Mesa (master): nvc0/ir: fix up mul+add -> mad algebraic opt, enable for integers

[Ilia Mirkin]

Module: Mesa
Branch: master
Commit: f97f755192210ce3690e67abccefa133d398d373
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=f97f755192210ce3690e67abccefa133d398d373

Author: Ilia Mirkin <imirkin at alum.mit.edu>
Date:   Mon Dec  7 18:15:51 2015 -0500

nvc0/ir: fix up mul+add -> mad algebraic opt, enable for integers

For some reason this has been disabled for integers ever since codegen
was merged, despite there being emission code for IMAD. Seems to work.

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

---

 .../drivers/nouveau/codegen/nv50_ir_peephole.cpp       |   16 +++++++++++-----
 .../drivers/nouveau/codegen/nv50_ir_target_gm107.cpp   |    5 -----
 .../drivers/nouveau/codegen/nv50_ir_target_nvc0.cpp    |    2 --
 3 files changed, 11 insertions(+), 12 deletions(-)

diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
index 9f44628..2f21257 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
@@ -957,7 +957,8 @@ ConstantFolding::opnd(Instruction *i, ImmediateValue &imm0, int s)
          if (i->op != OP_CVT)
             i->src(0).mod = 0;
       } else
-      if (imm0.isInteger(1) || imm0.isInteger(-1)) {
+      if (i->subOp != NV50_IR_SUBOP_MUL_HIGH &&
+          (imm0.isInteger(1) || imm0.isInteger(-1))) {
          if (imm0.isNegative())
             i->src(t).mod = i->src(t).mod ^ Modifier(NV50_IR_MOD_NEG);
          if (s == 0) {
@@ -1589,12 +1590,11 @@ AlgebraicOpt::tryADDToMADOrSAD(Instruction *add, operation toOp)
    else
       return false;
 
-   if ((src0->getUniqueInsn() && src0->getUniqueInsn()->bb != add->bb) ||
-       (src1->getUniqueInsn() && src1->getUniqueInsn()->bb != add->bb))
-      return false;
-
    src = add->getSrc(s);
 
+   if (src->getUniqueInsn() && src->getUniqueInsn()->bb != add->bb)
+      return false;
+
    if (src->getInsn()->postFactor)
       return false;
    if (toOp == OP_SAD) {
@@ -1605,6 +1605,10 @@ AlgebraicOpt::tryADDToMADOrSAD(Instruction *add, operation toOp)
          return false;
    }
 
+   if (typeSizeof(add->dType) != typeSizeof(src->getInsn()->dType) ||
+       isFloatType(add->dType) != isFloatType(src->getInsn()->dType))
+      return false;
+
    mod[0] = add->src(0).mod;
    mod[1] = add->src(1).mod;
    mod[2] = src->getUniqueInsn()->src(0).mod;
@@ -1615,6 +1619,8 @@ AlgebraicOpt::tryADDToMADOrSAD(Instruction *add, operation toOp)
 
    add->op = toOp;
    add->subOp = src->getInsn()->subOp; // potentially mul-high
+   add->dType = src->getInsn()->dType; // sign matters for imad hi
+   add->sType = src->getInsn()->sType;
 
    add->setSrc(2, add->src(s ? 0 : 1));
 
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_target_gm107.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_target_gm107.cpp
index 202d744..92caeb2 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_target_gm107.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_target_gm107.cpp
@@ -55,11 +55,6 @@ bool
 TargetGM107::isOpSupported(operation op, DataType ty) const
 {
    switch (op) {
-   case OP_MAD:
-   case OP_FMA:
-      if (ty != TYPE_F32)
-         return false;
-      break;
    case OP_SAD:
    case OP_POW:
    case OP_SQRT:
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nvc0.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nvc0.cpp
index 0f1f4f8..19637ce 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nvc0.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nvc0.cpp
@@ -395,8 +395,6 @@ TargetNVC0::isAccessSupported(DataFile file, DataType ty) const
 bool
 TargetNVC0::isOpSupported(operation op, DataType ty) const
 {
-   if ((op == OP_MAD || op == OP_FMA) && (ty != TYPE_F32))
-      return false;
    if (op == OP_SAD && ty != TYPE_S32 && ty != TYPE_U32)
       return false;
    if (op == OP_POW || op == OP_SQRT || op == OP_DIV || op == OP_MOD)