nvc0/ir: make it possible to have the flags def in def0

There's all kinds of logic that doesn't like there being holes in defs
or srcs lists. Avoid them. This also fixes the sched logic for maxwell.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>

diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir.h b/src/gallium/drivers/nouveau/codegen/nv50_ir.h
index b7fd19d2a4c0e23495260fb2d221022f3ed2092b..e161a5a04922725ef00d33679b845c555c1cabc3 100644 (file)
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir.h
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir.h
@@ -662,7 +662,7 @@ public:
    inline const Symbol *asSym() const;
    inline const ImmediateValue *asImm() const;
 
-   inline bool inFile(DataFile f) { return reg.file == f; }
+   inline bool inFile(DataFile f) const { return reg.file == f; }
 
    static inline Value *get(Iterator&);
 

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 03286509edb0dd56f2ca3cdab70d2cae0af56b8c..4210321ae1347d867b48f24736f90c240bcf5a32 100644 (file)
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gk110.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gk110.cpp
@@ -198,7 +198,7 @@ void CodeEmitterGK110::srcAddr32(const ValueRef& src, const int pos)
 
 void CodeEmitterGK110::defId(const ValueDef& def, const int pos)
 {
-   code[pos / 32] |= (def.get() ? DDATA(def).id : GK110_GPR_ZERO) << (pos % 32);
+   code[pos / 32] |= (def.get() && def.getFile() != FILE_FLAGS ? DDATA(def).id : GK110_GPR_ZERO) << (pos % 32);
 }
 
 bool CodeEmitterGK110::isLIMM(const ValueRef& ref, DataType ty, bool mod)
@@ -703,7 +703,7 @@ CodeEmitterGK110::emitUADD(const Instruction *i)
       if (addOp & 2)
          code[1] |= 1 << 27;
 
-      assert(!i->defExists(1));
+      assert(i->flagsDef < 0);
       assert(i->flagsSrc < 0);
 
       SAT_(39);
@@ -714,7 +714,7 @@ CodeEmitterGK110::emitUADD(const Instruction *i)
 
       code[1] |= addOp << 19;
 
-      if (i->defExists(1))
+      if (i->flagsDef >= 0)
          code[1] |= 1 << 18; // write carry
       if (i->flagsSrc >= 0)
          code[1] |= 1 << 14; // add carry

diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gm107.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gm107.cpp
index 8fec6a85800a8e224fb45f00bc31cc55ac61ecf4..f36838cbf8d4f39606590782f89c2f774c48f6a4 100644 (file)
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gm107.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gm107.cpp
@@ -252,7 +252,8 @@ CodeEmitterGM107::emitInsn(uint32_t hi, bool pred)
 void
 CodeEmitterGM107::emitGPR(int pos, const Value *val)
 {
-   emitField(pos, 8, val ? val->reg.data.id : 255);
+   emitField(pos, 8, val && !val->inFile(FILE_FLAGS) ?
+             val->reg.data.id : 255);
 }
 
 void

diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_nvc0.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_nvc0.cpp
index 145868813e31776f89e4cf7e26468d1e2d11e393..5467447e354302e1b77f8a54329af7208923306c 100644 (file)
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_nvc0.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_nvc0.cpp
@@ -195,13 +195,15 @@ CodeEmitterNVC0::srcAddr32(const ValueRef& src, int pos, int shr)
 
 void CodeEmitterNVC0::defId(const ValueDef& def, const int pos)
 {
-   code[pos / 32] |= (def.get() ? DDATA(def).id : 63) << (pos % 32);
+   code[pos / 32] |= (def.get() && def.getFile() != FILE_FLAGS ? DDATA(def).id : 63) << (pos % 32);
 }
 
-void CodeEmitterNVC0::defId(const Instruction *insn, int d, int pos)
+void CodeEmitterNVC0::defId(const Instruction *insn, int d, const int pos)
 {
-   int r = insn->defExists(d) ? DDATA(insn->def(d)).id : 63;
-   code[pos / 32] |= r << (pos % 32);
+   if (insn->defExists(d))
+      defId(insn->def(d), pos);
+   else
+      code[pos / 32] |= 63 << (pos % 32);
 }
 
 bool CodeEmitterNVC0::isLIMM(const ValueRef& ref, DataType ty)
@@ -716,11 +718,11 @@ CodeEmitterNVC0::emitUADD(const Instruction *i)
    if (i->encSize == 8) {
       if (isLIMM(i->src(1), TYPE_U32)) {
          emitForm_A(i, HEX64(08000000, 00000002));
-         if (i->defExists(1))
+         if (i->flagsDef >= 0)
             code[1] |= 1 << 26; // write carry
       } else {
          emitForm_A(i, HEX64(48000000, 00000003));
-         if (i->defExists(1))
+         if (i->flagsDef >= 0)
             code[1] |= 1 << 16; // write carry
       }
       code[0] |= addOp;

diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
index f2843c7f0d9c44230ff2baa4846fbbf1e3b9441a..755895d2496a5fb014ad9d3b8ca904205ccbd09b 100644 (file)
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
@@ -249,7 +249,7 @@ NVC0LegalizeSSA::handleSET(CmpInstruction *cmp)
    bld.mkSplit(src0, 4, cmp->getSrc(0));
    bld.mkSplit(src1, 4, cmp->getSrc(1));
    bld.mkOp2(OP_SUB, hTy, NULL, src0[0], src1[0])
-      ->setFlagsDef(1, (carry = bld.getSSA(1, FILE_FLAGS)));
+      ->setFlagsDef(0, (carry = bld.getSSA(1, FILE_FLAGS)));
    cmp->setFlagsSrc(cmp->srcCount(), carry);
    cmp->setSrc(0, src0[1]);
    cmp->setSrc(1, src1[1]);