/* * Copyright © 2010 Intel Corporation * Copyright © 2018 Broadcom * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), * to deal in the Software without restriction, including without limitation * the rights to use, copy, modify, merge, publish, distribute, sublicense, * and/or sell copies of the Software, and to permit persons to whom the * Software is furnished to do so, subject to the following conditions: * * The above copyright notice and this permission notice (including the next * paragraph) shall be included in all copies or substantial portions of the * Software. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER * DEALINGS IN THE SOFTWARE. */ #include "nir.h" #include "nir_builder.h" /** nir_lower_alu.c * * NIR's home for miscellaneous ALU operation lowering implementations. * * Most NIR ALU lowering occurs in nir_opt_algebraic.py, since it's generally * easy to write them there. However, if terms appear multiple times in the * lowered code, it can get very verbose and cause a lot of work for CSE, so * it may end up being easier to write out in C code. * * The shader must be in SSA for this pass. */ static bool lower_alu_instr(nir_builder *b, nir_alu_instr *instr, UNUSED void *cb_data) { nir_def *lowered = NULL; b->cursor = nir_before_instr(&instr->instr); b->fp_math_ctrl = instr->fp_math_ctrl; switch (instr->op) { case nir_op_bitfield_reverse: if (b->shader->options->lower_bitfield_reverse) { assert(instr->def.bit_size == 32); /* For more details, see: * * http://graphics.stanford.edu/~seander/bithacks.html#ReverseParallel */ nir_def *c1 = nir_imm_int(b, 1); nir_def *c2 = nir_imm_int(b, 2); nir_def *c4 = nir_imm_int(b, 4); nir_def *c8 = nir_imm_int(b, 8); nir_def *c16 = nir_imm_int(b, 16); nir_def *c33333333 = nir_imm_int(b, 0x33333333); nir_def *c55555555 = nir_imm_int(b, 0x55555555); nir_def *c0f0f0f0f = nir_imm_int(b, 0x0f0f0f0f); nir_def *c00ff00ff = nir_imm_int(b, 0x00ff00ff); lowered = nir_ssa_for_alu_src(b, instr, 0); /* Swap odd and even bits. */ lowered = nir_ior(b, nir_iand(b, nir_ushr(b, lowered, c1), c55555555), nir_ishl(b, nir_iand(b, lowered, c55555555), c1)); /* Swap consecutive pairs. */ lowered = nir_ior(b, nir_iand(b, nir_ushr(b, lowered, c2), c33333333), nir_ishl(b, nir_iand(b, lowered, c33333333), c2)); /* Swap nibbles. */ lowered = nir_ior(b, nir_iand(b, nir_ushr(b, lowered, c4), c0f0f0f0f), nir_ishl(b, nir_iand(b, lowered, c0f0f0f0f), c4)); /* Swap bytes. */ lowered = nir_ior(b, nir_iand(b, nir_ushr(b, lowered, c8), c00ff00ff), nir_ishl(b, nir_iand(b, lowered, c00ff00ff), c8)); lowered = nir_ior(b, nir_ushr(b, lowered, c16), nir_ishl(b, lowered, c16)); } break; case nir_op_bit_count: if (b->shader->options->lower_bit_count) { /* For more details, see: * * http://graphics.stanford.edu/~seander/bithacks.html#CountBitsSetParallel */ lowered = nir_ssa_for_alu_src(b, instr, 0); unsigned bit_size = lowered->bit_size; lowered = nir_isub(b, lowered, nir_iand_imm(b, nir_ushr_imm(b, lowered, 1), 0x55555555)); lowered = nir_iadd(b, nir_iand_imm(b, lowered, 0x33333333), nir_iand_imm(b, nir_ushr_imm(b, lowered, 2), 0x33333333)); lowered = nir_iadd(b, lowered, nir_ushr_imm(b, lowered, 4)); lowered = nir_iand_imm(b, lowered, 0x0f0f0f0f); lowered = nir_imul_imm(b, lowered, 0x01010101); lowered = nir_u2u32(b, nir_ushr_imm(b, lowered, bit_size - 8)); } break; case nir_op_imul_high: case nir_op_umul_high: if (b->shader->options->lower_mul_high) { nir_def *src0 = nir_ssa_for_alu_src(b, instr, 0); nir_def *src1 = nir_ssa_for_alu_src(b, instr, 1); if (src0->bit_size < 32) { /* Just do the math in 32-bit space and shift the result */ nir_alu_type base_type = nir_op_infos[instr->op].output_type; nir_def *src0_32 = nir_type_convert(b, src0, base_type, base_type | 32, nir_rounding_mode_undef); nir_def *src1_32 = nir_type_convert(b, src1, base_type, base_type | 32, nir_rounding_mode_undef); nir_def *dest_32 = nir_imul(b, src0_32, src1_32); nir_def *dest_shifted = nir_ishr_imm(b, dest_32, src0->bit_size); lowered = nir_type_convert(b, dest_shifted, base_type, base_type | src0->bit_size, nir_rounding_mode_undef); } else { nir_def *cshift = nir_imm_int(b, src0->bit_size / 2); nir_def *cmask = nir_imm_intN_t(b, (1ull << (src0->bit_size / 2)) - 1, src0->bit_size); /* Taken from Figure 8-2 in Hacker's Delight, which is derived * from Knuth's Algorithm M. */ bool is_signed = instr->op == nir_op_imul_high; nir_def *src0l = nir_iand(b, src0, cmask); nir_def *src1l = nir_iand(b, src1, cmask); nir_def *src0h = nir_shr(b, is_signed, src0, cshift); nir_def *src1h = nir_shr(b, is_signed, src1, cshift); nir_def *lo = nir_imul(b, src0l, src1l); nir_def *m1 = nir_imul(b, src0l, src1h); nir_def *m2 = nir_imul(b, src0h, src1l); nir_def *hi = nir_imul(b, src0h, src1h); nir_def *t = nir_iadd(b, m2, nir_ushr(b, lo, cshift)); nir_def *w1 = nir_iand(b, t, cmask); nir_def *w2 = nir_shr(b, is_signed, t, cshift); w1 = nir_iadd(b, m1, w1); hi = nir_iadd(b, hi, nir_iadd(b, nir_shr(b, is_signed, w1, cshift), w2)); lowered = hi; } } break; case nir_op_fmin: case nir_op_fmax: { if (!b->shader->options->lower_fminmax_signed_zero || !nir_alu_instr_is_signed_zero_preserve(instr)) break; nir_def *s0 = nir_ssa_for_alu_src(b, instr, 0); nir_def *s1 = nir_ssa_for_alu_src(b, instr, 1); bool max = instr->op == nir_op_fmax; /* Lower the fmin/fmax to a no_signed_zero fmin/fmax. This ensures that * nir_lower_alu is idempotent, and allows the backend to implement * soundly the no_signed_zero subset of fmin/fmax. */ b->fp_math_ctrl &= ~nir_fp_preserve_signed_zero; nir_def *fminmax = max ? nir_fmax(b, s0, s1) : nir_fmin(b, s0, s1); b->fp_math_ctrl = instr->fp_math_ctrl; /* If we have a constant source, we can usually optimize */ if (s0->num_components == 1 && s0->bit_size == 32) { for (unsigned i = 0; i < 2 && lowered == NULL; ++i) { if (!nir_src_is_const(instr->src[i].src)) continue; uint32_t x = nir_alu_src_as_uint(instr->src[i]); bool pos_zero = x == fui(+0.0); bool neg_zero = x == fui(-0.0); nir_def *zero = i == 0 ? s0 : s1; nir_def *other = i == 0 ? s1 : s0; if (!pos_zero && !neg_zero) { /* The lowering is only required when both sources are zero, so * if we have a nonzero constant source, skip the lowering. */ lowered = fminmax; } else if (pos_zero && max) { b->fp_math_ctrl &= ~nir_fp_preserve_signed_zero; /* max(x, +0.0) = +0.0 < x ? x : +0.0 */ lowered = nir_bcsel(b, nir_flt(b, zero, other), other, zero); } else if (neg_zero && !max) { b->fp_math_ctrl &= ~nir_fp_preserve_signed_zero; /* min(x, -0.0) = x < -0.0 ? x : -0.0 */ lowered = nir_bcsel(b, nir_flt(b, other, zero), other, zero); } } } b->fp_math_ctrl = instr->fp_math_ctrl; /* Fallback on the emulation */ if (!lowered) { nir_def *iminmax = max ? nir_imax(b, s0, s1) : nir_imin(b, s0, s1); lowered = nir_bcsel(b, nir_feq(b, s0, s1), iminmax, fminmax); } break; } default: break; } if (lowered) { nir_def_replace(&instr->def, lowered); return true; } else { return false; } } bool nir_lower_alu(nir_shader *shader) { if (!shader->options->lower_bitfield_reverse && !shader->options->lower_bit_count && !shader->options->lower_mul_high && !shader->options->lower_fminmax_signed_zero) return false; return nir_shader_alu_pass(shader, lower_alu_instr, nir_metadata_control_flow, NULL); }