nv50_ir_target_nv50.cpp

/*
 * Copyright 2011 Christoph Bumiller
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in
 * all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
 * OTHER DEALINGS IN THE SOFTWARE.
 */

#include "codegen/nv50_ir_target_nv50.h"

namespace nv50_ir {

Target *getTargetNV50(unsigned int chipset)
{
   return new TargetNV50(chipset);
}

TargetNV50::TargetNV50(unsigned int card) : Target(true, true, false)
{
   chipset = card;

   wposMask = 0;
   for (unsigned int i = 0; i <= SV_LAST; ++i)
      sysvalLocation[i] = ~0;

   initOpInfo();
}

#if 0
// BULTINS / LIBRARY FUNCTIONS:

// TODO
static const uint32_t nvc0_builtin_code[] =
{
};

static const uint16_t nvc0_builtin_offsets[NV50_BUILTIN_COUNT] =
{
};
#endif

void
TargetNV50::getBuiltinCode(const uint32_t **code, uint32_t *size) const
{
   *code = NULL;
   *size = 0;
}

uint32_t
TargetNV50::getBuiltinOffset(int builtin) const
{
   return 0;
}

struct nv50_opProperties
{
   operation op;
   unsigned int mNeg    : 4;
   unsigned int mAbs    : 4;
   unsigned int mNot    : 4;
   unsigned int mSat    : 4;
   unsigned int fConst  : 3;
   unsigned int fShared : 3;
   unsigned int fAttrib : 3;
   unsigned int fImm    : 3;
};

static const struct nv50_opProperties _initProps[] =
{
   //           neg  abs  not  sat  c[]  s[], a[], imm
   { OP_ADD,    0x3, 0x0, 0x0, 0x8, 0x2, 0x1, 0x1, 0x2 },
   { OP_SUB,    0x3, 0x0, 0x0, 0x8, 0x2, 0x1, 0x1, 0x2 },
   { OP_MUL,    0x3, 0x0, 0x0, 0x0, 0x2, 0x1, 0x1, 0x2 },
   { OP_MAX,    0x3, 0x3, 0x0, 0x0, 0x2, 0x1, 0x1, 0x0 },
   { OP_MIN,    0x3, 0x3, 0x0, 0x0, 0x2, 0x1, 0x1, 0x0 },
   { OP_MAD,    0x7, 0x0, 0x0, 0x8, 0x6, 0x1, 0x1, 0x0 }, // special constraint
   { OP_ABS,    0x0, 0x0, 0x0, 0x0, 0x0, 0x1, 0x1, 0x0 },
   { OP_NEG,    0x0, 0x1, 0x0, 0x0, 0x0, 0x1, 0x1, 0x0 },
   { OP_CVT,    0x1, 0x1, 0x0, 0x8, 0x0, 0x1, 0x1, 0x0 },
   { OP_AND,    0x0, 0x0, 0x3, 0x0, 0x0, 0x0, 0x0, 0x2 },
   { OP_OR,     0x0, 0x0, 0x3, 0x0, 0x0, 0x0, 0x0, 0x2 },
   { OP_XOR,    0x0, 0x0, 0x3, 0x0, 0x0, 0x0, 0x0, 0x2 },
   { OP_SHL,    0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x2 },
   { OP_SHR,    0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x2 },
   { OP_SET,    0x3, 0x3, 0x0, 0x0, 0x2, 0x1, 0x1, 0x0 },
   { OP_PREEX2, 0x1, 0x1, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 },
   { OP_PRESIN, 0x1, 0x1, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 },
   { OP_EX2,    0x0, 0x0, 0x0, 0x8, 0x0, 0x0, 0x0, 0x0 },
   { OP_LG2,    0x1, 0x1, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 },
   { OP_RCP,    0x1, 0x1, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 },
   { OP_RSQ,    0x1, 0x1, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 },
   { OP_DFDX,   0x1, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 },
   { OP_DFDY,   0x1, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 },
};

void TargetNV50::initOpInfo()
{
   unsigned int i, j;

