gv100/ir: add support for barrier thread state files for OP_CVT

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11061>

diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gv100.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gv100.cpp
index f18c116..da8f71c 100644 (file)
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gv100.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gv100.cpp
@@ -313,6 +313,12 @@ CodeEmitterGV100::emitMOV()
          emitField(90,  1, 1);
          emitPRED (87, insn->src(0));
          break;
+      case FILE_BARRIER:
+      case FILE_THREAD_STATE:
+         emitInsn (0x355);
+         emitBTS  (24, insn->src(0));
+         emitGPR  (16, insn->def(0));
+         break;
       default:
          assert(!"bad src file");
          break;
@@ -329,6 +335,31 @@ CodeEmitterGV100::emitMOV()
       emitGPR  (24, insn->src(0));
       emitGPR  (32);
       break;
+   case FILE_BARRIER:
+   case FILE_THREAD_STATE:
+      switch (insn->src(0).getFile()) {
+      case FILE_GPR:
+         emitInsn (0x356);
+         emitGPR  (32, insn->src(0));
+         emitBTS  (24, insn->def(0));
+         break;
+      case FILE_BARRIER:
+         emitInsn (0xf56);
+         emitBTS  (24, insn->def(0));
+         emitBTS  (16, insn->src(0));
+         break;
+      case FILE_THREAD_STATE:
+         assert(insn->def(0).getFile() == FILE_BARRIER);
+         emitInsn (0xf55);
+         emitBTS  (24, insn->src(0));
+         emitBTS  (16, insn->def(0));
+         break;
+      default:
+         assert(!"bad src file");
+         break;
+      }
+      emitField(84, 1, insn->getDef(0)->reg.data.ts == TS_PQUAD_MACTIVE ? 1 : 0);
+      break;
    default:
       assert(!"bad dst file");
       break;
@@ -1753,7 +1784,11 @@ CodeEmitterGV100::emitInstruction(Instruction *i)
    case OP_FLOOR:
    case OP_TRUNC:
       if (insn->op == OP_CVT && (insn->def(0).getFile() == FILE_PREDICATE ||
-                                 insn->src(0).getFile() == FILE_PREDICATE)) {
+                                 insn->def(0).getFile() == FILE_BARRIER ||
+                                 insn->def(0).getFile() == FILE_THREAD_STATE ||
+                                 insn->src(0).getFile() == FILE_PREDICATE ||
+                                 insn->src(0).getFile() == FILE_BARRIER ||
+                                 insn->src(0).getFile() == FILE_THREAD_STATE)) {
          emitMOV();
       } else if (isFloatType(insn->dType)) {
          if (isFloatType(insn->sType)) {

diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gv100.h b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gv100.h
index 15ab717..d6a903e 100644 (file)
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gv100.h
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gv100.h
@@ -228,6 +228,23 @@ private:
       emitSYS(pos, ref.get() ? ref.rep() : (const Value *)NULL);
    }
 
+   inline void emitBTS(int pos, const Value *val) {
+      if (val->inFile(FILE_THREAD_STATE)) {
+         TSSemantic ts = val->reg.data.ts == TS_PQUAD_MACTIVE ? TS_MACTIVE : val->reg.data.ts;
+         emitField(pos, 5, ts | 0x10);
+      } else {
+         emitField(pos, 5, val->reg.data.id);
+      }
+   }
+
+   inline void emitBTS(int pos, const ValueRef &ref) {
+      emitBTS(pos, ref.get() ? ref.rep() : (const Value *)NULL);
+   }
+
+   inline void emitBTS(int pos, const ValueDef &def) {
+      emitBTS(pos, def.get() ? def.rep() : (const Value *)NULL);
+   }
+
    inline void emitGPR(int pos, const Value *val, int off) {
       emitField(pos, 8, val && !val->inFile(FILE_FLAGS) ?
                 val->reg.data.id + off: 255);