Handle more than 64 registers - Part 2 (#102297)

* Make regMaskTP struct for non-arm64 platforms

* some refactoring

* jit format

* fix missing paranethesis in arm

* fix riscv64 and loongarch build

* minor change

* review feedback

src/coreclr/jit/codegenarm.cpp

@@ -1927,14 +1927,15 @@ void CodeGen::genAllocLclFrame(unsigned frameSize, regNumber initReg, bool* pIni
 
 void CodeGen::genPushFltRegs(regMaskTP regMask)
 {
-    assert(regMask != 0);                        // Don't call uness we have some registers to push
-    assert((regMask & RBM_ALLFLOAT) == regMask); // Only floasting point registers should be in regMask
+    assert(regMask != 0);                        // Don't call unless we have some registers to push
+    assert((regMask & RBM_ALLFLOAT) == regMask); // Only floating point registers should be in regMask
 
     regNumber lowReg = genRegNumFromMask(genFindLowestBit(regMask));
     int       slots  = genCountBits(regMask);
+
     // regMask should be contiguously set
-    regMaskTP tmpMask = ((regMask >> lowReg) + 1); // tmpMask should have a single bit set
-    assert((tmpMask & (tmpMask - 1)) == 0);
+    regMaskSmall tmpMask = ((regMask.getLow() >> lowReg) + 1); // tmpMask should have a single bit set
+    assert(genMaxOneBit(tmpMask));
     assert(lowReg == REG_F16); // Currently we expect to start at F16 in the unwind codes
 
     // Our calling convention requires that we only use vpush for TYP_DOUBLE registers
@@ -1952,8 +1953,8 @@ void CodeGen::genPopFltRegs(regMaskTP regMask)
     regNumber lowReg = genRegNumFromMask(genFindLowestBit(regMask));
     int       slots  = genCountBits(regMask);
     // regMask should be contiguously set
-    regMaskTP tmpMask = ((regMask >> lowReg) + 1); // tmpMask should have a single bit set
-    assert((tmpMask & (tmpMask - 1)) == 0);
+    regMaskSmall tmpMask = ((regMask.getLow() >> lowReg) + 1); // tmpMask should have a single bit set
+    assert(genMaxOneBit(tmpMask));
 
     // Our calling convention requires that we only use vpop for TYP_DOUBLE registers
     noway_assert(floatRegCanHoldType(lowReg, TYP_DOUBLE));
@@ -2192,7 +2193,7 @@ void CodeGen::genPopCalleeSavedRegisters(bool jmpEpilog)
         genUsedPopToReturn = false;
     }
 
-    assert(FitsIn<int>(maskPopRegsInt));
+    assert(FitsIn<int>(maskPopRegsInt.getLow()));
     inst_IV(INS_pop, (int)maskPopRegsInt);
     compiler->unwindPopMaskInt(maskPopRegsInt);
 }
@@ -2320,7 +2321,7 @@ void CodeGen::genFuncletProlog(BasicBlock* block)
     regMaskTP maskStackAlloc = genStackAllocRegisterMask(genFuncletInfo.fiSpDelta, maskPushRegsFloat);
     maskPushRegsInt |= maskStackAlloc;
 
-    assert(FitsIn<int>(maskPushRegsInt));
+    assert(FitsIn<int>(maskPushRegsInt.getLow()));
     inst_IV(INS_push, (int)maskPushRegsInt);
     compiler->unwindPushMaskInt(maskPushRegsInt);
 
@@ -2437,7 +2438,7 @@ void CodeGen::genFuncletEpilog()
         compiler->unwindPopMaskFloat(maskPopRegsFloat);
     }
 
-    assert(FitsIn<int>(maskPopRegsInt));
+    assert(FitsIn<int>(maskPopRegsInt.getLow()));
     inst_IV(INS_pop, (int)maskPopRegsInt);
     compiler->unwindPopMaskInt(maskPopRegsInt);
 

src/coreclr/jit/codegenarm64.cpp

@@ -922,7 +922,7 @@ void CodeGen::genSaveCalleeSavedRegistersHelp(regMaskTP regsToSaveMask, int lowe
     assert((spDelta % 16) == 0);
 
     // We also can save FP and LR, even though they are not in RBM_CALLEE_SAVED.
-    assert(regsToSaveCount <= genCountBits(RBM_CALLEE_SAVED | RBM_FP | RBM_LR));
+    assert(regsToSaveCount <= genCountBits(regMaskTP(RBM_CALLEE_SAVED | RBM_FP | RBM_LR)));
 
     // Save integer registers at higher addresses than floating-point registers.
 
