void setCAddress14(const ValueRef&);
void setShortImmediate(const Instruction *, const int s);
void setImmediate32(const Instruction *, const int s, Modifier);
+ void setSUConst16(const Instruction *, const int s);
void modNegAbsF32_3b(const Instruction *, const int s);
void emitInterpMode(const Instruction *);
void emitLoadStoreType(DataType ty, const int pos);
void emitCachingMode(CacheMode c, const int pos);
+ void emitSUGType(DataType, const int pos);
+ void emitSUCachingMode(CacheMode c);
inline uint8_t getSRegEncoding(const ValueRef&);
void emitLOAD(const Instruction *);
void emitSTORE(const Instruction *);
void emitMOV(const Instruction *);
+ void emitATOM(const Instruction *);
+ void emitCCTL(const Instruction *);
void emitINTERP(const Instruction *);
+ void emitAFETCH(const Instruction *);
void emitPFETCH(const Instruction *);
void emitVFETCH(const Instruction *);
void emitEXPORT(const Instruction *);
void emitISAD(const Instruction *);
void emitFMAD(const Instruction *);
void emitDMAD(const Instruction *);
+ void emitMADSP(const Instruction *i);
void emitNOT(const Instruction *);
void emitLogicOp(const Instruction *, uint8_t subOp);
void emitINSBF(const Instruction *);
void emitEXTBF(const Instruction *);
void emitBFIND(const Instruction *);
+ void emitPERMT(const Instruction *);
void emitShift(const Instruction *);
void emitSFnOp(const Instruction *, uint8_t subOp);
void emitPIXLD(const Instruction *);
+ void emitBAR(const Instruction *);
+ void emitMEMBAR(const Instruction *);
+
void emitFlow(const Instruction *);
+ void emitVOTE(const Instruction *);
+
+ void emitSULDGB(const TexInstruction *);
+ void emitSUSTGx(const TexInstruction *);
+ void emitSUCLAMPMode(uint16_t);
+ void emitSUCalc(Instruction *);
+
+ void emitVSHL(const Instruction *);
+ void emitVectorSubOp(const Instruction *);
+
inline void defId(const ValueDef&, const int pos);
inline void srcId(const ValueRef&, const int pos);
inline void srcId(const ValueRef *, const int pos);
srcId(i->src(s), s ? ((s == 2) ? 42 : s1) : 10);
break;
default:
+ if (i->op == OP_SELP) {
+ assert(s == 2 && i->src(s).getFile() == FILE_PREDICATE);
+ srcId(i->src(s), 42);
+ }
// ignore here, can be predicate or flags, but must not be address
break;
}
}
}
+void
+CodeEmitterGK110::emitMADSP(const Instruction *i)
+{
+ emitForm_21(i, 0x140, 0xa40);
+
+ if (i->subOp == NV50_IR_SUBOP_MADSP_SD) {
+ code[1] |= 0x00c00000;
+ } else {
+ code[1] |= (i->subOp & 0x00f) << 19; // imadp1
+ code[1] |= (i->subOp & 0x0f0) << 20; // imadp2
+ code[1] |= (i->subOp & 0x100) << 11; // imadp3
+ code[1] |= (i->subOp & 0x200) << 15; // imadp3
+ code[1] |= (i->subOp & 0xc00) << 12; // imadp3
+ }
+
+ if (i->flagsDef >= 0)
+ code[1] |= 1 << 18;
+}
+
void
CodeEmitterGK110::emitFMUL(const Instruction *i)
{
