[Mesa-dev] [PATCH] nv50/ir: Improve performance of signed division by powers of two

[Rhys Perry]

Signed-off-by: Rhys Perry <pendingchaos02 at gmail.com>
---
 .../drivers/nouveau/codegen/nv50_ir_peephole.cpp   | 29 +++++++++++++++++++---
 1 file changed, 25 insertions(+), 4 deletions(-)

diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
index 39177bd044..7a18a5fe73 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
@@ -1095,10 +1095,35 @@ ConstantFolding::opnd(Instruction *i, ImmediateValue &imm0, int s)
          i->op = OP_MOV;
          i->setSrc(1, NULL);
       } else
+      if (imm0.reg.data.s32 == -1) {
+         i->op = OP_NEG;
+         i->setSrc(1, NULL);
+      } else
       if (i->dType == TYPE_U32 && imm0.isPow2()) {
          i->op = OP_SHR;
          i->setSrc(1, bld.mkImm(util_logbase2(imm0.reg.data.u32)));
       } else
+      if (i->dType == TYPE_S32 && util_is_power_of_two_or_zero(abs(imm0.reg.data.s32))) {
+         Value *a = i->getSrc(0);
+         int32_t b = imm0.reg.data.s32;
+
+         if (b < 0) {
+            a = bld.getSSA();
+            bld.mkOp1(OP_NEG, TYPE_S32, a, i->getSrc(0));
+            b = -b;
+         }
+
+         Value *sign = bld.getSSA();
+         Value *tmp0 = bld.getSSA();
+         Value *tmp1 = bld.getSSA();
+         bld.mkOp2(OP_SHR, TYPE_U32, sign, a, bld.mkImm(31));
+         bld.mkOp2(OP_ADD, TYPE_U32, tmp0, a, bld.mkImm(b - 1));
+         bld.mkOp3(OP_SELP, TYPE_U32, tmp1, tmp0, a, sign);
+
+         i->op = OP_SHR;
+         i->setSrc(0, tmp1);
+         i->setSrc(1, bld.mkImm(util_logbase2(b)));
+      } else
       if (i->dType == TYPE_U32) {
          Instruction *mul;
          Value *tA, *tB;
@@ -1129,10 +1154,6 @@ ConstantFolding::opnd(Instruction *i, ImmediateValue &imm0, int s)
             bld.mkOp2(OP_SHR, TYPE_U32, i->getDef(0), tB, bld.mkImm(s));
 
          delete_Instruction(prog, i);
-      } else
-      if (imm0.reg.data.s32 == -1) {
-         i->op = OP_NEG;
-         i->setSrc(1, NULL);
       } else {
          LValue *tA, *tB;
          LValue *tD;
-- 
2.14.4