   static const operation commutativeList[] =
   {
      OP_ADD, OP_MUL, OP_MAD, OP_FMA, OP_AND, OP_OR, OP_XOR, OP_MAX, OP_MIN,
      OP_SET_AND, OP_SET_OR, OP_SET_XOR, OP_SET, OP_SELP, OP_SLCT
   };
   static const operation shortFormList[] =
   {
      OP_MOV, OP_ADD, OP_SUB, OP_MUL, OP_MAD, OP_SAD, OP_RCP, OP_LINTERP,
      OP_PINTERP, OP_TEX, OP_TXF
   };
   static const operation noDestList[] =
   {
      OP_STORE, OP_WRSV, OP_EXPORT, OP_BRA, OP_CALL, OP_RET, OP_EXIT,
      OP_DISCARD, OP_CONT, OP_BREAK, OP_PRECONT, OP_PREBREAK, OP_PRERET,
      OP_JOIN, OP_JOINAT, OP_BRKPT, OP_MEMBAR, OP_EMIT, OP_RESTART,
      OP_QUADON, OP_QUADPOP, OP_TEXBAR, OP_SUSTB, OP_SUSTP, OP_SUREDP,
      OP_SUREDB, OP_BAR
   };
   static const operation noPredList[] =
   {
      OP_CALL, OP_PREBREAK, OP_PRERET, OP_QUADON, OP_QUADPOP, OP_JOINAT,
      OP_EMIT, OP_RESTART
   };

   for (i = 0; i < DATA_FILE_COUNT; ++i)
      nativeFileMap[i] = (DataFile)i;
   nativeFileMap[FILE_PREDICATE] = FILE_FLAGS;

   for (i = 0; i < OP_LAST; ++i) {
      opInfo[i].variants = NULL;
      opInfo[i].op = (operation)i;
      opInfo[i].srcTypes = 1 << (int)TYPE_F32;
      opInfo[i].dstTypes = 1 << (int)TYPE_F32;
      opInfo[i].immdBits = 0xffffffff;
      opInfo[i].srcNr = operationSrcNr[i];

      for (j = 0; j < opInfo[i].srcNr; ++j) {
         opInfo[i].srcMods[j] = 0;
         opInfo[i].srcFiles[j] = 1 << (int)FILE_GPR;
      }
      opInfo[i].dstMods = 0;
      opInfo[i].dstFiles = 1 << (int)FILE_GPR;

      opInfo[i].hasDest = 1;
      opInfo[i].vector = (i >= OP_TEX && i <= OP_TEXCSAA);
      opInfo[i].commutative = false; /* set below */
      opInfo[i].pseudo = (i < OP_MOV);
      opInfo[i].predicate = !opInfo[i].pseudo;
      opInfo[i].flow = (i >= OP_BRA && i <= OP_JOIN);
      opInfo[i].minEncSize = 8; /* set below */
   }
   for (i = 0; i < ARRAY_SIZE(commutativeList); ++i)
      opInfo[commutativeList[i]].commutative = true;
   for (i = 0; i < ARRAY_SIZE(shortFormList); ++i)
      opInfo[shortFormList[i]].minEncSize = 4;
   for (i = 0; i < ARRAY_SIZE(noDestList); ++i)
      opInfo[noDestList[i]].hasDest = 0;
   for (i = 0; i < ARRAY_SIZE(noPredList); ++i)
      opInfo[noPredList[i]].predicate = 0;

   for (i = 0; i < ARRAY_SIZE(_initProps); ++i) {
      const struct nv50_opProperties *prop = &_initProps[i];

      for (int s = 0; s < 3; ++s) {
         if (prop->mNeg & (1 << s))
            opInfo[prop->op].srcMods[s] |= NV50_IR_MOD_NEG;
         if (prop->mAbs & (1 << s))
            opInfo[prop->op].srcMods[s] |= NV50_IR_MOD_ABS;
         if (prop->mNot & (1 << s))
            opInfo[prop->op].srcMods[s] |= NV50_IR_MOD_NOT;
         if (prop->fConst & (1 << s))
            opInfo[prop->op].srcFiles[s] |= 1 << (int)FILE_MEMORY_CONST;
         if (prop->fShared & (1 << s))
            opInfo[prop->op].srcFiles[s] |= 1 << (int)FILE_MEMORY_SHARED;
         if (prop->fAttrib & (1 << s))
            opInfo[prop->op].srcFiles[s] |= 1 << (int)FILE_SHADER_INPUT;
         if (prop->fImm & (1 << s))
            opInfo[prop->op].srcFiles[s] |= 1 << (int)FILE_IMMEDIATE;
      }
      if (prop->mSat & 8)
         opInfo[prop->op].dstMods = NV50_IR_MOD_SAT;
   }