assert(!i->src(0).mod.neg() && !i->src(1).mod.neg());
assert(!i->src(0).mod.abs() && !i->src(1).mod.abs());
- if (isLIMM(i->src(1), TYPE_S32)) {
+ if (i->src(1).getFile() == FILE_IMMEDIATE) {
emitForm_L(i, 0x280, 2, Modifier(0));
- assert(i->subOp != NV50_IR_SUBOP_MUL_HIGH);
-
+ if (i->subOp == NV50_IR_SUBOP_MUL_HIGH)
+ code[1] |= 1 << 24;
if (i->sType == TYPE_S32)
code[1] |= 3 << 25;
} else {
if (i->sType == TYPE_S32)
code[1] |= (1 << 19) | (1 << 24);
- if (code[0] & 0x1) {
- assert(!i->subOp);
- SAT_(39);
- } else {
- if (i->subOp == NV50_IR_SUBOP_MUL_HIGH)
- code[1] |= 1 << 25;
- SAT_(35);
- }
+ if (i->subOp == NV50_IR_SUBOP_MUL_HIGH)
+ code[1] |= 1 << 25;
+
+ if (i->flagsDef >= 0) code[1] |= 1 << 18;
+ if (i->flagsSrc >= 0) code[1] |= 1 << 20;
+
+ SAT_(35);
}
void
void
CodeEmitterGK110::emitLogicOp(const Instruction *i, uint8_t subOp)
{
+ if (i->def(0).getFile() == FILE_PREDICATE) {
+ code[0] = 0x00000002 | (subOp << 27);
+ code[1] = 0x84800000;
+
+ emitPredicate(i);
+
+ defId(i->def(0), 5);
+ srcId(i->src(0), 14);
+ if (i->src(0).mod == Modifier(NV50_IR_MOD_NOT)) code[0] |= 1 << 17;
+ srcId(i->src(1), 32);
+ if (i->src(1).mod == Modifier(NV50_IR_MOD_NOT)) code[1] |= 1 << 3;
+
+ if (i->defExists(1)) {
+ defId(i->def(1), 2);
+ } else {
+ code[0] |= 7 << 2;
+ }
+ // (a OP b) OP c
+ if (i->predSrc != 2 && i->srcExists(2)) {
+ code[1] |= subOp << 16;
+ srcId(i->src(2), 42);
+ if (i->src(2).mod == Modifier(NV50_IR_MOD_NOT)) code[1] |= 1 << 13;
+ } else {
+ code[1] |= 7 << 10;
+ }
+ } else
if (isLIMM(i->src(1), TYPE_S32)) {
emitForm_L(i, 0x200, 0, i->src(1).mod);
code[1] |= subOp << 24;
code[1] |= 0x1000;
}
+void
+CodeEmitterGK110::emitPERMT(const Instruction *i)
+{
+ emitForm_21(i, 0x1e0, 0xb60);
+
+ code[1] |= i->subOp << 19;
+}
+
void
CodeEmitterGK110::emitShift(const Instruction *i)
{
code[0] |= typeSizeofLog2(dType) << 10;
code[0] |= typeSizeofLog2(i->sType) << 12;
+ code[1] |= i->subOp << 12;
if (isSignedIntType(dType))
code[0] |= 0x4000;
code[0] = (code[0] & ~0xfc) | ((code[0] << 3) & 0xe0);
if (i->defExists(1))
defId(i->def(1), 2);
- else
- code[0] |= 0x1c;
+ else
+ code[0] |= 0x1c;
} else {
switch (i->sType) {
case TYPE_F32: op2 = 0x000; op1 = 0x800; break;
}
FTZ_(3a);
- if (i->dType == TYPE_F32)
- code[1] |= 1 << 23;
+ if (i->dType == TYPE_F32) {
+ if (isFloatType(i->sType))
+ code[1] |= 1 << 23;
+ else
+ code[1] |= 1 << 15;
+ }
}
if (i->sType == TYPE_S32)
code[1] |= 1 << 19;
}
}
+static void
+selpFlip(const FixupEntry *entry, uint32_t *code, const FixupData& data)
+{
+ int loc = entry->loc;
+ if (data.force_persample_interp)
+ code[loc + 1] |= 1 << 13;
+ else
+ code[loc + 1] &= ~(1 << 13);
+}
+
void CodeEmitterGK110::emitSELP(const Instruction *i)
{
emitForm_21(i, 0x250, 0x050);
- if ((i->cc == CC_NOT_P) ^ (bool)(i->src(2).mod & Modifier(NV50_IR_MOD_NOT)))
