nv50/ir: add support for const-folding OP_CVT with F64 source/dest

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>

diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_build_util.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_build_util.cpp
index ece6ce40643ab59debb2bebf6f7cbf78d87c3899..dca799dd9b5548dcb96d2615c46f77f25b4443ea 100644 (file)
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_build_util.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_build_util.cpp
@@ -392,12 +392,24 @@ BuildUtil::mkImm(float f)
    return mkImm(u.u32);
 }
 
+ImmediateValue *
+BuildUtil::mkImm(double d)
+{
+   return new_ImmediateValue(prog, d);
+}
+
 Value *
 BuildUtil::loadImm(Value *dst, float f)
 {
    return mkOp1v(OP_MOV, TYPE_F32, dst ? dst : getScratch(), mkImm(f));
 }
 
+Value *
+BuildUtil::loadImm(Value *dst, double d)
+{
+   return mkOp1v(OP_MOV, TYPE_F64, dst ? dst : getScratch(), mkImm(d));
+}
+
 Value *
 BuildUtil::loadImm(Value *dst, uint32_t u)
 {

diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_build_util.h b/src/gallium/drivers/nouveau/codegen/nv50_ir_build_util.h
index 0d544581697f2571ec9f86d6bcb1e17fba1a1d68..8f3bf77949cebe7ddff3e48e8b3f6b459db45d6b 100644 (file)
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_build_util.h
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_build_util.h
@@ -90,12 +90,14 @@ public:
    void mkClobber(DataFile file, uint32_t regMask, int regUnitLog2);
 
    ImmediateValue *mkImm(float);
+   ImmediateValue *mkImm(double);
    ImmediateValue *mkImm(uint32_t);
    ImmediateValue *mkImm(uint64_t);
 
    ImmediateValue *mkImm(int i) { return mkImm((uint32_t)i); }
 
    Value *loadImm(Value *dst, float);
+   Value *loadImm(Value *dst, double);
    Value *loadImm(Value *dst, uint32_t);
    Value *loadImm(Value *dst, uint64_t);
 

diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
index b952c760a21c0be852ca0b4986ac75049c1689d2..f0955978dc8bb9edfadc2f0edb6556301da282fb 100644 (file)
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
@@ -1164,6 +1164,11 @@ ConstantFolding::opnd(Instruction *i, ImmediateValue &imm0, int s)
 #define CASE(type, dst, fmin, fmax, imin, imax, umin, umax) \
    case type: \
       switch (i->sType) { \
+      case TYPE_F64: \
+         res.data.dst = util_iround(i->saturate ? \
+                                    CLAMP(imm0.reg.data.f64, fmin, fmax) : \
+                                    imm0.reg.data.f64); \
+         break; \
       case TYPE_F32: \
          res.data.dst = util_iround(i->saturate ? \
                                     CLAMP(imm0.reg.data.f32, fmin, fmax) : \
@@ -1201,6 +1206,11 @@ ConstantFolding::opnd(Instruction *i, ImmediateValue &imm0, int s)
       CASE(TYPE_S32, s32, INT32_MIN, INT32_MAX, INT32_MIN, INT32_MAX, 0, INT32_MAX);
       case TYPE_F32:
          switch (i->sType) {
+         case TYPE_F64:
+            res.data.f32 = i->saturate ?
+               CLAMP(imm0.reg.data.f64, 0.0f, 1.0f) :
+               imm0.reg.data.f64;
+            break;
          case TYPE_F32:
             res.data.f32 = i->saturate ?
                CLAMP(imm0.reg.data.f32, 0.0f, 1.0f) :
@@ -1215,6 +1225,27 @@ ConstantFolding::opnd(Instruction *i, ImmediateValue &imm0, int s)
          }
          i->setSrc(0, bld.mkImm(res.data.f32));
          break;
+      case TYPE_F64:
+         switch (i->sType) {
+         case TYPE_F64:
+            res.data.f64 = i->saturate ?
+               CLAMP(imm0.reg.data.f64, 0.0f, 1.0f) :
+               imm0.reg.data.f64;
+            break;
+         case TYPE_F32:
+            res.data.f64 = i->saturate ?
+               CLAMP(imm0.reg.data.f32, 0.0f, 1.0f) :
+               imm0.reg.data.f32;
+            break;
+         case TYPE_U16: res.data.f64 = (double) imm0.reg.data.u16; break;
+         case TYPE_U32: res.data.f64 = (double) imm0.reg.data.u32; break;
+         case TYPE_S16: res.data.f64 = (double) imm0.reg.data.s16; break;
+         case TYPE_S32: res.data.f64 = (double) imm0.reg.data.s32; break;
+         default:
+            return;
+         }
+         i->setSrc(0, bld.mkImm(res.data.f64));
+         break;
       default:
          return;
       }