diff options
author | Ilia Mirkin <[email protected]> | 2017-11-10 22:10:46 -0500 |
---|---|---|
committer | Ilia Mirkin <[email protected]> | 2017-11-26 01:10:03 -0500 |
commit | 50e913b9c5d311334281da89b4e9969d48fd62b6 (patch) | |
tree | b3cc1d2f1105f0e184059845ff3fa19bde791290 /src | |
parent | 3079993727ddfe6b09e1f0ec8f834691fd654b46 (diff) |
nv50/ir: add optimization for modulo by a non-power-of-2 value
We can still use the optimized division methods which make use of
multiplication with overflow.
Signed-off-by: Ilia Mirkin <[email protected]>
Reviewed-by: Tobias Klausmann <[email protected]>
Diffstat (limited to 'src')
-rw-r--r-- | src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp | 15 |
1 files changed, 15 insertions, 0 deletions
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp index 2448c737e73..cfd623ea877 100644 --- a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp +++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp @@ -1192,6 +1192,21 @@ ConstantFolding::opnd(Instruction *i, ImmediateValue &imm0, int s) delete_Instruction(prog, i); } + } else if (s == 1) { + // In this case, we still want the optimized lowering that we get + // from having division by an immediate. + // + // a % b == a - (a/b) * b + bld.setPosition(i, false); + Value *div = bld.mkOp2v(OP_DIV, i->sType, bld.getSSA(), + i->getSrc(0), i->getSrc(1)); + newi = bld.mkOp2(OP_ADD, i->sType, i->getDef(0), i->getSrc(0), + bld.mkOp2v(OP_MUL, i->sType, bld.getSSA(), div, i->getSrc(1))); + // TODO: Check that target supports this. In this case, we know that + // all backends do. + newi->src(1).mod = Modifier(NV50_IR_MOD_NEG); + + delete_Instruction(prog, i); } break; |