gpu-compute, hsail: pass GPUDynInstPtr to getRegisterIndex()

for HSAIL an operand's indices into the register files may be calculated
trivially, because the operands are always read from a register file, or are
an immediate.

for machine ISA, however, an op selector may specify special registers, or
may specify special SGPRs with an alias op selector value. the location of
some of the special registers values are dependent on the size of the RF
in some cases. here we add a way for the underlying getRegisterIndex()
method to know about the size of the RFs, so that it may find the relative
positions of the special register values.

src/arch/hsail/insts/branch.hh

@@ -95,7 +95,9 @@ namespace HsailISA
             return target.opSize();
         }
 
-        int getRegisterIndex(int operandIndex) override {
+        int
+        getRegisterIndex(int operandIndex, GPUDynInstPtr gpuDynInst) override
+        {
             assert(operandIndex >= 0 && operandIndex < getNumOperands());
             return target.regIndex();
         }
@@ -223,7 +225,9 @@ namespace HsailISA
             else
                 return 1;
         }
-        int getRegisterIndex(int operandIndex) override {
+        int
+        getRegisterIndex(int operandIndex, GPUDynInstPtr gpuDynInst) override
+        {
             assert(operandIndex >= 0 && operandIndex < getNumOperands());
             if (!operandIndex)
                 return target.regIndex();
@@ -370,7 +374,9 @@ namespace HsailISA
             assert(operandIndex >= 0 && operandIndex < getNumOperands());
             return target.opSize();
         }
-        int getRegisterIndex(int operandIndex) override {
+        int
+        getRegisterIndex(int operandIndex, GPUDynInstPtr gpuDynInst) override
+        {
             assert(operandIndex >= 0 && operandIndex < getNumOperands());
             return target.regIndex();
         }

src/arch/hsail/insts/decl.hh

@@ -178,7 +178,9 @@ namespace HsailISA
             else
                 return dest.opSize();
         }
-        int getRegisterIndex(int operandIndex) {
+        int
+        getRegisterIndex(int operandIndex, GPUDynInstPtr gpuDynInst)
+        {
             assert(operandIndex >= 0 && operandIndex < getNumOperands());
 
             if (operandIndex < NumSrcOperands)
@@ -313,7 +315,10 @@ namespace HsailISA
             else
                 return dest.opSize();
         }
-        int getRegisterIndex(int operandIndex) {
+
+        int
+        getRegisterIndex(int operandIndex, GPUDynInstPtr gpuDynInst)
+        {
             assert((operandIndex >= 0) && (operandIndex < getNumOperands()));
             if (!operandIndex)
                 return src0.regIndex();
@@ -477,7 +482,10 @@ namespace HsailISA
             else
                 return dest.opSize();
         }
-        int getRegisterIndex(int operandIndex) {
+
+        int
+        getRegisterIndex(int operandIndex, GPUDynInstPtr gpuDynInst)
+        {
             assert((operandIndex >= 0) && (operandIndex < getNumOperands()));
             if (!operandIndex)
                 return src0.regIndex();
@@ -643,7 +651,7 @@ namespace HsailISA
             return -1;
 
         //handle positive and negative numbers
-        T tmp = (src0 < 0) ? (~src0) : (src0);
+        T tmp = ((int64_t)src0 < 0) ? (~src0) : (src0);
 
         //the starting pos is MSB
         int pos = 8 * sizeof(T) - 1;
@@ -732,7 +740,12 @@ namespace HsailISA
         bool isSrcOperand(int operandIndex) { return false; }
         bool isDstOperand(int operandIndex) { return false; }
         int getOperandSize(int operandIndex) { return 0; }
-        int getRegisterIndex(int operandIndex) { return -1; }
+
+        int
+        getRegisterIndex(int operandIndex, GPUDynInstPtr gpuDynInst)
+        {
+            return -1;
+        }
 
         int numSrcRegOperands() { return 0; }
         int numDstRegOperands() { return 0; }
@@ -777,10 +790,14 @@ namespace HsailISA
             assert((operandIndex >= 0) && (operandIndex < getNumOperands()));
             return dest.opSize();
         }
-        int getRegisterIndex(int operandIndex) {
+
+        int
+        getRegisterIndex(int operandIndex, GPUDynInstPtr gpuDynInst)
+        {
             assert((operandIndex >= 0) && (operandIndex < getNumOperands()));
             return dest.regIndex();
         }
+
         int numSrcRegOperands() { return 0; }
         int numDstRegOperands() { return dest.isVectorRegister(); }
         int getNumOperands() { return 1; }
@@ -848,10 +865,14 @@ namespace HsailISA
             assert((operandIndex >= 0) && (operandIndex < getNumOperands()));
             return dest.opSize();
         }
-        int getRegisterIndex(int operandIndex) {
+
+        int
+        getRegisterIndex(int operandIndex, GPUDynInstPtr gpuDynInst)
+        {
             assert((operandIndex >= 0) && (operandIndex < getNumOperands()));
             return dest.regIndex();
         }
+
         int numSrcRegOperands() { return 0; }
         int numDstRegOperands() { return dest.isVectorRegister(); }
         int getNumOperands() { return 1; }
@@ -1171,8 +1192,13 @@ namespace HsailISA
         bool isScalarRegister(int operandIndex) { return false; }
         bool isSrcOperand(int operandIndex) { return false; }
         bool isDstOperand(int operandIndex) { return false; }
-        int  getOperandSize(int operandIndex) { return 0; }
+        int getOperandSize(int operandIndex) { return 0; }
-        int  getRegisterIndex(int operandIndex) { return -1; }
+
+        int
+        getRegisterIndex(int operandIndex, GPUDynInstPtr gpuDynInst)
+        {
+            return -1;
+        }
 
