From 76957389fc6952e59c1f0f1cbdf74f6949a7a956 Mon Sep 17 00:00:00 2001
From: Ilia Mirkin <imirkin@alum.mit.edu>
Date: Sun, 22 Feb 2015 19:49:49 -0500
Subject: [PATCH] nv50/ir: add fp64 opcode emission support for G200 (NVA0)

Need to emulate rcp/rsq before providing full fp64 support

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
---
 .../nouveau/codegen/nv50_ir_emit_nv50.cpp     | 94 +++++++++++++++++--
 1 file changed, 84 insertions(+), 10 deletions(-)

diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_nv50.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_nv50.cpp
index 9f1e4b803d5..ee115b581b8 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_nv50.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_nv50.cpp
@@ -96,9 +96,12 @@ private:
    void emitUADD(const Instruction *);
    void emitAADD(const Instruction *);
    void emitFADD(const Instruction *);
+   void emitDADD(const Instruction *);
    void emitIMUL(const Instruction *);
    void emitFMUL(const Instruction *);
+   void emitDMUL(const Instruction *);
    void emitFMAD(const Instruction *);
+   void emitDMAD(const Instruction *);
    void emitIMAD(const Instruction *);
    void emitISAD(const Instruction *);
 
@@ -954,11 +957,13 @@ CodeEmitterNV50::emitMINMAX(const Instruction *i)
          assert(0);
          break;
       }
-      code[1] |= i->src(0).mod.abs() << 20;
-      code[1] |= i->src(0).mod.neg() << 26;
-      code[1] |= i->src(1).mod.abs() << 19;
-      code[1] |= i->src(1).mod.neg() << 27;
    }
+
+   code[1] |= i->src(0).mod.abs() << 20;
+   code[1] |= i->src(0).mod.neg() << 26;
+   code[1] |= i->src(1).mod.abs() << 19;
+   code[1] |= i->src(1).mod.neg() << 27;
+
    emitForm_MAD(i);
 }
 
@@ -993,6 +998,26 @@ CodeEmitterNV50::emitFMAD(const Instruction *i)
    }
 }
 
+void
+CodeEmitterNV50::emitDMAD(const Instruction *i)
+{
+   const int neg_mul = i->src(0).mod.neg() ^ i->src(1).mod.neg();
+   const int neg_add = i->src(2).mod.neg();
+
+   assert(i->encSize == 8);
+   assert(!i->saturate);
+
+   code[1] = 0x40000000;
+   code[0] = 0xe0000000;
+
+   code[1] |= neg_mul << 26;
+   code[1] |= neg_add << 27;
+
+   roundMode_MAD(i);
+
+   emitForm_MAD(i);
+}
+
 void
 CodeEmitterNV50::emitFADD(const Instruction *i)
 {
@@ -1027,6 +1052,25 @@ CodeEmitterNV50::emitFADD(const Instruction *i)
    }
 }
 
+void
+CodeEmitterNV50::emitDADD(const Instruction *i)
+{
+   const int neg0 = i->src(0).mod.neg();
+   const int neg1 = i->src(1).mod.neg() ^ ((i->op == OP_SUB) ? 1 : 0);
+
+   assert(!(i->src(0).mod | i->src(1).mod).abs());
+   assert(!i->saturate);
+   assert(i->encSize == 8);
+
+   code[1] = 0x60000000;
+   code[0] = 0xe0000000;
+
+   emitForm_ADD(i);
+
+   code[1] |= neg0 << 26;
+   code[1] |= neg1 << 27;
+}
+
 void
 CodeEmitterNV50::emitUADD(const Instruction *i)
 {
@@ -1120,6 +1164,25 @@ CodeEmitterNV50::emitFMUL(const Instruction *i)
    }
 }
 
+void
+CodeEmitterNV50::emitDMUL(const Instruction *i)
+{
+   const int neg = (i->src(0).mod ^ i->src(1).mod).neg();
+
+   assert(!i->saturate);
+   assert(i->encSize == 8);
+
+   code[1] = 0x80000000;
+   code[0] = 0xe0000000;
+
+   if (neg)
+      code[1] |= 0x08000000;
+
+   roundMode_CVT(i->rnd);
+
+   emitForm_MAD(i);
+}
+
 void
 CodeEmitterNV50::emitIMAD(const Instruction *i)
 {
@@ -1181,9 +1244,11 @@ CodeEmitterNV50::emitSET(const Instruction *i)
    code[0] = 0x30000000;
    code[1] = 0x60000000;
 
-   emitCondCode(i->asCmp()->setCond, i->sType, 32 + 14);
-
    switch (i->sType) {
+   case TYPE_F64:
+      code[0] = 0xe0000000;
+      code[1] = 0xe0000000;
+      break;
    case TYPE_F32: code[0] |= 0x80000000; break;
    case TYPE_S32: code[1] |= 0x0c000000; break;
    case TYPE_U32: code[1] |= 0x04000000; break;
@@ -1193,6 +1258,9 @@ CodeEmitterNV50::emitSET(const Instruction *i)
       assert(0);
       break;
    }
+
+   emitCondCode(i->asCmp()->setCond, i->sType, 32 + 14);
+
    if (i->src(0).mod.neg()) code[1] |= 0x04000000;
    if (i->src(1).mod.neg()) code[1] |= 0x08000000;
    if (i->src(0).mod.abs()) code[1] |= 0x00100000;
@@ -1756,7 +1824,9 @@ CodeEmitterNV50::emitInstruction(Instruction *insn)
       break;
    case OP_ADD:
    case OP_SUB:
-      if (isFloatType(insn->dType))
+      if (insn->dType == TYPE_F64)
+         emitDADD(insn);
+      else if (isFloatType(insn->dType))
          emitFADD(insn);
       else if (insn->getDef(0)->reg.file == FILE_ADDRESS)
          emitAADD(insn);
@@ -1764,14 +1834,18 @@ CodeEmitterNV50::emitInstruction(Instruction *insn)
          emitUADD(insn);
       break;
    case OP_MUL:
-      if (isFloatType(insn->dType))
+      if (insn->dType == TYPE_F64)
+         emitDMUL(insn);
+      else if (isFloatType(insn->dType))
          emitFMUL(insn);
       else
          emitIMUL(insn);
       break;
    case OP_MAD:
    case OP_FMA:
-      if (isFloatType(insn->dType))
+      if (insn->dType == TYPE_F64)
+         emitDMAD(insn);
+      else if (isFloatType(insn->dType))
          emitFMAD(insn);
       else
          emitIMAD(insn);
@@ -1943,7 +2017,7 @@ CodeEmitterNV50::getMinEncodingSize(const Instruction *i) const
 {
    const Target::OpInfo &info = targ->getOpInfo(i);
 
-   if (info.minEncSize > 4)
+   if (info.minEncSize > 4 || i->dType == TYPE_F64)
       return 8;
 
    // check constraints on dst and src operands
-- 
2.30.2