search for: half_input

Displaying 4 results from an estimated 4 matches for "half_input".

[Mesa-dev] [PATCH 2/2] nvc0/ir: improve precision of double RCP/RSQ results
2015 Feb 23
2
[Mesa-dev] [PATCH 2/2] nvc0/ir: improve precision of double RCP/RSQ results
...), two, guess), > + bld.mkOp2v(OP_MUL, TYPE_F64, bld.getSSA(8), input, > + bld.mkOp2v(OP_MUL, TYPE_F64, bld.getSSA(8), guess, guess))); > + } else { > + // RSQ: x_{n+1} = x_n (1.5 - 0.5 * input * x_n^2) > + Value *half_input = bld.getSSA(8), *three_half = bld.getSSA(8); > + bld.mkCvt(OP_CVT, TYPE_F64, half_input, TYPE_F32, bld.loadImm(NULL, -0.5f)); > + bld.mkCvt(OP_CVT, TYPE_F64, three_half, TYPE_F32, bld.loadImm(NULL, 1.5f)); > + > + half_input = bld.mkOp2v(OP_MUL, TYPE_F64, bld.getSSA(8),...
[PATCH 1/2] nv50/ir: add fp64 support on G200 (NVA0)
2015 Feb 23
2
[PATCH 1/2] nv50/ir: add fp64 support on G200 (NVA0)
...0, (pred = bld.getSSA(1, FILE_FLAGS))); + bld.mkOp1(OP_NEG, TYPE_F64, neg, guess) + ->setPredicate(CC_S, pred); + bld.mkMov(copy, guess) + ->setPredicate(CC_NS, pred); + guess = bld.mkOp2v(OP_UNION, TYPE_U64, bld.getSSA(8), neg, copy); + } else { + Value *half_input = bld.getSSA(8), *three_half = bld.getSSA(8); + bld.mkCvt(OP_CVT, TYPE_F64, half_input, TYPE_F32, bld.loadImm(NULL, -0.5f)); + bld.mkCvt(OP_CVT, TYPE_F64, three_half, TYPE_F32, bld.loadImm(NULL, 1.5f)); + + half_input = bld.mkOp2v(OP_MUL, TYPE_F64, bld.getSSA(8), half_input, abs); +...
[PATCH 2/2] nvc0/ir: improve precision of double RCP/RSQ results
2015 Feb 23
0
[PATCH 2/2] nvc0/ir: improve precision of double RCP/RSQ results
...L, TYPE_F64, bld.getSSA(8), two, guess), + bld.mkOp2v(OP_MUL, TYPE_F64, bld.getSSA(8), input, + bld.mkOp2v(OP_MUL, TYPE_F64, bld.getSSA(8), guess, guess))); + } else { + // RSQ: x_{n+1} = x_n (1.5 - 0.5 * input * x_n^2) + Value *half_input = bld.getSSA(8), *three_half = bld.getSSA(8); + bld.mkCvt(OP_CVT, TYPE_F64, half_input, TYPE_F32, bld.loadImm(NULL, -0.5f)); + bld.mkCvt(OP_CVT, TYPE_F64, three_half, TYPE_F32, bld.loadImm(NULL, 1.5f)); + + half_input = bld.mkOp2v(OP_MUL, TYPE_F64, bld.getSSA(8), half_input, input);...
[Mesa-dev] [PATCH 2/2] nvc0/ir: improve precision of double RCP/RSQ results
2015 Feb 23
0
[Mesa-dev] [PATCH 2/2] nvc0/ir: improve precision of double RCP/RSQ results
...> + bld.mkOp2v(OP_MUL, TYPE_F64, bld.getSSA(8), input, >> + bld.mkOp2v(OP_MUL, TYPE_F64, bld.getSSA(8), guess, guess))); >> + } else { >> + // RSQ: x_{n+1} = x_n (1.5 - 0.5 * input * x_n^2) >> + Value *half_input = bld.getSSA(8), *three_half = bld.getSSA(8); >> + bld.mkCvt(OP_CVT, TYPE_F64, half_input, TYPE_F32, bld.loadImm(NULL, -0.5f)); >> + bld.mkCvt(OP_CVT, TYPE_F64, three_half, TYPE_F32, bld.loadImm(NULL, 1.5f)); >> + >> + half_input = bld.mkOp2v(OP_MUL, TYPE_F64,...