AMDGPU: Add instruction definitions for GPR indexing

[arsenm]

Rename to specify VGPRs, add missing test

[arsenm]

Print mode name instead of number. I'm not sure how this is supposed to be printed, but this is better for tests

[arsenm]

Fix assembler tests

[• tstellarAMD]

LGTM.

[arsenm]

r284027

lib/Target/AMDGPU/AMDGPU.td

 >;
 
 def Feature16BitInsts : SubtargetFeature<"16-bit-insts",
   "Has16BitInsts",
   "true",
   "Has i16/f16 instructions"
 >;
 
+def FeatureGPRIndexMode : SubtargetFeature<"gpr-index-mode",
+  "HasGPRIndexMode",
+  "true",
+  "Has GPR mode register indexing"
+>;
+
 //===------------------------------------------------------------===//
 // Subtarget Features (options and debugging)
 //===------------------------------------------------------------===//
 
 // Some instructions do not support denormals despite this flag. Using
 // fp32 denormals also causes instructions to run at the double
 // precision rate for the device.
 def FeatureFP32Denormals : SubtargetFeature<"fp32-denormals",
def FeatureSeaIslands : SubtargetFeatureGeneration<"SEA_ISLANDS",
   FeatureWavefrontSize64, FeatureGCN, FeatureFlatAddressSpace,
   FeatureGCN1Encoding, FeatureCIInsts]
 >;
 
 def FeatureVolcanicIslands : SubtargetFeatureGeneration<"VOLCANIC_ISLANDS",
   [FeatureFP64, FeatureLocalMemorySize65536,
    FeatureWavefrontSize64, FeatureFlatAddressSpace, FeatureGCN,
    FeatureGCN3Encoding, FeatureCIInsts, Feature16BitInsts,
-   FeatureSMemRealTime
+   FeatureSMemRealTime, FeatureGPRIndexMode
   ]
 >;
 
 //===----------------------------------------------------------------------===//
 // Debugger related subtarget features.
 //===----------------------------------------------------------------------===//
 
 def FeatureDebuggerInsertNops : SubtargetFeature<

lib/Target/AMDGPU/AMDGPUSubtarget.h

protected:
   bool FP64;
   bool IsGCN;
   bool GCN1Encoding;
   bool GCN3Encoding;
   bool CIInsts;
   bool SGPRInitBug;
   bool HasSMemRealTime;
   bool Has16BitInsts;
+  bool HasGPRIndexMode;
   bool FlatAddressSpace;
   bool R600ALUInst;
   bool CaymanISA;
   bool CFALUBug;
   bool HasVertexCache;
   short TexVTXClauseSize;
 
   // Dummy feature to use for assembler in tablegen.
public:
   bool hasSMemRealTime() const {
     return HasSMemRealTime;
   }
 
   bool has16BitInsts() const {
     return Has16BitInsts;
   }
 
+  bool hasGPRIndexMode() const {
+    return HasGPRIndexMode;
+  }
+
   bool hasScalarCompareEq64() const {
     return getGeneration() >= VOLCANIC_ISLANDS;
   }
 
   bool enableSIScheduler() const {
     return EnableSIScheduler;
   }
 

lib/Target/AMDGPU/AMDGPUSubtarget.cpp

: AMDGPUGenSubtargetInfo(TT, GPU, FS),
     FP64(false),
     IsGCN(false),
     GCN1Encoding(false),
     GCN3Encoding(false),
     CIInsts(false),
     SGPRInitBug(false),
     HasSMemRealTime(false),
     Has16BitInsts(false),
+    HasGPRIndexMode(false),
     FlatAddressSpace(false),
 
     R600ALUInst(false),
     CaymanISA(false),
     CFALUBug(false),
     HasVertexCache(false),
     TexVTXClauseSize(0),
 

lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp

public:
   }
 
   bool isSWaitCnt() const;
   bool isHwreg() const;
   bool isSendMsg() const;
   bool isSMRDOffset() const;
   bool isSMRDLiteralOffset() const;
   bool isDPPCtrl() const;
+  bool isGPRIdxMode() const;
 
