nvc0/ir: remove useless NVC0LoweringPass::gMemBase

[mesa.git] / src / gallium / drivers / nouveau / codegen / nv50_ir_lowering_nvc0.cpp

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 2604296d46e494bcef0368fe1547470db3e8ddee..7c9fad3442c23a3b94422cbd2da9d2ddeba968ef 100644 (file)
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
@@ -598,7 +598,6 @@ NVC0LegalizePostRA::visit(BasicBlock *bb)
 NVC0LoweringPass::NVC0LoweringPass(Program *prog) : targ(prog->getTarget())
 {
    bld.setProgram(prog);
-   gMemBase = NULL;
 }
 
 bool
@@ -751,6 +750,16 @@ NVC0LoweringPass::handleTEX(TexInstruction *i)
          i->tex.rIndirectSrc = 0;
          i->tex.sIndirectSrc = -1;
       }
+      // Move the indirect reference to right after the coords
+      else if (i->tex.rIndirectSrc >= 0 && chipset >= NVISA_GM107_CHIPSET) {
+         Value *hnd = i->getIndirectR();
+
+         i->setIndirectR(NULL);
+         i->moveSources(arg, 1);
+         i->setSrc(arg, hnd);
+         i->tex.rIndirectSrc = 0;
+         i->tex.sIndirectSrc = -1;
+      }
    } else
    // (nvc0) generate and move the tsc/tic/array source to the front
    if (i->tex.target.isArray() || i->tex.rIndirectSrc >= 0 || i->tex.sIndirectSrc >= 0) {
@@ -824,7 +833,7 @@ NVC0LoweringPass::handleTEX(TexInstruction *i)
          for (n = 0; n < i->tex.useOffsets; n++) {
             for (c = 0; c < 2; ++c) {
                if ((n % 2) == 0 && c == 0)
-                  offs[n / 2] = i->offset[n][c].get();
+                  bld.mkMov(offs[n / 2] = bld.getScratch(), i->offset[n][c].get());
                else
                   bld.mkOp3(OP_INSBF, TYPE_U32,
                             offs[n / 2],
@@ -1969,9 +1978,7 @@ NVC0LoweringPass::handleSurfaceOpNVE4(TexInstruction *su)
       }
       Instruction *red = bld.mkOp(OP_ATOM, su->dType, bld.getSSA());
       red->subOp = su->subOp;
-      if (!gMemBase)
-         gMemBase = bld.mkSymbol(FILE_MEMORY_GLOBAL, 0, TYPE_U32, 0);
-      red->setSrc(0, gMemBase);
+      red->setSrc(0, bld.mkSymbol(FILE_MEMORY_GLOBAL, 0, TYPE_U32, 0));
       red->setSrc(1, su->getSrc(3));
       if (su->subOp == NV50_IR_SUBOP_ATOM_CAS)
          red->setSrc(2, su->getSrc(4));
@@ -2695,13 +2702,30 @@ NVC0LoweringPass::visit(Instruction *i)
 
    /* Kepler+ has a special opcode to compute a new base address to be used
     * for indirect loads.
+    *
+    * Maxwell+ has an additional similar requirement for indirect
+    * interpolation ops in frag shaders.
     */
-   if (targ->getChipset() >= NVISA_GK104_CHIPSET && !i->perPatch &&
-       (i->op == OP_VFETCH || i->op == OP_EXPORT) && i->src(0).isIndirect(0)) {
+   bool doAfetch = false;
+   if (targ->getChipset() >= NVISA_GK104_CHIPSET &&
+       !i->perPatch &&
+       (i->op == OP_VFETCH || i->op == OP_EXPORT) &&
+       i->src(0).isIndirect(0)) {
+      doAfetch = true;
+   }
+   if (targ->getChipset() >= NVISA_GM107_CHIPSET &&
+       (i->op == OP_LINTERP || i->op == OP_PINTERP) &&
+       i->src(0).isIndirect(0)) {
+      doAfetch = true;
+   }
+
+   if (doAfetch) {
+      Value *addr = cloneShallow(func, i->getSrc(0));
       Instruction *afetch = bld.mkOp1(OP_AFETCH, TYPE_U32, bld.getSSA(),
-                                      cloneShallow(func, i->getSrc(0)));
+                                      i->getSrc(0));
       afetch->setIndirect(0, 0, i->getIndirect(0, 0));
-      i->src(0).get()->reg.data.offset = 0;
+      addr->reg.data.offset = 0;
+      i->setSrc(0, addr);
       i->setIndirect(0, 0, afetch->getDef(0));
    }