nvir/gv100: enable support for tu1xx

[mesa.git] / src / gallium / drivers / nouveau / codegen / nv50_ir_ra.cpp

diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_ra.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_ra.cpp
index f4379c137c5d697ecb52164bf8edd27a39bf7a0a..4e5b21d917642a6ad2a36f37704dab5b6ff24a8b 100644 (file)
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_ra.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_ra.cpp
@@ -624,8 +624,6 @@ RegAlloc::BuildIntervalsPass::collectLiveValues(BasicBlock *bb)
       // trickery to save a loop of OR'ing liveSets
       // aliasing works fine with BitSet::setOr
       for (Graph::EdgeIterator ei = bb->cfg.outgoing(); !ei.end(); ei.next()) {
-         if (ei.getType() == Graph::Edge::DUMMY)
-            continue;
          if (bbA) {
             bb->liveSet.setOr(&bbA->liveSet, &bbB->liveSet);
             bbA = bb;
@@ -990,6 +988,8 @@ GCRA::coalesce(ArrayList& insns)
    case 0x110:
    case 0x120:
    case 0x130:
+   case 0x140:
+   case 0x160:
       ret = doCoalesce(insns, JOIN_MASK_UNION);
       break;
    default:
@@ -2299,13 +2299,25 @@ RegAlloc::InsertConstraintsPass::texConstraintGM107(TexInstruction *tex)
    if (isTextureOp(tex->op))
       textureMask(tex);
 
-   if (isScalarTexGM107(tex)) {
-      handleScalarTexGM107(tex);
-      return;
-   }
+   if (targ->getChipset() < NVISA_GV100_CHIPSET) {
+      if (isScalarTexGM107(tex)) {
+         handleScalarTexGM107(tex);
+         return;
+      }
 
-   assert(!tex->tex.scalar);
-   condenseDefs(tex);
+      assert(!tex->tex.scalar);
+      condenseDefs(tex);
+   } else {
+      if (isTextureOp(tex->op)) {
+         int defCount = tex->defCount(0xff);
+         if (defCount > 3)
+            condenseDefs(tex, 2, 3);
+         if (defCount > 1)
+            condenseDefs(tex, 0, 1);
+      } else {
+         condenseDefs(tex);
+      }
+   }
 
    if (isSurfaceOp(tex->op)) {
       int s = tex->tex.target.getDim() +
@@ -2341,9 +2353,19 @@ RegAlloc::InsertConstraintsPass::texConstraintGM107(TexInstruction *tex)
             if (!tex->tex.target.isArray() && tex->tex.useOffsets)
                s++;
          }
-         n = tex->srcCount(0xff) - s;
+         n = tex->srcCount(0xff, true) - s;
+         // TODO: Is this necessary? Perhaps just has to be aligned to the
+         // level that the first arg is, not necessarily to 4. This
+         // requirement has not been rigorously verified, as it has been on
+         // Kepler.
+         if (n > 0 && n < 3) {
+            if (tex->srcExists(n + s)) // move potential predicate out of the way
+               tex->moveSources(n + s, 3 - n);
+            while (n < 3)
+               tex->setSrc(s + n++, new_LValue(func, FILE_GPR));
+         }
       } else {
-         s = tex->srcCount(0xff);
+         s = tex->srcCount(0xff, true);
          n = 0;
       }
 
@@ -2366,14 +2388,18 @@ RegAlloc::InsertConstraintsPass::texConstraintNVE0(TexInstruction *tex)
    } else
    if (isTextureOp(tex->op)) {
       int n = tex->srcCount(0xff, true);
-      if (n > 4) {
-         condenseSrcs(tex, 0, 3);
-         if (n > 5) // NOTE: first call modified positions already
-            condenseSrcs(tex, 4 - (4 - 1), n - 1 - (4 - 1));
-      } else
-      if (n > 1) {
-         condenseSrcs(tex, 0, n - 1);
+      int s = n > 4 ? 4 : n;
+      if (n > 4 && n < 7) {
+         if (tex->srcExists(n)) // move potential predicate out of the way
+            tex->moveSources(n, 7 - n);
+
+         while (n < 7)
+            tex->setSrc(n++, new_LValue(func, FILE_GPR));
       }
+      if (s > 1)
+         condenseSrcs(tex, 0, s - 1);
+      if (n > 4)
+         condenseSrcs(tex, 1, n - s);
    }
 }
 
@@ -2473,6 +2499,8 @@ RegAlloc::InsertConstraintsPass::visit(BasicBlock *bb)
          case 0x110:
          case 0x120:
          case 0x130:
+         case 0x140:
+         case 0x160:
             texConstraintGM107(tex);
             break;
          default:
@@ -2510,6 +2538,7 @@ RegAlloc::InsertConstraintsPass::insertConstraintMove(Instruction *cst, int s)
    assert(cst->getSrc(s)->defs.size() == 1); // still SSA
 
    Instruction *defi = cst->getSrc(s)->defs.front()->getInsn();
+
    bool imm = defi->op == OP_MOV &&
       defi->src(0).getFile() == FILE_IMMEDIATE;
    bool load = defi->op == OP_LOAD &&