   StringRef getExpressionAsToken() const {
     assert(isExpr());
     const MCSymbolRefExpr *S = cast<MCSymbolRefExpr>(Expr);
     return S->getSymbol().getName();
   }
 
 
return ((Imm >= 0x000) && (Imm <= 0x0ff)) ||
            (Imm == 0x140) ||
            (Imm == 0x141) ||
            (Imm == 0x142) ||
            (Imm == 0x143);
   }
   return false;
 }
 
+bool AMDGPUOperand::isGPRIdxMode() const {
+  return isImm() && isUInt<4>(getImm());
+}
+
 AMDGPUAsmParser::OperandMatchResultTy
 AMDGPUAsmParser::parseDPPCtrl(OperandVector &Operands) {
   SMLoc S = Parser.getTok().getLoc();
   StringRef Prefix;
   int64_t Int;
 
   if (getLexer().getKind() == AsmToken::Identifier) {
     Prefix = Parser.getTok().getString();

lib/Target/AMDGPU/SIInstructions.td

def isGCN : Predicate<"Subtarget->getGeneration() "
                       ">= SISubtarget::SOUTHERN_ISLANDS">,
             AssemblerPredicate<"FeatureGCN">;
 def isSI : Predicate<"Subtarget->getGeneration() "
                       "== SISubtarget::SOUTHERN_ISLANDS">,
            AssemblerPredicate<"FeatureSouthernIslands">;
 
 def has16BankLDS : Predicate<"Subtarget->getLDSBankCount() == 16">;
 def has32BankLDS : Predicate<"Subtarget->getLDSBankCount() == 32">;
+def hasGPRIndexMode : Predicate<"Subtarget->hasGPRIndexMode()">,
+                      AssemblerPredicate<"FeatureGPRIndexMode">;
 
 include "VOPInstructions.td"
 include "SOPInstructions.td"
 include "SMInstructions.td"
 include "FLATInstructions.td"
 include "BUFInstructions.td"
 
 let SubtargetPredicate = isGCN in {

lib/Target/AMDGPU/SOPInstructions.td

 //===-- SOPInstructions.td - SOP Instruction Defintions -------------------===//
 //
 //                     The LLVM Compiler Infrastructure
 //
 // This file is distributed under the University of Illinois Open Source
 // License. See LICENSE.TXT for details.
 //
 //===----------------------------------------------------------------------===//
 
+def GPRIdxModeMatchClass : AsmOperandClass {
+  let Name = "GPRIdxMode";
+  let PredicateMethod = "isGPRIdxMode";
+  let RenderMethod = "addImmOperands";
+}
+
+def GPRIdxMode : Operand<i32> {
+  let PrintMethod = "printU4ImmOperand";
+  let ParserMatchClass = GPRIdxModeMatchClass;
+  let OperandType = "OPERAND_IMMEDIATE";
+}
+
 //===----------------------------------------------------------------------===//
 // SOP1 Instructions
 //===----------------------------------------------------------------------===//
 
 class SOP1_Pseudo <string opName, dag outs, dag ins,
                    string asmOps, list<dag> pattern=[]> :
   InstSI <outs, ins, "", pattern>,
   SIMCInstr<opName, SIEncodingFamily.NONE> {
class SOP1_Real<bits<8> op, SOP1_Pseudo ps> :
   let Inst{31-23} = 0x17d; //encoding;
 }
 
 class SOP1_32 <string opName, list<dag> pattern=[]> : SOP1_Pseudo <
   opName, (outs SReg_32:$sdst), (ins SSrc_b32:$src0),
   "$sdst, $src0", pattern
 >;
 
+// 32-bit input, no output.
+class SOP1_0_32 <string opName, list<dag> pattern = []> : SOP1_Pseudo <
+  opName, (outs), (ins SSrc_b32:$src0),
+  "$src0", pattern> {
+  let has_sdst = 0;
+}
+
 class SOP1_64 <string opName, list<dag> pattern=[]> : SOP1_Pseudo <
   opName, (outs SReg_64:$sdst), (ins SSrc_b64:$src0),
   "$sdst, $src0", pattern
 >;
 