+ if (i->src(2).mod & Modifier(NV50_IR_MOD_NOT))
code[1] |= 1 << 13;
+
+ if (i->subOp == 1) {
+ addInterp(0, 0, selpFlip);
+ }
}
void CodeEmitterGK110::emitTEXBAR(const Instruction *i)
defId(i->def(0), 2);
srcId(i->src(0), 10);
- srcId(i->srcExists(1) ? i->src(1) : i->src(0), 23);
+ srcId((i->srcExists(1) && i->predSrc != 1) ? i->src(1) : i->src(0), 23);
if (i->op == OP_QUADOP && progType != Program::TYPE_FRAGMENT)
code[1] |= 1 << 9; // dall
code[1] |= 0x00070000;
}
+void
+CodeEmitterGK110::emitBAR(const Instruction *i)
+{
+ code[0] = 0x00000002;
+ code[1] = 0x85400000;
+
+ switch (i->subOp) {
+ case NV50_IR_SUBOP_BAR_ARRIVE: code[1] |= 0x08; break;
+ case NV50_IR_SUBOP_BAR_RED_AND: code[1] |= 0x50; break;
+ case NV50_IR_SUBOP_BAR_RED_OR: code[1] |= 0x90; break;
+ case NV50_IR_SUBOP_BAR_RED_POPC: code[1] |= 0x10; break;
+ default:
+ assert(i->subOp == NV50_IR_SUBOP_BAR_SYNC);
+ break;
+ }
+
+ emitPredicate(i);
+
+ // barrier id
+ if (i->src(0).getFile() == FILE_GPR) {
+ srcId(i->src(0), 10);
+ } else {
+ ImmediateValue *imm = i->getSrc(0)->asImm();
+ assert(imm);
+ code[0] |= imm->reg.data.u32 << 10;
+ code[1] |= 0x8000;
+ }
+
+ // thread count
+ if (i->src(1).getFile() == FILE_GPR) {
+ srcId(i->src(1), 23);
+ } else {
+ ImmediateValue *imm = i->getSrc(0)->asImm();
+ assert(imm);
+ assert(imm->reg.data.u32 <= 0xfff);
+ code[0] |= imm->reg.data.u32 << 23;
+ code[1] |= imm->reg.data.u32 >> 9;
+ code[1] |= 0x4000;
+ }
+
+ if (i->srcExists(2) && (i->predSrc != 2)) {
+ srcId(i->src(2), 32 + 10);
+ if (i->src(2).mod == Modifier(NV50_IR_MOD_NOT))
+ code[1] |= 1 << 13;
+ } else {
+ code[1] |= 7 << 10;
+ }
+}
+
+void CodeEmitterGK110::emitMEMBAR(const Instruction *i)
+{
+ code[0] = 0x00000002 | NV50_IR_SUBOP_MEMBAR_SCOPE(i->subOp) << 8;
+ code[1] = 0x7cc00000;
+
+ emitPredicate(i);
+}
+
void
CodeEmitterGK110::emitFlow(const Instruction *i)
{
} else
if (mask & 2) {
int32_t pcRel = f->target.bb->binPos - (codeSize + 8);
+ if (writeIssueDelays && !(f->target.bb->binPos & 0x3f))
+ pcRel += 8;
// currently we don't want absolute branches
assert(!f->absolute);
code[0] |= (pcRel & 0x1ff) << 23;
}
}
+void
+CodeEmitterGK110::emitVOTE(const Instruction *i)
+{
+ assert(i->src(0).getFile() == FILE_PREDICATE &&
+ i->def(1).getFile() == FILE_PREDICATE);
+
+ code[0] = 0x00000002;
+ code[1] = 0x86c00000 | (i->subOp << 19);
+
+ emitPredicate(i);
+
+ defId(i->def(0), 2);
+ defId(i->def(1), 48);
+ if (i->src(0).mod == Modifier(NV50_IR_MOD_NOT))
+ code[1] |= 1 << 13;
+ srcId(i->src(0), 42);
+}
+
+void
+CodeEmitterGK110::emitSUGType(DataType ty, const int pos)
+{
+ uint8_t n = 0;
+
+ switch (ty) {
+ case TYPE_S32: n = 1; break;
+ case TYPE_U8: n = 2; break;
+ case TYPE_S8: n = 3; break;
+ default:
+ assert(ty == TYPE_U32);
+ break;
+ }