@@ -1035,7 +1035,7 @@ void CodeGen::genRestoreCalleeSavedRegistersHelp(regMaskTP regsToRestoreMask, in
     assert((spDelta % 16) == 0);
 
     // We also can restore FP and LR, even though they are not in RBM_CALLEE_SAVED.
-    assert(regsToRestoreCount <= genCountBits(RBM_CALLEE_SAVED | RBM_FP | RBM_LR));
+    assert(regsToRestoreCount <= genCountBits(regMaskTP(RBM_CALLEE_SAVED | RBM_FP | RBM_LR)));
 
     // Point past the end, to start. We predecrement to find the offset to load from.
     static_assert_no_msg(REGSIZE_BYTES == FPSAVE_REGSIZE_BYTES);

src/coreclr/jit/codegenarmarch.cpp

@@ -4687,7 +4687,7 @@ void CodeGen::genPushCalleeSavedRegisters()
 
     maskPushRegsInt |= genStackAllocRegisterMask(compiler->compLclFrameSize, maskPushRegsFloat);
 
-    assert(FitsIn<int>(maskPushRegsInt));
+    assert(FitsIn<int>(maskPushRegsInt.getLow()));
     inst_IV(INS_push, (int)maskPushRegsInt);
     compiler->unwindPushMaskInt(maskPushRegsInt);
 

src/coreclr/jit/codegenloongarch64.cpp

@@ -535,7 +535,7 @@ void CodeGen::genSaveCalleeSavedRegistersHelp(regMaskTP regsToSaveMask, int lowe
         return;
     }
 
-    assert(genCountBits(regsToSaveMask) <= genCountBits(RBM_CALLEE_SAVED));
+    assert(genCountBits(regsToSaveMask) <= genCountBits(regMaskTP(RBM_CALLEE_SAVED)));
 
     // Save integer registers at higher addresses than floating-point registers.
 
@@ -626,7 +626,7 @@ void CodeGen::genRestoreCalleeSavedRegistersHelp(regMaskTP regsToRestoreMask, in
 
     unsigned regsToRestoreCount = genCountBits(regsToRestoreMask);
     // The FP and RA are not in RBM_CALLEE_SAVED.
-    assert(regsToRestoreCount <= genCountBits(RBM_CALLEE_SAVED));
+    assert(regsToRestoreCount <= genCountBits(regMaskTP(RBM_CALLEE_SAVED)));
 
     // Point past the end, to start. We predecrement to find the offset to load from.
     static_assert_no_msg(REGSIZE_BYTES == FPSAVE_REGSIZE_BYTES);

src/coreclr/jit/codegenriscv64.cpp

@@ -607,7 +607,7 @@ void CodeGen::genSaveCalleeSavedRegistersHelp(regMaskTP regsToSaveMask, int lowe
 
     assert((spDelta % STACK_ALIGN) == 0);
 
-    assert(regsToSaveCount <= genCountBits(RBM_CALLEE_SAVED));
+    assert(regsToSaveCount <= genCountBits(regMaskTP(RBM_CALLEE_SAVED)));
 
     // Save integer registers at higher addresses than floating-point registers.
     regMaskTP maskSaveRegsFloat = regsToSaveMask & RBM_ALLFLOAT;
@@ -718,7 +718,7 @@ void CodeGen::genRestoreCalleeSavedRegistersHelp(regMaskTP regsToRestoreMask, in
     assert((spDelta % STACK_ALIGN) == 0);
 
     // We also can restore FP and RA, even though they are not in RBM_CALLEE_SAVED.
-    assert(regsToRestoreCount <= genCountBits(RBM_CALLEE_SAVED | RBM_FP | RBM_RA));
+    assert(regsToRestoreCount <= genCountBits(regMaskTP(RBM_CALLEE_SAVED | RBM_FP | RBM_RA)));
 