 // 64-bit input, 32-bit output.
 class SOP1_32_64 <string opName, list<dag> pattern=[]> : SOP1_Pseudo <
   opName, (outs SReg_32:$sdst), (ins SSrc_b64:$src0),

 
 def S_CBRANCH_JOIN : SOP1_1  <"s_cbranch_join">;
 def S_MOV_REGRD_B32 : SOP1_32 <"s_mov_regrd_b32">;
 let Defs = [SCC] in {
 def S_ABS_I32 : SOP1_32 <"s_abs_i32">;
 } // End Defs = [SCC]
 def S_MOV_FED_B32 : SOP1_32 <"s_mov_fed_b32">;
 
+let SubtargetPredicate = hasGPRIndexMode in {
+def S_SET_GPR_IDX_IDX : SOP1_0_32<"s_set_gpr_idx_idx">;
+}
 
 //===----------------------------------------------------------------------===//
 // SOP2 Instructions
 //===----------------------------------------------------------------------===//
 
 class SOP2_Pseudo<string opName, dag outs, dag ins,
                   string asmOps, list<dag> pattern=[]> :
   InstSI<outs, ins, "", pattern>,
class SOPCe <bits<7> op> : Enc32 {
   bits<8> src1;
 
   let Inst{7-0} = src0;
   let Inst{15-8} = src1;
   let Inst{22-16} = op;
   let Inst{31-23} = 0x17e;
 }
 
-class SOPC <bits<7> op, dag outs, dag ins, string asm, list<dag> pattern> :
+class SOPC <bits<7> op, dag outs, dag ins, string asm,
+            list<dag> pattern = []> :
   InstSI<outs, ins, asm, pattern>, SOPCe <op> {
   let mayLoad = 0;
   let mayStore = 0;
   let hasSideEffects = 0;
   let SALU = 1;
   let SOPC = 1;
   let isCodeGenOnly = 0;
   let Defs = [SCC];

 
 def S_BITCMP0_B32 : SOPC_32 <0x0c, "s_bitcmp0_b32">;
 def S_BITCMP1_B32 : SOPC_32 <0x0d, "s_bitcmp1_b32">;
 def S_BITCMP0_B64 : SOPC_64_32 <0x0e, "s_bitcmp0_b64">;
 def S_BITCMP1_B64 : SOPC_64_32 <0x0f, "s_bitcmp1_b64">;
 def S_SETVSKIP : SOPC_32 <0x10, "s_setvskip">;
 
 let SubtargetPredicate = isVI in {
+def S_SET_GPR_IDX_ON : SOPC <0x11,
+  (outs),
+  (ins SSrc_b32:$src0, GPRIdxMode:$src1),
+  "s_set_gpr_idx_on $src0, $src1"> {
+  let Defs = [M0]; // No scc def
+  let Uses = [M0]; // Other bits of m0 unmodified.
+  let hasSideEffects = 1; // Sets mode.gpr_idx_en
+}
+
 def S_CMP_EQ_U64 : SOPC_CMP_64 <0x12, "s_cmp_eq_u64", COND_EQ>;
 def S_CMP_LG_U64 : SOPC_CMP_64 <0x13, "s_cmp_lg_u64", COND_NE>;
 }
 
 //===----------------------------------------------------------------------===//
 // SOPP Instructions
 //===----------------------------------------------------------------------===//
 
def S_DECPERFLEVEL : SOPP <0x00000015, (ins i32imm:$simm16), "s_decperflevel $simm16",
   [(int_amdgcn_s_decperflevel SIMM16bit:$simm16)]> {
   let hasSideEffects = 1;
   let mayLoad = 1;
   let mayStore = 1;
 }
 def S_TTRACEDATA : SOPP <0x00000016, (ins), "s_ttracedata"> {
   let simm16 = 0;
 }
+
+let SubtargetPredicate = hasGPRIndexMode in {
+def S_SET_GPR_IDX_OFF : SOPP<0x1c, (ins), "s_set_gpr_idx_off"> {
+  let simm16 = 0;
+}
+}
 } // End hasSideEffects
 