+ code[pos / 32] |= n << (pos % 32);
+}
+
+void
+CodeEmitterGK110::emitSUCachingMode(CacheMode c)
+{
+ uint8_t n = 0;
+
+ switch (c) {
+ case CACHE_CA:
+// case CACHE_WB:
+ n = 0;
+ break;
+ case CACHE_CG:
+ n = 1;
+ break;
+ case CACHE_CS:
+ n = 2;
+ break;
+ case CACHE_CV:
+// case CACHE_WT:
+ n = 3;
+ break;
+ default:
+ assert(!"invalid caching mode");
+ break;
+ }
+ code[0] |= (n & 1) << 31;
+ code[1] |= (n & 2) >> 1;
+}
+
+void
+CodeEmitterGK110::setSUConst16(const Instruction *i, const int s)
+{
+ const uint32_t offset = i->getSrc(s)->reg.data.offset;
+
+ assert(offset == (offset & 0xfffc));
+
+ code[0] |= offset << 21;
+ code[1] |= offset >> 11;
+ code[1] |= i->getSrc(s)->reg.fileIndex << 5;
+}
+
+void
+CodeEmitterGK110::emitSULDGB(const TexInstruction *i)
+{
+ code[0] = 0x00000002;
+ code[1] = 0x30000000 | (i->subOp << 14);
+
+ if (i->src(1).getFile() == FILE_MEMORY_CONST) {
+ emitLoadStoreType(i->dType, 0x38);
+ emitCachingMode(i->cache, 0x36);
+
+ // format
+ setSUConst16(i, 1);
+ } else {
+ assert(i->src(1).getFile() == FILE_GPR);
+ code[1] |= 0x49800000;
+
+ emitLoadStoreType(i->dType, 0x21);
+ emitSUCachingMode(i->cache);
+
+ srcId(i->src(1), 23);
+ }
+
+ emitSUGType(i->sType, 0x34);
+
+ emitPredicate(i);
+ defId(i->def(0), 2); // destination
+ srcId(i->src(0), 10); // address
+
+ // surface predicate
+ if (!i->srcExists(2) || (i->predSrc == 2)) {
+ code[1] |= 0x7 << 10;
+ } else {
+ if (i->src(2).mod == Modifier(NV50_IR_MOD_NOT))
+ code[1] |= 1 << 13;
+ srcId(i->src(2), 32 + 10);
+ }
+}
+
+void
+CodeEmitterGK110::emitSUSTGx(const TexInstruction *i)
+{
+ assert(i->op == OP_SUSTP);
+
+ code[0] = 0x00000002;
+ code[1] = 0x38000000;
+
+ if (i->src(1).getFile() == FILE_MEMORY_CONST) {
+ code[0] |= i->subOp << 2;
+
+ if (i->op == OP_SUSTP)
+ code[0] |= i->tex.mask << 4;
+
+ emitSUGType(i->sType, 0x8);
+ emitCachingMode(i->cache, 0x36);
+
+ // format
+ setSUConst16(i, 1);
+ } else {
+ assert(i->src(1).getFile() == FILE_GPR);
+
+ code[0] |= i->subOp << 23;
+ code[1] |= 0x41c00000;
+
+ if (i->op == OP_SUSTP)
+ code[0] |= i->tex.mask << 25;
+
+ emitSUGType(i->sType, 0x1d);
+ emitSUCachingMode(i->cache);
+
+ srcId(i->src(1), 2);
+ }
+
+ emitPredicate(i);
+ srcId(i->src(0), 10); // address
+ srcId(i->src(3), 42); // values
+
+ // surface predicate
+ if (!i->srcExists(2) || (i->predSrc == 2)) {
+ code[1] |= 0x7 << 18;
+ } else {
+ if (i->src(2).mod == Modifier(NV50_IR_MOD_NOT))
+ code[1] |= 1 << 21;
+ srcId(i->src(2), 32 + 18);
+ }
+}
+
+void
+CodeEmitterGK110::emitSUCLAMPMode(uint16_t subOp)
+{
+ uint8_t m;
+ switch (subOp & ~NV50_IR_SUBOP_SUCLAMP_2D) {
+ case NV50_IR_SUBOP_SUCLAMP_SD(0, 1): m = 0; break;
+ case NV50_IR_SUBOP_SUCLAMP_SD(1, 1): m = 1; break;
+ case NV50_IR_SUBOP_SUCLAMP_SD(2, 1): m = 2; break;
+ case NV50_IR_SUBOP_SUCLAMP_SD(3, 1): m = 3; break;