         void
         execute(GPUDynInstPtr gpuDynInst)

src/arch/hsail/insts/mem.hh

@@ -146,7 +146,8 @@ namespace HsailISA
             return((operandIndex == 0) ? dest.opSize() :
                    this->addr.opSize());
         }
-        int getRegisterIndex(int operandIndex) override
+        int
+        getRegisterIndex(int operandIndex, GPUDynInstPtr gpuDynInst) override
         {
             assert((operandIndex >= 0) && (operandIndex < getNumOperands()));
             return((operandIndex == 0) ? dest.regIndex() :
@@ -377,7 +378,8 @@ namespace HsailISA
             return((operandIndex == 0) ? dest.opSize() :
                    this->addr.opSize());
         }
-        int getRegisterIndex(int operandIndex) override
+        int
+        getRegisterIndex(int operandIndex, GPUDynInstPtr gpuDynInst) override
         {
             assert((operandIndex >= 0) && (operandIndex < getNumOperands()));
             return((operandIndex == 0) ? dest.regIndex() :
@@ -670,7 +672,8 @@ namespace HsailISA
                        AddrOperandType>::dest.opSize());
             return 0;
         }
-        int getRegisterIndex(int operandIndex) override
+        int
+        getRegisterIndex(int operandIndex, GPUDynInstPtr gpuDynInst) override
         {
             assert((operandIndex >= 0) && (operandIndex < getNumOperands()));
             if ((num_dest_operands != getNumOperands()) &&
@@ -934,7 +937,8 @@ namespace HsailISA
             assert(operandIndex >= 0 && operandIndex < getNumOperands());
             return !operandIndex ? src.opSize() : this->addr.opSize();
         }
-        int getRegisterIndex(int operandIndex) override
+        int
+        getRegisterIndex(int operandIndex, GPUDynInstPtr gpuDynInst) override
         {
             assert(operandIndex >= 0 && operandIndex < getNumOperands());
             return !operandIndex ? src.regIndex() : this->addr.regIndex();
@@ -1144,7 +1148,8 @@ namespace HsailISA
                        AddrOperandType>::src.opSize();
             return 0;
         }
-        int getRegisterIndex(int operandIndex) override
+        int
+        getRegisterIndex(int operandIndex, GPUDynInstPtr gpuDynInst) override
         {
             assert((operandIndex >= 0) && (operandIndex < getNumOperands()));
             if (operandIndex == num_src_operands)
@@ -1433,7 +1438,8 @@ namespace HsailISA
             else
                 return(dest.opSize());
         }
-        int getRegisterIndex(int operandIndex)
+        int
+        getRegisterIndex(int operandIndex, GPUDynInstPtr gpuDynInst)
         {
             assert((operandIndex >= 0) && (operandIndex < getNumOperands()));
             if (operandIndex < NumSrcOperands)

src/gpu-compute/condition_register_state.cc

@@ -62,19 +62,19 @@ ConditionRegisterState::init(uint32_t _size)
 }
 
 void
-ConditionRegisterState::exec(GPUStaticInst *ii, Wavefront *w)
+ConditionRegisterState::exec(GPUDynInstPtr ii, Wavefront *w)
 {
     // iterate over all operands
     for (auto i = 0; i < ii->getNumOperands(); ++i) {
         // is this a condition register destination operand?
         if (ii->isCondRegister(i) && ii->isDstOperand(i)) {
             // mark the register as busy
-            markReg(ii->getRegisterIndex(i), 1);
+            markReg(ii->getRegisterIndex(i, ii), 1);
             uint32_t pipeLen =  w->computeUnit->spBypassLength();
 