     // Point past the end, to start. We predecrement to find the offset to load from.
     static_assert_no_msg(REGSIZE_BYTES == FPSAVE_REGSIZE_BYTES);

src/coreclr/jit/compiler.hpp

@@ -99,7 +99,6 @@ inline bool genExactlyOneBit(T value)
     return ((value != 0) && genMaxOneBit(value));
 }
 
-#ifdef TARGET_ARM64
 inline regMaskTP genFindLowestBit(regMaskTP value)
 {
     return regMaskTP(genFindLowestBit(value.getLow()));
@@ -124,7 +123,6 @@ inline bool genExactlyOneBit(regMaskTP value)
 {
     return genExactlyOneBit(value.getLow());
 }
-#endif
 
 /*****************************************************************************
  *
@@ -169,17 +167,10 @@ inline unsigned uhi32(uint64_t value)
  *  A rather simple routine that counts the number of bits in a given number.
  */
 
-inline unsigned genCountBits(uint64_t bits)
-{
-    return BitOperations::PopCount(bits);
-}
-
-#ifdef TARGET_ARM64
 inline unsigned genCountBits(regMaskTP mask)
 {
     return BitOperations::PopCount(mask.getLow());
 }
-#endif
 
 /*****************************************************************************
  *
@@ -948,19 +939,11 @@ inline regNumber genRegNumFromMask(regMaskTP mask)
 
     /* Convert the mask to a register number */
 
-#ifdef TARGET_ARM64
     regNumber regNum = (regNumber)genLog2(mask.getLow());
 
     /* Make sure we got it right */
     assert(genRegMask(regNum) == mask.getLow());
 
-#else
-    regNumber regNum = (regNumber)genLog2(mask);
-
-    /* Make sure we got it right */
-    assert(genRegMask(regNum) == mask);
-#endif
-
     return regNum;
 }
 
@@ -4505,46 +4488,30 @@ inline void* operator new[](size_t sz, Compiler* compiler, CompMemKind cmk)
 
 inline void printRegMask(regMaskTP mask)
 {
-#ifdef TARGET_ARM64
     printf(REG_MASK_ALL_FMT, mask.getLow());
-#else
-    printf(REG_MASK_ALL_FMT, mask);
-#endif
 }
 
 inline char* regMaskToString(regMaskTP mask, Compiler* context)
 {
     const size_t cchRegMask = 24;
     char*        regmask    = new (context, CMK_Unknown) char[cchRegMask];
 
-#ifdef TARGET_ARM64
     sprintf_s(regmask, cchRegMask, REG_MASK_ALL_FMT, mask.getLow());
-#else
-    sprintf_s(regmask, cchRegMask, REG_MASK_ALL_FMT, mask);
-#endif
 
     return regmask;
 }
 
 inline void printRegMaskInt(regMaskTP mask)
 {
-#ifdef TARGET_ARM64
     printf(REG_MASK_INT_FMT, (mask & RBM_ALLINT).getLow());
-#else
-    printf(REG_MASK_INT_FMT, (mask & RBM_ALLINT));
-#endif
 }
 
 inline char* regMaskIntToString(regMaskTP mask, Compiler* context)
 {
     const size_t cchRegMask = 24;
     char*        regmask    = new (context, CMK_Unknown) char[cchRegMask];
 
-#ifdef TARGET_ARM64
     sprintf_s(regmask, cchRegMask, REG_MASK_INT_FMT, (mask & RBM_ALLINT).getLow());
-#else
-    sprintf_s(regmask, cchRegMask, REG_MASK_INT_FMT, (mask & RBM_ALLINT));
-#endif
 
     return regmask;
 }

src/coreclr/jit/emit.cpp

@@ -8787,7 +8787,7 @@ void emitter::emitRecordGCcall(BYTE* codePos, unsigned char callInstrSize)
     callDsc* call;
 