   if (chipset >= 0xa0)
      opInfo[OP_MUL].dstMods = NV50_IR_MOD_SAT;
}

unsigned int
TargetNV50::getFileSize(DataFile file) const
{
   switch (file) {
   case FILE_NULL:          return 0;
   case FILE_GPR:           return 254; // in 16-bit units **
   case FILE_PREDICATE:     return 0;
   case FILE_FLAGS:         return 4;
   case FILE_ADDRESS:       return 4;
   case FILE_BARRIER:       return 0;
   case FILE_IMMEDIATE:     return 0;
   case FILE_MEMORY_CONST:  return 65536;
   case FILE_SHADER_INPUT:  return 0x200;
   case FILE_SHADER_OUTPUT: return 0x200;
   case FILE_MEMORY_BUFFER: return 0xffffffff;
   case FILE_MEMORY_GLOBAL: return 0xffffffff;
   case FILE_MEMORY_SHARED: return 16 << 10;
   case FILE_MEMORY_LOCAL:  return 48 << 10;
   case FILE_SYSTEM_VALUE:  return 16;
   default:
      assert(!"invalid file");
      return 0;
   }
   // ** only first 128 units encodable for 16-bit regs
}

unsigned int
TargetNV50::getFileUnit(DataFile file) const
{
   if (file == FILE_GPR || file == FILE_ADDRESS)
      return 1;
   if (file == FILE_SYSTEM_VALUE)
      return 2;
   return 0;
}

uint32_t
TargetNV50::getSVAddress(DataFile shaderFile, const Symbol *sym) const
{
   switch (sym->reg.data.sv.sv) {
   case SV_FACE:
      return 0x3fc;
   case SV_POSITION:
   {
      uint32_t addr = sysvalLocation[sym->reg.data.sv.sv];
      for (int c = 0; c < sym->reg.data.sv.index; ++c)
         if (wposMask & (1 << c))
            addr += 4;
      return addr;
   }
   case SV_PRIMITIVE_ID:
      return shaderFile == FILE_SHADER_INPUT ? 0x18 :
         sysvalLocation[sym->reg.data.sv.sv];
   case SV_NCTAID:
      return sym->reg.data.sv.index >= 2 ? 0x10 : 0x8 + 2 * sym->reg.data.sv.index;
   case SV_CTAID:
      return sym->reg.data.sv.index >= 2 ? 0x12 : 0xc + 2 * sym->reg.data.sv.index;
   case SV_NTID:
      return 0x2 + 2 * sym->reg.data.sv.index;
   case SV_TID:
   case SV_COMBINED_TID:
      return 0;
   case SV_SAMPLE_POS:
      return 0; /* sample position is handled differently */
   case SV_THREAD_KILL:
      return 0;
   default:
      return sysvalLocation[sym->reg.data.sv.sv];
   }
}

// long:  rrr, arr, rcr, acr, rrc, arc, gcr, grr
// short: rr, ar, rc, gr
// immd:  ri, gi
bool
TargetNV50::insnCanLoad(const Instruction *i, int s,
                        const Instruction *ld) const
{
   DataFile sf = ld->src(0).getFile();

   // immediate 0 can be represented by GPR $r63/$r127
   // this does not work with global memory ld/st/atom
   if (sf == FILE_IMMEDIATE && ld->getSrc(0)->reg.data.u64 == 0)
      return (!i->isPseudo() &&
              !i->asTex() &&
              i->op != OP_EXPORT &&
              i->op != OP_STORE &&
              ((i->op != OP_ATOM && i->op != OP_LOAD) ||
               i->src(0).getFile() != FILE_MEMORY_GLOBAL));

   if (sf == FILE_IMMEDIATE && (i->predSrc >= 0 || i->flagsDef >= 0))
      return false;
   if (s >= opInfo[i->op].srcNr)
      return false;
   if (!(opInfo[i->op].srcFiles[s] & (1 << (int)sf)))
      return false;
   if (s == 2 && i->src(1).getFile() != FILE_GPR)
      return false;

   // NOTE: don't rely on flagsDef
   if (sf == FILE_IMMEDIATE)
      for (int d = 0; i->defExists(d); ++d)
         if (i->def(d).getFile() == FILE_FLAGS)
            return false;

   unsigned mode = 0;

   for (int z = 0; z < Target::operationSrcNr[i->op]; ++z) {
      DataFile zf = (z == s) ? sf : i->src(z).getFile();
      switch (zf) {
      case FILE_GPR:
         break;
      case FILE_MEMORY_SHARED:
      case FILE_SHADER_INPUT:
         mode |= 1 << (z * 2);
         break;
      case FILE_MEMORY_CONST:
         mode |= 2 << (z * 2);
         break;
      case FILE_IMMEDIATE:
         mode |= 3 << (z * 2);
      default:
         break;
      }
   }