             // schedule an event for marking the register as ready
             w->computeUnit->
-                registerEvent(w->simdId, ii->getRegisterIndex(i),
+                registerEvent(w->simdId, ii->getRegisterIndex(i, ii),
                               ii->getOperandSize(i),
                               w->computeUnit->shader->tick_cnt +
                               w->computeUnit->shader->ticks(pipeLen), 0);

src/gpu-compute/condition_register_state.hh

@@ -87,7 +87,7 @@ class ConditionRegisterState
     }
 
     int numRegs() { return c_reg.size(); }
-    void exec(GPUStaticInst *ii, Wavefront *w);
+    void exec(GPUDynInstPtr ii, Wavefront *w);
 
   private:
     ComputeUnit* computeUnit;

src/gpu-compute/gpu_dyn_inst.cc

@@ -102,10 +102,16 @@ GPUDynInst::isScalarRegister(int operandIdx)
     return _staticInst->isScalarRegister(operandIdx);
 }
 
-int
+bool
-GPUDynInst::getRegisterIndex(int operandIdx)
+GPUDynInst::isCondRegister(int operandIdx)
 {
-    return _staticInst->getRegisterIndex(operandIdx);
+    return _staticInst->isCondRegister(operandIdx);
+}
+
+int
+GPUDynInst::getRegisterIndex(int operandIdx, GPUDynInstPtr gpuDynInst)
+{
+    return _staticInst->getRegisterIndex(operandIdx, gpuDynInst);
 }
 
 int

src/gpu-compute/gpu_dyn_inst.hh

@@ -194,7 +194,8 @@ class GPUDynInst : public GPUExecContext
     int getNumOperands();
     bool isVectorRegister(int operandIdx);
     bool isScalarRegister(int operandIdx);
-    int getRegisterIndex(int operandIdx);
+    bool isCondRegister(int operandIdx);
+    int getRegisterIndex(int operandIdx, GPUDynInstPtr gpuDynInst);
     int getOperandSize(int operandIdx);
     bool isDstOperand(int operandIdx);
     bool isSrcOperand(int operandIdx);

src/gpu-compute/gpu_static_inst.hh

@@ -83,7 +83,10 @@ class GPUStaticInst : public GPUStaticInstFlags
     virtual bool isSrcOperand(int operandIndex) = 0;
     virtual bool isDstOperand(int operandIndex) = 0;
     virtual int getOperandSize(int operandIndex) = 0;
-    virtual int getRegisterIndex(int operandIndex) = 0;
+
+    virtual int getRegisterIndex(int operandIndex,
+                                 GPUDynInstPtr gpuDynInst) = 0;
+
     virtual int numDstRegOperands() = 0;
     virtual int numSrcRegOperands() = 0;
 
@@ -286,7 +289,13 @@ class KernelLaunchStaticInst : public GPUStaticInst
     bool isSrcOperand(int operandIndex) { return false; }
     bool isDstOperand(int operandIndex) { return false; }
     int getOperandSize(int operandIndex) { return 0; }
-    int getRegisterIndex(int operandIndex) { return 0; }
+
+    int
+    getRegisterIndex(int operandIndex, GPUDynInstPtr gpuDynInst) override
+    {
+        return 0;
+    }
+
     int numDstRegOperands() { return 0; }
     int numSrcRegOperands() { return 0; }
     bool isValid() const { return true; }

src/gpu-compute/vector_register_file.cc

@@ -121,7 +121,7 @@ VectorRegisterFile::operandsReady(Wavefront *w, GPUDynInstPtr ii) const
 {
     for (int i = 0; i < ii->getNumOperands(); ++i) {
         if (ii->isVectorRegister(i)) {
-            uint32_t vgprIdx = ii->getRegisterIndex(i);
+            uint32_t vgprIdx = ii->getRegisterIndex(i, ii);
             uint32_t pVgpr = w->remap(vgprIdx, ii->getOperandSize(i), 1);
 
             if (regBusy(pVgpr, ii->getOperandSize(i)) == 1) {
@@ -160,7 +160,7 @@ VectorRegisterFile::exec(GPUDynInstPtr ii, Wavefront *w)
     // iterate over all register destination operands
     for (int i = 0; i < ii->getNumOperands(); ++i) {
         if (ii->isVectorRegister(i) && ii->isDstOperand(i)) {
-            uint32_t physReg = w->remap(ii->getRegisterIndex(i),
+            uint32_t physReg = w->remap(ii->getRegisterIndex(i, ii),
                                         ii->getOperandSize(i), 1);
 
             // mark the destination vector register as busy
@@ -216,7 +216,7 @@ VectorRegisterFile::updateResources(Wavefront *w, GPUDynInstPtr ii)
     // iterate over all register destination operands
     for (int i = 0; i < ii->getNumOperands(); ++i) {
         if (ii->isVectorRegister(i) && ii->isDstOperand(i)) {
-            uint32_t physReg = w->remap(ii->getRegisterIndex(i),
+            uint32_t physReg = w->remap(ii->getRegisterIndex(i, ii),
                                         ii->getOperandSize(i), 1);
             // set the in-flight status of the destination vector register
             preMarkReg(physReg, ii->getOperandSize(i), 1);