+let SubtargetPredicate = hasGPRIndexMode in {
+def S_SET_GPR_IDX_MODE : SOPP<0x1d, (ins GPRIdxMode:$simm16),
+  "s_set_gpr_idx_mode $simm16"> {
+  let Defs = [M0];
+}
+}
 
 let Predicates = [isGCN] in {
 
 //===----------------------------------------------------------------------===//
 // S_GETREG_B32 Intrinsic Pattern.
 //===----------------------------------------------------------------------===//
 def : Pat <
   (int_amdgcn_s_getreg imm:$simm16),

 def S_MOVRELS_B32_vi       : SOP1_Real_vi <0x2a, S_MOVRELS_B32>;
 def S_MOVRELS_B64_vi       : SOP1_Real_vi <0x2b, S_MOVRELS_B64>;
 def S_MOVRELD_B32_vi       : SOP1_Real_vi <0x2c, S_MOVRELD_B32>;
 def S_MOVRELD_B64_vi       : SOP1_Real_vi <0x2d, S_MOVRELD_B64>;
 def S_CBRANCH_JOIN_vi      : SOP1_Real_vi <0x2e, S_CBRANCH_JOIN>;
 def S_MOV_REGRD_B32_vi     : SOP1_Real_vi <0x2f, S_MOV_REGRD_B32>;
 def S_ABS_I32_vi           : SOP1_Real_vi <0x30, S_ABS_I32>;
 def S_MOV_FED_B32_vi       : SOP1_Real_vi <0x31, S_MOV_FED_B32>;
+def S_SET_GPR_IDX_IDX_vi   : SOP1_Real_vi <0x32, S_SET_GPR_IDX_IDX>;
 
 def S_ADD_U32_vi           : SOP2_Real_vi <0x00, S_ADD_U32>;
 def S_ADD_I32_vi           : SOP2_Real_vi <0x02, S_ADD_I32>;
 def S_SUB_U32_vi           : SOP2_Real_vi <0x01, S_SUB_U32>;
 def S_SUB_I32_vi           : SOP2_Real_vi <0x03, S_SUB_I32>;
 def S_ADDC_U32_vi          : SOP2_Real_vi <0x04, S_ADDC_U32>;
 def S_SUBB_U32_vi          : SOP2_Real_vi <0x05, S_SUBB_U32>;
 def S_MIN_I32_vi           : SOP2_Real_vi <0x06, S_MIN_I32>;

test/MC/AMDGPU/sop1.s

-// RUN: llvm-mc -arch=amdgcn -show-encoding %s | FileCheck --check-prefix=GCN --check-prefix=SICI %s
-// RUN: llvm-mc -arch=amdgcn -mcpu=SI -show-encoding %s | FileCheck --check-prefix=GCN --check-prefix=SICI %s
+// RUN: not llvm-mc -arch=amdgcn -show-encoding %s | FileCheck --check-prefix=GCN --check-prefix=SICI %s
+// RUN: not llvm-mc -arch=amdgcn -show-encoding %s 2>&1 | FileCheck --check-prefix=NOSICI %s
 // RUN: not llvm-mc -arch=amdgcn -mcpu=fiji -show-encoding %s 2>&1 | FileCheck --check-prefix=GCN --check-prefix=VI %s
 // RUN: not llvm-mc -arch=amdgcn -mcpu=fiji -show-encoding %s 2>&1 | FileCheck --check-prefix=NOVI %s
 
 s_mov_b32 s1, s2
 // SICI: s_mov_b32 s1, s2 ; encoding: [0x02,0x03,0x81,0xbe]
 // VI:   s_mov_b32 s1, s2 ; encoding: [0x02,0x00,0x81,0xbe]
 
 s_mov_b32 s1, 1

 // VI:   s_cbranch_join s[4:5] ; encoding: [0x04,0x2e,0x80,0xbe]
 
 s_abs_i32 s1, s2
 // SICI: s_abs_i32 s1, s2 ; encoding: [0x02,0x34,0x81,0xbe]
 // VI:   s_abs_i32 s1, s2 ; encoding: [0x02,0x30,0x81,0xbe]
 