+ case NV50_IR_SUBOP_SUCLAMP_SD(4, 1): m = 4; break;
+ case NV50_IR_SUBOP_SUCLAMP_PL(0, 1): m = 5; break;
+ case NV50_IR_SUBOP_SUCLAMP_PL(1, 1): m = 6; break;
+ case NV50_IR_SUBOP_SUCLAMP_PL(2, 1): m = 7; break;
+ case NV50_IR_SUBOP_SUCLAMP_PL(3, 1): m = 8; break;
+ case NV50_IR_SUBOP_SUCLAMP_PL(4, 1): m = 9; break;
+ case NV50_IR_SUBOP_SUCLAMP_BL(0, 1): m = 10; break;
+ case NV50_IR_SUBOP_SUCLAMP_BL(1, 1): m = 11; break;
+ case NV50_IR_SUBOP_SUCLAMP_BL(2, 1): m = 12; break;
+ case NV50_IR_SUBOP_SUCLAMP_BL(3, 1): m = 13; break;
+ case NV50_IR_SUBOP_SUCLAMP_BL(4, 1): m = 14; break;
+ default:
+ return;
+ }
+ code[1] |= m << 20;
+ if (subOp & NV50_IR_SUBOP_SUCLAMP_2D)
+ code[1] |= 1 << 24;
+}
+
+void
+CodeEmitterGK110::emitSUCalc(Instruction *i)
+{
+ ImmediateValue *imm = NULL;
+ uint64_t opc1, opc2;
+
+ if (i->srcExists(2)) {
+ imm = i->getSrc(2)->asImm();
+ if (imm)
+ i->setSrc(2, NULL); // special case, make emitForm_21 not assert
+ }
+
+ switch (i->op) {
+ case OP_SUCLAMP: opc1 = 0xb00; opc2 = 0x580; break;
+ case OP_SUBFM: opc1 = 0xb68; opc2 = 0x1e8; break;
+ case OP_SUEAU: opc1 = 0xb6c; opc2 = 0x1ec; break;
+ default:
+ assert(0);
+ return;
+ }
+ emitForm_21(i, opc2, opc1);
+
+ if (i->op == OP_SUCLAMP) {
+ if (i->dType == TYPE_S32)
+ code[1] |= 1 << 19;
+ emitSUCLAMPMode(i->subOp);
+ }
+
+ if (i->op == OP_SUBFM && i->subOp == NV50_IR_SUBOP_SUBFM_3D)
+ code[1] |= 1 << 18;
+
+ if (i->op != OP_SUEAU) {
+ const uint8_t pos = i->op == OP_SUBFM ? 19 : 16;
+ if (i->def(0).getFile() == FILE_PREDICATE) { // p, #
+ code[0] |= 255 << 2;
+ code[1] |= i->getDef(1)->reg.data.id << pos;
+ } else
+ if (i->defExists(1)) { // r, p
+ assert(i->def(1).getFile() == FILE_PREDICATE);
+ code[1] |= i->getDef(1)->reg.data.id << pos;
+ } else { // r, #
+ code[1] |= 7 << pos;
+ }
+ }
+
+ if (imm) {
+ assert(i->op == OP_SUCLAMP);
+ i->setSrc(2, imm);
+ code[1] |= (imm->reg.data.u32 & 0x3f) << 10; // sint6
+ }
+}
+
+
+void
+CodeEmitterGK110::emitVectorSubOp(const Instruction *i)
+{
+ switch (NV50_IR_SUBOP_Vn(i->subOp)) {
+ case 0:
+ code[1] |= (i->subOp & 0x000f) << 7; // vsrc1
+ code[1] |= (i->subOp & 0x00e0) >> 6; // vsrc2
+ code[1] |= (i->subOp & 0x0100) << 13; // vsrc2
+ code[1] |= (i->subOp & 0x3c00) << 12; // vdst
+ break;
+ default:
+ assert(0);
+ break;
+ }
+}
+
+void
+CodeEmitterGK110::emitVSHL(const Instruction *i)
+{
+ code[0] = 0x00000002;
+ code[1] = 0xb8000000;
+
+ assert(NV50_IR_SUBOP_Vn(i->subOp) == 0);
+
+ if (isSignedType(i->dType)) code[1] |= 1 << 25;
+ if (isSignedType(i->sType)) code[1] |= 1 << 19;
+
+ emitVectorSubOp(i);
+
+ emitPredicate(i);
+ defId(i->def(0), 2);
+ srcId(i->src(0), 10);
+
+ if (i->getSrc(1)->reg.file == FILE_IMMEDIATE) {
+ ImmediateValue *imm = i->getSrc(1)->asImm();
+ assert(imm);
+ code[0] |= (imm->reg.data.u32 & 0x01ff) << 23;
