nv50/ir: only unspill once ahead of a group of instructions

[mesa.git] / src / gallium / drivers / nouveau / codegen / nv50_ir_emit_gk110.cpp

diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gk110.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gk110.cpp
index 6bb9620d5f728cb9475f0cf3e9d8f120305de017..b1064bf0a92fc13e93745353091ac9171df80bb8 100644 (file)
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gk110.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gk110.cpp
@@ -75,8 +75,10 @@ private:
    void emitLOAD(const Instruction *);
    void emitSTORE(const Instruction *);
    void emitMOV(const Instruction *);
+   void emitMEMBAR(const Instruction *);
 
    void emitINTERP(const Instruction *);
+   void emitAFETCH(const Instruction *);
    void emitPFETCH(const Instruction *);
    void emitVFETCH(const Instruction *);
    void emitEXPORT(const Instruction *);
@@ -120,6 +122,8 @@ private:
 
    void emitPIXLD(const Instruction *);
 
+   void emitBAR(const Instruction *);
+
    void emitFlow(const Instruction *);
 
    inline void defId(const ValueDef&, const int pos);
@@ -572,8 +576,8 @@ CodeEmitterGK110::emitIMUL(const Instruction *i)
    if (isLIMM(i->src(1), TYPE_S32)) {
       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 {
@@ -692,14 +696,9 @@ CodeEmitterGK110::emitIMAD(const Instruction *i)
    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;
+   SAT_(35);
 }
 
 void
@@ -930,6 +929,7 @@ CodeEmitterGK110::emitCVT(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;
@@ -967,8 +967,8 @@ CodeEmitterGK110::emitSET(const CmpInstruction *i)
       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;
@@ -990,8 +990,12 @@ CodeEmitterGK110::emitSET(const CmpInstruction *i)
       }
       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;
@@ -1245,6 +1249,13 @@ CodeEmitterGK110::emitPIXLD(const Instruction *i)
    code[1] |= 0x00070000;
 }
 
+void
+CodeEmitterGK110::emitBAR(const Instruction *i)
+{
+   /* TODO */
+   emitNOP(i);
+}
+
 void
 CodeEmitterGK110::emitFlow(const Instruction *i)
 {
@@ -1316,6 +1327,8 @@ 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;
@@ -1323,6 +1336,23 @@ CodeEmitterGK110::emitFlow(const Instruction *i)
    }
 }
 
+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)
 {
@@ -1403,6 +1433,30 @@ CodeEmitterGK110::emitInterpMode(const Instruction *i)
    code[1] |= (i->ipa & 0xc) << (19 - 2);
 }
 
+static void
+interpApply(const InterpEntry *entry, uint32_t *code,
+      bool force_persample_interp, bool flatshade)
+{
+   int ipa = entry->ipa;
+   int reg = entry->reg;
+   int loc = entry->loc;
+
+   if (flatshade &&
+       (ipa & NV50_IR_INTERP_MODE_MASK) == NV50_IR_INTERP_SC) {
+      ipa = NV50_IR_INTERP_FLAT;
+      reg = 0xff;
+   } else if (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)
 {
@@ -1414,10 +1468,13 @@ 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);
@@ -1583,6 +1640,7 @@ CodeEmitterGK110::getSRegEncoding(const ValueRef& ref)
    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;
@@ -1625,6 +1683,14 @@ CodeEmitterGK110::emitMOV(const Instruction *i)
    }
 }
 
+void CodeEmitterGK110::emitMEMBAR(const Instruction *i)
+{
+   code[0] = 0x00000002 | NV50_IR_SUBOP_MEMBAR_SCOPE(i->subOp) << 8;
+   code[1] = 0x7cc00000;
+
+   emitPredicate(i);
+}
+
 bool
 CodeEmitterGK110::emitInstruction(Instruction *insn)
 {
@@ -1692,6 +1758,9 @@ CodeEmitterGK110::emitInstruction(Instruction *insn)
    case OP_EXPORT:
       emitEXPORT(insn);
       break;
+   case OP_AFETCH:
+      emitAFETCH(insn);
+      break;
    case OP_PFETCH:
       emitPFETCH(insn);
       break;
@@ -1850,6 +1919,12 @@ CodeEmitterGK110::emitInstruction(Instruction *insn)
       emitNOP(insn);
       insn->join = 1;
       break;
+   case OP_BAR:
+      emitBAR(insn);
+      break;
+   case OP_MEMBAR:
+      emitMEMBAR(insn);
+      break;
    case OP_PHI:
    case OP_UNION:
    case OP_CONSTRAINT:
@@ -1862,7 +1937,7 @@ CodeEmitterGK110::emitInstruction(Instruction *insn)
       ERROR("operation should have been lowered\n");
       return false;
    default:
-      ERROR("unknow op\n");
+      ERROR("unknown op: %u\n", insn->op);
       return false;
    }