 s_mov_fed_b32 s1, s2
 // SICI: s_mov_fed_b32 s1, s2 ; encoding: [0x02,0x35,0x81,0xbe]
+
+s_set_gpr_idx_idx s0
+// VI: s_set_gpr_idx_idx s0 ; encoding: [0x00,0x32,0x80,0xbe]
+// NOSICI: error: instruction not supported on this GPU

test/MC/AMDGPU/sopc.s

 
 s_cmp_eq_u64 s[0:1], s[2:3]
 // VI: s_cmp_eq_u64 s[0:1], s[2:3] ; encoding: [0x00,0x02,0x12,0xbf]
 // NOSICI: error: instruction not supported on this GPU
 
 s_cmp_lg_u64 s[0:1], s[2:3]
 // VI: s_cmp_lg_u64 s[0:1], s[2:3] ; encoding: [0x00,0x02,0x13,0xbf]
 // NOSICI: error: instruction not supported on this GPU
+
+s_set_gpr_idx_on s0, 0
+// VI: s_set_gpr_idx_on s0, 0x0 ; encoding: [0x00,0x00,0x11,0xbf]
+// NOSICI: error: instruction not supported on this GPU
+
+s_set_gpr_idx_on s0, 1
+// VI: s_set_gpr_idx_on s0, 0x1 ; encoding: [0x00,0x01,0x11,0xbf]
+// NOSICI: error: instruction not supported on this GPU
+
+s_set_gpr_idx_on s0, 3
+// VI: s_set_gpr_idx_on s0, 0x3 ; encoding: [0x00,0x03,0x11,0xbf]
+// NOSICI: error: instruction not supported on this GPU
+
+s_set_gpr_idx_on s0, 15
+// VI: s_set_gpr_idx_on s0, 0xf ; encoding: [0x00,0x0f,0x11,0xbf]
+// NOSICI: error: instruction not supported on this GPU

test/MC/AMDGPU/sopp.s

-// RUN: llvm-mc -arch=amdgcn -show-encoding %s | FileCheck --check-prefix=GCN --check-prefix=SICI %s
-// RUN: llvm-mc -arch=amdgcn -mcpu=SI -show-encoding %s | FileCheck --check-prefix=GCN --check-prefix=SICI %s
+// RUN: not llvm-mc -arch=amdgcn -show-encoding %s | FileCheck --check-prefix=GCN --check-prefix=SICI %s
+// RUN: not llvm-mc -arch=amdgcn -show-encoding %s 2>&1 | FileCheck %s --check-prefix=NOSICI
 // RUN: llvm-mc -arch=amdgcn -mcpu=fiji -show-encoding %s | FileCheck --check-prefix=GCN --check-prefix=VI %s
 
 //===----------------------------------------------------------------------===//
 // Edge Cases
 //===----------------------------------------------------------------------===//
 
 s_nop 0
 // GCN: s_nop 0 ; encoding: [0x00,0x00,0x80,0xbf]

 s_incperflevel 5
 // GCN: s_incperflevel 5 ; encoding: [0x05,0x00,0x94,0xbf]
 
 s_decperflevel 6
 // GCN: s_decperflevel 6 ; encoding: [0x06,0x00,0x95,0xbf]
 
 s_ttracedata
 // GCN: s_ttracedata ; encoding: [0x00,0x00,0x96,0xbf]
+
+s_set_gpr_idx_off
+// VI:  s_set_gpr_idx_off ; encoding: [0x00,0x00,0x9c,0xbf]
+// NOSICI: error: instruction not supported on this GPU
+
+s_set_gpr_idx_mode 0
+// VI: s_set_gpr_idx_mode 0x0 ; encoding: [0x00,0x00,0x9d,0xbf]
+// NOSICI: error: instruction not supported on this GPU
+
+s_set_gpr_idx_mode 15
+// VI: s_set_gpr_idx_mode 0xf ; encoding: [0x0f,0x00,0x9d,0xbf]
+// NOSICI: error: instruction not supported on this GPU