+ code[1] |= (imm->reg.data.u32 & 0xfe00) >> 9;
+ } else {
+ assert(i->getSrc(1)->reg.file == FILE_GPR);
+ code[1] |= 1 << 21;
+ srcId(i->src(1), 23);
+ }
+ srcId(i->src(2), 42);
+
+ if (i->saturate)
+ code[0] |= 1 << 22;
+ if (i->flagsDef >= 0)
+ code[1] |= 1 << 18;
+}
+
+void
+CodeEmitterGK110::emitAFETCH(const Instruction *i)
+{
+ uint32_t offset = i->src(0).get()->reg.data.offset & 0x7ff;
+
+ code[0] = 0x00000002 | (offset << 23);
+ code[1] = 0x7d000000 | (offset >> 9);
+
+ if (i->getSrc(0)->reg.file == FILE_SHADER_OUTPUT)
+ code[1] |= 0x8;
+
+ emitPredicate(i);
+
+ defId(i->def(0), 2);
+ srcId(i->src(0).getIndirect(0), 10);
+}
+
void
CodeEmitterGK110::emitPFETCH(const Instruction *i)
{
emitPredicate(i);
+ const int src1 = (i->predSrc == 1) ? 2 : 1; // if predSrc == 1, !srcExists(2)
+
defId(i->def(0), 2);
- srcId(i->src(1), 10);
+ srcId(i, src1, 10);
}
void
code[1] |= (i->ipa & 0xc) << (19 - 2);
}
+static void
+interpApply(const FixupEntry *entry, uint32_t *code, const FixupData& data)
+{
+ int ipa = entry->ipa;
+ int reg = entry->reg;
+ int loc = entry->loc;
+
+ if (data.flatshade &&
+ (ipa & NV50_IR_INTERP_MODE_MASK) == NV50_IR_INTERP_SC) {
+ ipa = NV50_IR_INTERP_FLAT;
+ reg = 0xff;
+ } else if (data.force_persample_interp &&
+ (ipa & NV50_IR_INTERP_SAMPLE_MASK) == NV50_IR_INTERP_DEFAULT &&
+ (ipa & NV50_IR_INTERP_MODE_MASK) != NV50_IR_INTERP_FLAT) {
+ ipa |= NV50_IR_INTERP_CENTROID;
+ }
+ code[loc + 1] &= ~(0xf << 19);
+ code[loc + 1] |= (ipa & 0x3) << 21;
+ code[loc + 1] |= (ipa & 0xc) << (19 - 2);
+ code[loc + 0] &= ~(0xff << 23);
+ code[loc + 0] |= reg << 23;
+}
+
void
CodeEmitterGK110::emitINTERP(const Instruction *i)
{
if (i->saturate)
code[1] |= 1 << 18;
- if (i->op == OP_PINTERP)
+ if (i->op == OP_PINTERP) {
srcId(i->src(1), 23);
- else
+ addInterp(i->ipa, SDATA(i->src(1)).id, interpApply);
+ } else {
code[0] |= 0xff << 23;
+ addInterp(i->ipa, 0xff, interpApply);
+ }
srcId(i->src(0).getIndirect(0), 10);
emitInterpMode(i);
switch (i->src(0).getFile()) {
case FILE_MEMORY_GLOBAL: code[1] = 0xe0000000; code[0] = 0x00000000; break;
case FILE_MEMORY_LOCAL: code[1] = 0x7a800000; code[0] = 0x00000002; break;
- case FILE_MEMORY_SHARED: code[1] = 0x7ac00000; code[0] = 0x00000002; break;
+ case FILE_MEMORY_SHARED:
+ code[0] = 0x00000002;
+ if (i->subOp == NV50_IR_SUBOP_STORE_UNLOCKED)
+ code[1] = 0x78400000;
+ else
+ code[1] = 0x7ac00000;
+ break;
default:
assert(!"invalid memory file");
break;
}
- if (i->src(0).getFile() != FILE_MEMORY_GLOBAL)
- offset &= 0xffffff;
-
if (code[0] & 0x2) {
+ offset &= 0xffffff;
emitLoadStoreType(i->dType, 0x33);
if (i->src(0).getFile() == FILE_MEMORY_LOCAL)
emitCachingMode(i->cache, 0x2f);
code[0] |= offset << 23;
code[1] |= offset >> 9;
+ // Unlocked store on shared memory can fail.
