x05,0x02]
+0xf9 0x00 0x00 0x02 0x00 0x46 0x05 0x02
+
+# GFX9: v_add_f32_sdwa v0, v0, v0 clamp div:2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x00,0x00,0x02,0x00,0xe6,0x05,0x02]
+0xf9 0x00 0x00 0x02 0x00 0xe6 0x05 0x02