nvc0/ir: Properly handle a "split form" of predicate destination

[mesa.git] / src / gallium / drivers / nouveau / codegen / nv50_ir_lowering_nv50.cpp

diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nv50.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nv50.cpp
index 12c5f6996037c5c19dd3e4f26ba5e92fc4a5e79a..36ab837f6e2592bc4ae6613930eaacf93d73b1bf 100644 (file)
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nv50.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nv50.cpp
@@ -372,7 +372,8 @@ NV50LegalizeSSA::propagateWriteToOutput(Instruction *st)
       return;
 
    for (int s = 0; di->srcExists(s); ++s)
-      if (di->src(s).getFile() == FILE_IMMEDIATE)
+      if (di->src(s).getFile() == FILE_IMMEDIATE ||
+          di->src(s).getFile() == FILE_MEMORY_LOCAL)
          return;
 
    if (prog->getType() == Program::TYPE_GEOMETRY) {
@@ -682,7 +683,7 @@ void NV50LoweringPreSSA::loadTexMsInfo(uint32_t off, Value **ms,
                                        Value **ms_x, Value **ms_y) {
    // This loads the texture-indexed ms setting from the constant buffer
    Value *tmp = new_LValue(func, FILE_GPR);
-   uint8_t b = prog->driver->io.resInfoCBSlot;
+   uint8_t b = prog->driver->io.auxCBSlot;
    off += prog->driver->io.suInfoBase;
    if (prog->getType() > Program::TYPE_VERTEX)
       off += 16 * 2 * 4;
@@ -724,6 +725,23 @@ NV50LoweringPreSSA::handleTEX(TexInstruction *i)
    const int dref = arg;
    const int lod = i->tex.target.isShadow() ? (arg + 1) : arg;
 
+   /* Only normalize in the non-explicit derivatives case.
+    */
+   if (i->tex.target.isCube() && i->op != OP_TXD) {
+      Value *src[3], *val;
+      int c;
+      for (c = 0; c < 3; ++c)
+         src[c] = bld.mkOp1v(OP_ABS, TYPE_F32, bld.getSSA(), i->getSrc(c));
+      val = bld.getScratch();
+      bld.mkOp2(OP_MAX, TYPE_F32, val, src[0], src[1]);
+      bld.mkOp2(OP_MAX, TYPE_F32, val, src[2], val);
+      bld.mkOp1(OP_RCP, TYPE_F32, val, val);
+      for (c = 0; c < 3; ++c) {
+         i->setSrc(c, bld.mkOp2v(OP_MUL, TYPE_F32, bld.getSSA(),
+                                 i->getSrc(c), val));
+      }
+   }
+
    // handle MS, which means looking up the MS params for this texture, and
    // adjusting the input coordinates to point at the right sample.
    if (i->tex.target.isMS()) {
@@ -934,12 +952,14 @@ NV50LoweringPreSSA::handleTXD(TexInstruction *i)
 
    handleTEX(i);
    i->op = OP_TEX; // no need to clone dPdx/dPdy later
+   i->tex.derivAll = true;
 
    for (c = 0; c < dim; ++c)
       crd[c] = bld.getScratch();
 
    bld.mkOp(OP_QUADON, TYPE_NONE, NULL);
    for (l = 0; l < 4; ++l) {
+      Value *src[3], *val;
       // mov coordinates from lane l to all lanes
       for (c = 0; c < dim; ++c)
          bld.mkQuadop(0x00, crd[c], l, i->getSrc(c), zero);
@@ -949,10 +969,24 @@ NV50LoweringPreSSA::handleTXD(TexInstruction *i)
       // add dPdy from lane l to lanes dy
       for (c = 0; c < dim; ++c)
          bld.mkQuadop(qOps[l][1], crd[c], l, i->dPdy[c].get(), crd[c]);
+      // normalize cube coordinates if necessary
+      if (i->tex.target.isCube()) {
+         for (c = 0; c < 3; ++c)
+            src[c] = bld.mkOp1v(OP_ABS, TYPE_F32, bld.getSSA(), crd[c]);
+         val = bld.getScratch();
+         bld.mkOp2(OP_MAX, TYPE_F32, val, src[0], src[1]);
+         bld.mkOp2(OP_MAX, TYPE_F32, val, src[2], val);
+         bld.mkOp1(OP_RCP, TYPE_F32, val, val);
+         for (c = 0; c < 3; ++c)
+            src[c] = bld.mkOp2v(OP_MUL, TYPE_F32, bld.getSSA(), crd[c], val);
+      } else {
+         for (c = 0; c < dim; ++c)
+            src[c] = crd[c];
+      }
       // texture
       bld.insert(tex = cloneForward(func, i));
       for (c = 0; c < dim; ++c)
-         tex->setSrc(c, crd[c]);
+         tex->setSrc(c, src[c]);
       // save results
       for (c = 0; i->defExists(c); ++c) {
          Instruction *mov;
@@ -1174,7 +1208,7 @@ NV50LoweringPreSSA::handleRDSV(Instruction *i)
       bld.mkLoad(TYPE_F32,
                  def,
                  bld.mkSymbol(
-                       FILE_MEMORY_CONST, prog->driver->io.resInfoCBSlot,
+                       FILE_MEMORY_CONST, prog->driver->io.auxCBSlot,
                        TYPE_U32, prog->driver->io.sampleInfoBase + 4 * idx),
                  off);
       break;