   switch (mode) {
   case 0x00:
   case 0x01:
   case 0x03:
   case 0x08:
   case 0x0c:
   case 0x20:
   case 0x21:
      break;
   case 0x09:
      // Shader inputs get transformed to p[] in geometry shaders, and those
      // aren't allowed to be used at the same time as c[].
      if (ld->bb->getProgram()->getType() == Program::TYPE_GEOMETRY)
         return false;
      break;
   case 0x0d:
      if (ld->bb->getProgram()->getType() != Program::TYPE_GEOMETRY)
         return false;
      break;
   default:
      return false;
   }

   uint8_t ldSize;

   if ((i->op == OP_MUL || i->op == OP_MAD) && !isFloatType(i->dType)) {
      // 32-bit MUL will be split into 16-bit MULs
      if (ld->src(0).isIndirect(0))
         return false;
      if (sf == FILE_IMMEDIATE)
         return false;
      if (i->subOp == NV50_IR_SUBOP_MUL_HIGH && sf == FILE_MEMORY_CONST)
         return false;
      ldSize = 2;
   } else {
      ldSize = typeSizeof(ld->dType);
   }

   if (sf == FILE_IMMEDIATE) {
      if (ldSize == 2 && (i->op == OP_AND || i->op == OP_OR || i->op == OP_XOR))
         return false;
      return ldSize <= 4;
   }


   // Check if memory access is encodable:

   if (ldSize < 4 && sf == FILE_SHADER_INPUT) // no < 4-byte aligned a[] access
      return false;
   if (ld->getSrc(0)->reg.data.offset > (int32_t)(127 * ldSize))
      return false;

   if (ld->src(0).isIndirect(0)) {
      for (int z = 0; i->srcExists(z); ++z)
         if (i->src(z).isIndirect(0))
            return false;

      // s[] access only possible in CP, $aX always applies
      if (sf == FILE_MEMORY_SHARED)
         return true;
      if (!ld->bb) // can't check type ...
         return false;
      Program::Type pt = ld->bb->getProgram()->getType();

      // $aX applies to c[] only in VP, FP, GP if p[] is not accessed
      if (pt == Program::TYPE_COMPUTE)
         return false;
      if (pt == Program::TYPE_GEOMETRY) {
         if (sf == FILE_MEMORY_CONST)
            return i->src(s).getFile() != FILE_SHADER_INPUT;
         return sf == FILE_SHADER_INPUT;
      }
      return sf == FILE_MEMORY_CONST;
   }
   return true;
}

bool
TargetNV50::insnCanLoadOffset(const Instruction *i, int s, int offset) const
{
   if (!i->src(s).isIndirect(0))
      return true;
   offset += i->src(s).get()->reg.data.offset;
   if (i->op == OP_LOAD || i->op == OP_STORE || i->op == OP_ATOM) {
      // There are some restrictions in theory, but in practice they're never
      // going to be hit. However offsets on global/shared memory are just
      // plain not supported.
      return i->src(s).getFile() != FILE_MEMORY_GLOBAL &&
         i->src(s).getFile() != FILE_MEMORY_SHARED;
   }
   return offset >= 0 && offset <= (int32_t)(127 * i->src(s).get()->reg.size);
}

bool
TargetNV50::isAccessSupported(DataFile file, DataType ty) const
{
   if (ty == TYPE_B96 || ty == TYPE_NONE)
      return false;
   if (typeSizeof(ty) > 4)
      return (file == FILE_MEMORY_LOCAL) || (file == FILE_MEMORY_GLOBAL) ||
             (file == FILE_MEMORY_BUFFER);
   return true;
}

bool
TargetNV50::isOpSupported(operation op, DataType ty) const
{
   if (ty == TYPE_F64 && chipset < 0xa0)
      return false;

   switch (op) {
   case OP_PRERET:
      return chipset >= 0xa0;
   case OP_TXG:
      return chipset >= 0xa3 && chipset != 0xaa && chipset != 0xac;
   case OP_POW:
   case OP_SQRT:
   case OP_DIV:
   case OP_MOD:
   case OP_SET_AND:
   case OP_SET_OR:
   case OP_SET_XOR:
   case OP_SLCT:
   case OP_SELP:
   case OP_POPCNT:
   case OP_INSBF:
   case OP_EXTBF:
   case OP_EXIT: // want exit modifier instead (on NOP if required)
   case OP_MEMBAR:
   case OP_SHLADD:
   case OP_XMAD:
      return false;
   case OP_SAD:
      return ty == TYPE_S32;
   case OP_SET:
      return !isFloatType(ty);
   default:
      return true;
   }
}