 #ifdef JIT32_GCENCODER
-    unsigned regs = (emitThisGCrefRegs | emitThisByrefRegs) & ~RBM_INTRET;
+    unsigned regs = (unsigned)(emitThisGCrefRegs | emitThisByrefRegs) & ~RBM_INTRET;
 
     // The JIT32 GCInfo encoder allows us to (as the comment previously here said):
     // "Bail if this is a totally boring call", but the GCInfoEncoder/Decoder interface

src/coreclr/jit/emitarm.cpp

@@ -5771,15 +5771,15 @@ size_t emitter::emitOutputInstr(insGroup* ig, instrDesc* id, BYTE** dp)
     assert(REG_NA == (int)REG_NA);
 
     VARSET_TP GCvars(VarSetOps::UninitVal());
+    regMaskTP gcrefRegs;
+    regMaskTP byrefRegs;
 
     /* What instruction format have we got? */
 
     switch (fmt)
     {
-        int       imm;
-        BYTE*     addr;
-        regMaskTP gcrefRegs;
-        regMaskTP byrefRegs;
+        int   imm;
+        BYTE* addr;
 
         case IF_T1_A: // T1_A    ................
             sz   = SMALL_IDSC_SIZE;

src/coreclr/jit/emitxarch.cpp

@@ -15795,6 +15795,7 @@ BYTE* emitter::emitOutputRI(BYTE* dst, instrDesc* id)
                 break;
 
             case IF_RRW_CNS:
+            {
                 assert(id->idGCref() == GCT_BYREF);
 
 #ifdef DEBUG
@@ -15816,7 +15817,7 @@ BYTE* emitter::emitOutputRI(BYTE* dst, instrDesc* id)
                 // Mark it as holding a GCT_BYREF
                 emitGCregLiveUpd(GCT_BYREF, id->idReg1(), dst);
                 break;
-
+            }
             default:
 #ifdef DEBUG
                 emitDispIns(id, false, false, false);
@@ -16609,6 +16610,8 @@ size_t emitter::emitOutputInstr(insGroup* ig, instrDesc* id, BYTE** dp)
     assert(instrIs3opImul(id->idIns()) == 0 || size >= EA_4BYTE); // Has no 'w' bit
 
     VARSET_TP GCvars(VarSetOps::UninitVal());
+    regMaskTP gcrefRegs;
+    regMaskTP byrefRegs;
 
     // What instruction format have we got?
     switch (insFmt)
@@ -16621,9 +16624,6 @@ size_t emitter::emitOutputInstr(insGroup* ig, instrDesc* id, BYTE** dp)
         BYTE* addr;
         bool  recCall;
 
-        regMaskTP gcrefRegs;
-        regMaskTP byrefRegs;
-
         /********************************************************************/
         /*                        No operands                               */
         /********************************************************************/

src/coreclr/jit/gcencode.cpp

@@ -2874,7 +2874,7 @@ size_t GCInfo::gcMakeRegPtrTable(BYTE* dest, int mask, const InfoHdr& header, un
 
         if (compiler->lvaKeepAliveAndReportThis() && compiler->lvaTable[compiler->info.compThisArg].lvRegister)
         {
-            unsigned thisRegMask   = genRegMask(compiler->lvaTable[compiler->info.compThisArg].GetRegNum());
+            unsigned thisRegMask   = (unsigned)genRegMask(compiler->lvaTable[compiler->info.compThisArg].GetRegNum());
             unsigned thisPtrRegEnc = gceEncodeCalleeSavedRegs(thisRegMask) << 4;
 
             if (thisPtrRegEnc)

src/coreclr/jit/lsraxarch.cpp

@@ -442,7 +442,7 @@ int LinearScan::BuildNode(GenTree* tree)
             // Comparand is preferenced to RAX.
             // The remaining two operands can be in any reg other than RAX.
 
-            const unsigned nonRaxCandidates = availableIntRegs & ~RBM_RAX;
+            const regMaskTP nonRaxCandidates = availableIntRegs & ~RBM_RAX;
             BuildUse(addr, nonRaxCandidates);
             BuildUse(data, varTypeIsByte(tree) ? (nonRaxCandidates & RBM_BYTE_REGS) : nonRaxCandidates);
             BuildUse(comparand, RBM_RAX);