+ if (i->src(0).getFile() == FILE_MEMORY_SHARED &&
+ i->subOp == NV50_IR_SUBOP_STORE_UNLOCKED) {
+ assert(i->defExists(0));
+ defId(i->def(0), 32 + 16);
+ }
+
emitPredicate(i);
srcId(i->src(1), 2);
srcId(i->src(0).getIndirect(0), 10);
+ if (i->src(0).getFile() == FILE_MEMORY_GLOBAL &&
+ i->src(0).isIndirect(0) &&
+ i->getIndirect(0, 0)->reg.size == 8)
+ code[1] |= 1 << 23;
}
void
switch (i->src(0).getFile()) {
case FILE_MEMORY_GLOBAL: code[1] = 0xc0000000; code[0] = 0x00000000; break;
case FILE_MEMORY_LOCAL: code[1] = 0x7a000000; code[0] = 0x00000002; break;
- case FILE_MEMORY_SHARED: code[1] = 0x7ac00000; code[0] = 0x00000002; break;
+ case FILE_MEMORY_SHARED:
+ code[0] = 0x00000002;
+ if (i->subOp == NV50_IR_SUBOP_LOAD_LOCKED)
+ code[1] = 0x77400000;
+ else
+ code[1] = 0x7a400000;
+ break;
case FILE_MEMORY_CONST:
if (!i->src(0).isIndirect(0) && typeSizeof(i->dType) == 4) {
emitMOV(i);
code[0] |= offset << 23;
code[1] |= offset >> 9;
+ // Locked store on shared memory can fail.
+ if (i->src(0).getFile() == FILE_MEMORY_SHARED &&
+ i->subOp == NV50_IR_SUBOP_LOAD_LOCKED) {
+ assert(i->defExists(1));
+ defId(i->def(1), 32 + 16);
+ }
+
emitPredicate(i);
defId(i->def(0), 2);
- srcId(i->src(0).getIndirect(0), 10);
+ if (i->getIndirect(0, 0)) {
+ srcId(i->src(0).getIndirect(0), 10);
+ if (i->getIndirect(0, 0)->reg.size == 8)
+ code[1] |= 1 << 23;
+ } else {
+ code[0] |= 255 << 10;
+ }
}
uint8_t
case SV_VERTEX_COUNT: return 0x10;
case SV_INVOCATION_ID: return 0x11;
case SV_YDIR: return 0x12;
+ case SV_THREAD_KILL: return 0x13;
case SV_TID: return 0x21 + SDATA(ref).sv.index;
case SV_CTAID: return 0x25 + SDATA(ref).sv.index;
case SV_NTID: return 0x29 + SDATA(ref).sv.index;
}
}
+static inline bool
+uses64bitAddress(const Instruction *ldst)
+{
+ return ldst->src(0).getFile() == FILE_MEMORY_GLOBAL &&
+ ldst->src(0).isIndirect(0) &&
+ ldst->getIndirect(0, 0)->reg.size == 8;
+}
+
+void
+CodeEmitterGK110::emitATOM(const Instruction *i)
+{
+ const bool hasDst = i->defExists(0);
+ const bool exch = i->subOp == NV50_IR_SUBOP_ATOM_EXCH;
+
+ code[0] = 0x00000002;
+ if (i->subOp == NV50_IR_SUBOP_ATOM_CAS)
+ code[1] = 0x77800000;
+ else
+ code[1] = 0x68000000;
+
+ switch (i->subOp) {
+ case NV50_IR_SUBOP_ATOM_CAS: break;
+ case NV50_IR_SUBOP_ATOM_EXCH: code[1] |= 0x04000000; break;
+ default: code[1] |= i->subOp << 23; break;
+ }
+
+ switch (i->dType) {
+ case TYPE_U32: break;
+ case TYPE_S32: code[1] |= 0x00100000; break;
+ case TYPE_U64: code[1] |= 0x00200000; break;
+ case TYPE_F32: code[1] |= 0x00300000; break;
+ case TYPE_B128: code[1] |= 0x00400000; break; /* TODO: U128 */
