[Mesa-dev] [PATCH] gk110/ir: make sched calculator take regs >= 64 into account

[Ilia Mirkin]

GK110/GK208 have 256 registers, not 64. Find out the number of registers
from the target (which will report 255/63 as the last reg is hard-wired
to 0).

Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu>
---
 .../drivers/nouveau/codegen/nv50_ir_emit_nvc0.cpp       | 17 ++++++++++-------
 1 file changed, 10 insertions(+), 7 deletions(-)

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 be6fe95..1d99101 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_nvc0.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_nvc0.cpp
@@ -2588,11 +2588,12 @@ private:
          int imul; // integer MUL to MUL delay 3
       } res;
       struct ScoreData {
-         int r[64];
+         int r[256];
          int p[8];
          int c;
       } rd, wr;
       int base;
+      int regs;
 
       void rebase(const int base)
       {
@@ -2601,7 +2602,7 @@ private:
             return;
          this->base = 0;
 
-         for (int i = 0; i < 64; ++i) {
+         for (int i = 0; i <= regs; ++i) {
             rd.r[i] += delta;
             wr.r[i] += delta;
          }
@@ -2620,16 +2621,17 @@ private:
          res.imul += delta;
          res.tex += delta;
       }
-      void wipe()
+      void wipe(int regs)
       {
          memset(&rd, 0, sizeof(rd));
          memset(&wr, 0, sizeof(wr));
          memset(&res, 0, sizeof(res));
+         this->regs = regs;
       }
       int getLatest(const ScoreData& d) const
       {
          int max = 0;
-         for (int i = 0; i < 64; ++i)
+         for (int i = 0; i <= regs; ++i)
             if (d.r[i] > max)
                max = d.r[i];
          for (int i = 0; i < 8; ++i)
@@ -2664,7 +2666,7 @@ private:
       }
       void setMax(const RegScores *that)
       {
-         for (int i = 0; i < 64; ++i) {
+         for (int i = 0; i <= regs; ++i) {
             rd.r[i] = MAX2(rd.r[i], that->rd.r[i]);
             wr.r[i] = MAX2(wr.r[i], that->wr.r[i]);
          }
@@ -2685,7 +2687,7 @@ private:
       }
       void print(int cycle)
       {
-         for (int i = 0; i < 64; ++i) {
+         for (int i = 0; i <= regs; ++i) {
             if (rd.r[i] > cycle)
                INFO("rd $r%i @ %i\n", i, rd.r[i]);
             if (wr.r[i] > cycle)
@@ -2780,9 +2782,10 @@ SchedDataCalculator::getCycles(const Instruction *insn, int origDelay) const
 bool
 SchedDataCalculator::visit(Function *func)
 {
+   int regs = targ->getFileSize(FILE_GPR);
    scoreBoards.resize(func->cfg.getSize());
    for (size_t i = 0; i < scoreBoards.size(); ++i)
-      scoreBoards[i].wipe();
+      scoreBoards[i].wipe(regs);
    return true;
 }
 
-- 
2.3.6