bool
TargetNV50::isModSupported(const Instruction *insn, int s, Modifier mod) const
{
   if (!isFloatType(insn->dType)) {
      switch (insn->op) {
      case OP_ABS:
      case OP_NEG:
      case OP_CVT:
      case OP_CEIL:
      case OP_FLOOR:
      case OP_TRUNC:
      case OP_AND:
      case OP_OR:
      case OP_XOR:
         break;
      case OP_ADD:
         if (insn->src(s ? 0 : 1).mod.neg())
            return false;
         break;
      case OP_SUB:
         if (s == 0)
            return insn->src(1).mod.neg() ? false : true;
         break;
      case OP_SET:
         if (insn->sType != TYPE_F32)
            return false;
         break;
      default:
         return false;
      }
   }
   if (s >= opInfo[insn->op].srcNr || s >= 3)
      return false;
   return (mod & Modifier(opInfo[insn->op].srcMods[s])) == mod;
}

bool
TargetNV50::mayPredicate(const Instruction *insn, const Value *pred) const
{
   if (insn->getPredicate() || insn->flagsSrc >= 0)
      return false;
   for (int s = 0; insn->srcExists(s); ++s)
      if (insn->src(s).getFile() == FILE_IMMEDIATE)
         return false;
   return opInfo[insn->op].predicate;
}

bool
TargetNV50::isSatSupported(const Instruction *insn) const
{
   if (insn->op == OP_CVT)
      return true;
   if (insn->dType != TYPE_F32)
      return false;
   return opInfo[insn->op].dstMods & NV50_IR_MOD_SAT;
}

int TargetNV50::getLatency(const Instruction *i) const
{
   // TODO: tune these values
   if (i->op == OP_LOAD) {
      switch (i->src(0).getFile()) {
      case FILE_MEMORY_LOCAL:
      case FILE_MEMORY_GLOBAL:
      case FILE_MEMORY_BUFFER:
         return 100; // really 400 to 800
      default:
         return 22;
      }
   }
   return 22;
}

// These are "inverse" throughput values, i.e. the number of cycles required
// to issue a specific instruction for a full warp (32 threads).
//
// Assuming we have more than 1 warp in flight, a higher issue latency results
// in a lower result latency since the MP will have spent more time with other
// warps.
// This also helps to determine the number of cycles between instructions in
// a single warp.
//
int TargetNV50::getThroughput(const Instruction *i) const
{
   // TODO: tune these values
   if (i->dType == TYPE_F32) {
      switch (i->op) {
      case OP_RCP:
      case OP_RSQ:
      case OP_LG2:
      case OP_SIN:
      case OP_COS:
      case OP_PRESIN:
      case OP_PREEX2:
         return 16;
      default:
         return 4;
      }
   } else
   if (i->dType == TYPE_U32 || i->dType == TYPE_S32) {
      return 4;
   } else
   if (i->dType == TYPE_F64) {
      return 32;
   } else {
      return 1;
   }
}

static void
recordLocation(uint16_t *locs, uint8_t *masks,
               const struct nv50_ir_varying *var)
{
   uint16_t addr = var->slot[0] * 4;

   switch (var->sn) {
   case TGSI_SEMANTIC_POSITION: locs[SV_POSITION] = addr; break;
   case TGSI_SEMANTIC_INSTANCEID: locs[SV_INSTANCE_ID] = addr; break;
   case TGSI_SEMANTIC_VERTEXID: locs[SV_VERTEX_ID] = addr; break;
   case TGSI_SEMANTIC_PRIMID: locs[SV_PRIMITIVE_ID] = addr; break;
   case TGSI_SEMANTIC_LAYER: locs[SV_LAYER] = addr; break;
   case TGSI_SEMANTIC_VIEWPORT_INDEX: locs[SV_VIEWPORT_INDEX] = addr; break;
   default:
      break;
   }
   if (var->sn == TGSI_SEMANTIC_POSITION && masks)
      masks[0] = var->mask;
}

void
TargetNV50::parseDriverInfo(const struct nv50_ir_prog_info *info,
                            const struct nv50_ir_prog_info_out *info_out)
{
   unsigned int i;
   for (i = 0; i < info_out->numOutputs; ++i)
      recordLocation(sysvalLocation, NULL, &info_out->out[i]);
   for (i = 0; i < info_out->numInputs; ++i)
      recordLocation(sysvalLocation, &wposMask, &info_out->in[i]);
   for (i = 0; i < info_out->numSysVals; ++i)
      recordLocation(sysvalLocation, NULL, &info_out->sv[i]);

   if (sysvalLocation[SV_POSITION] >= 0x200) {
      // not assigned by driver, but we need it internally
      wposMask = 0x8;
      sysvalLocation[SV_POSITION] = 0;
   }

   Target::parseDriverInfo(info, info_out);
}

} // namespace nv50_ir