+ case TYPE_S64: code[1] |= 0x00500000; break;
+ default: assert(!"unsupported type"); break;
+ }
+
+ emitPredicate(i);
+
+ /* TODO: cas: check that src regs line up */
+ /* TODO: cas: flip bits if $r255 is used */
+ srcId(i->src(1), 23);
+
+ if (hasDst) {
+ defId(i->def(0), 2);
+ } else
+ if (!exch) {
+ code[0] |= 255 << 2;
+ }
+
+ if (hasDst || !exch) {
+ const int32_t offset = SDATA(i->src(0)).offset;
+ assert(offset < 0x80000 && offset >= -0x80000);
+ code[0] |= (offset & 1) << 31;
+ code[1] |= (offset & 0xffffe) >> 1;
+ } else {
+ srcAddr32(i->src(0), 31);
+ }
+
+ if (i->getIndirect(0, 0)) {
+ srcId(i->getIndirect(0, 0), 10);
+ if (i->getIndirect(0, 0)->reg.size == 8)
+ code[1] |= 1 << 19;
+ } else {
+ code[0] |= 255 << 10;
+ }
+}
+
+void
+CodeEmitterGK110::emitCCTL(const Instruction *i)
+{
+ int32_t offset = SDATA(i->src(0)).offset;
+
+ code[0] = 0x00000002 | (i->subOp << 2);
+
+ if (i->src(0).getFile() == FILE_MEMORY_GLOBAL) {
+ code[1] = 0x7b000000;
+ } else {
+ code[1] = 0x7c000000;
+ offset &= 0xffffff;
+ }
+ code[0] |= offset << 23;
+ code[1] |= offset >> 9;
+
+ if (uses64bitAddress(i))
+ code[1] |= 1 << 23;
+ srcId(i->src(0).getIndirect(0), 10);
+
+ emitPredicate(i);
+}
+
bool
CodeEmitterGK110::emitInstruction(Instruction *insn)
{
case OP_EXPORT:
emitEXPORT(insn);
break;
+ case OP_AFETCH:
+ emitAFETCH(insn);
+ break;
case OP_PFETCH:
emitPFETCH(insn);
break;
else
emitIMAD(insn);
break;
+ case OP_MADSP:
+ emitMADSP(insn);
+ break;
case OP_SAD:
emitISAD(insn);
break;
case OP_CEIL:
case OP_FLOOR:
case OP_TRUNC:
- case OP_CVT:
case OP_SAT:
emitCVT(insn);
break;
+ case OP_CVT:
+ if (insn->def(0).getFile() == FILE_PREDICATE ||
+ insn->src(0).getFile() == FILE_PREDICATE)
+ emitMOV(insn);
+ else
+ emitCVT(insn);
+ break;
case OP_RSQ:
emitSFnOp(insn, 5 + 2 * insn->subOp);
break;
case OP_BFIND:
emitBFIND(insn);
break;
+ case OP_PERMT:
+ emitPERMT(insn);
+ break;
case OP_JOIN:
emitNOP(insn);
insn->join = 1;
break;
+ case OP_BAR:
+ emitBAR(insn);
+ break;
+ case OP_MEMBAR:
+ emitMEMBAR(insn);
+ break;
+ case OP_ATOM:
+ emitATOM(insn);
+ break;
+ case OP_CCTL:
+ emitCCTL(insn);
+ break;
+ case OP_VOTE:
+ emitVOTE(insn);
+ break;
+ case OP_SULDB:
+ emitSULDGB(insn->asTex());
+ break;
+ case OP_SUSTB:
+ case OP_SUSTP:
+ emitSUSTGx(insn->asTex());
+ break;
+ case OP_SUBFM:
+ case OP_SUCLAMP:
+ case OP_SUEAU:
+ emitSUCalc(insn);
+ break;
+ case OP_VSHL:
+ emitVSHL(insn);
+ break;
case OP_PHI:
case OP_UNION:
case OP_CONSTRAINT:
ERROR("operation should have been lowered\n");
return false;
default:
- ERROR("unknow op\n");
+ ERROR("unknown op: %u\n", insn->op);
return false;
}
projects / mesa.git / blobdiff