src/compiler/nir/nir_lower_wpos_ytransform.c

   1 /*
   2  * Copyright © 2015 Red Hat
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  21  * SOFTWARE.
  22  */
  23
  24 #include "nir.h"
  25 #include "nir_builder.h"
  26
  27 /* Lower gl_FragCoord (and fddy) to account for driver's requested coordinate-
  28  * origin and pixel-center vs. shader.  If transformation is required, a
  29  * gl_FbWposYTransform uniform is inserted (with the specified state-slots)
  30  * and additional instructions are inserted to transform gl_FragCoord (and
  31  * fddy src arg).
  32  *
  33  * This is based on the logic in emit_wpos()/emit_wpos_adjustment() in TGSI
  34  * compiler.
  35  *
  36  * Run before nir_lower_io.
  37  */
  38
  39 typedef struct {
  40    const nir_lower_wpos_ytransform_options *options;
  41    nir_shader   *shader;
  42    nir_builder   b;
  43    nir_variable *transform;
  44 } lower_wpos_ytransform_state;
  45
  46 static nir_ssa_def *
  47 get_transform(lower_wpos_ytransform_state *state)
  48 {
  49    if (state->transform == NULL) {
  50       /* NOTE: name must be prefixed w/ "gl_" to trigger slot based
  51        * special handling in uniform setup:
  52        */
  53       nir_variable *var = nir_variable_create(state->shader,
  54                                               nir_var_uniform,
  55                                               glsl_vec4_type(),
  56                                               "gl_FbWposYTransform");
  57
  58       var->num_state_slots = 1;
  59       var->state_slots = ralloc_array(var, nir_state_slot, 1);
  60       memcpy(var->state_slots[0].tokens, state->options->state_tokens,
  61              sizeof(var->state_slots[0].tokens));
  62
  63       state->transform = var;
  64    }
  65    return nir_load_var(&state->b, state->transform);
  66 }
  67
  68 /* NIR equiv of TGSI CMP instruction: */
  69 static nir_ssa_def *
  70 nir_cmp(nir_builder *b, nir_ssa_def *src0, nir_ssa_def *src1, nir_ssa_def *src2)
  71 {
  72    return nir_bcsel(b, nir_flt(b, src0, nir_imm_float(b, 0.0)), src1, src2);
  73 }
  74
  75 /* see emit_wpos_adjustment() in st_mesa_to_tgsi.c */
  76 static void
  77 emit_wpos_adjustment(lower_wpos_ytransform_state *state,
  78                      nir_intrinsic_instr *intr,
  79                      bool invert, float adjX, float adjY[2])
  80 {
  81    nir_builder *b = &state->b;
  82    nir_variable *fragcoord = intr->variables[0]->var;
  83    nir_ssa_def *wpostrans, *wpos_temp, *wpos_temp_y, *wpos_input;
  84
  85    assert(intr->dest.is_ssa);
  86
  87    b->cursor = nir_before_instr(&intr->instr);
  88
  89    wpostrans = get_transform(state);
  90    wpos_input = nir_load_var(b, fragcoord);
  91
  92    /* First, apply the coordinate shift: */
  93    if (adjX || adjY[0] || adjY[1]) {
  94       if (adjY[0] != adjY[1]) {
  95          /* Adjust the y coordinate by adjY[1] or adjY[0] respectively
  96           * depending on whether inversion is actually going to be applied
  97           * or not, which is determined by testing against the inversion
  98           * state variable used below, which will be either +1 or -1.
  99           */
 100          nir_ssa_def *adj_temp;
 101
 102          adj_temp = nir_cmp(b,
 103                             nir_channel(b, wpostrans, invert ? 2 : 0),
 104                             nir_imm_vec4(b, adjX, adjY[0], 0.0f, 0.0f),
 105                             nir_imm_vec4(b, adjX, adjY[1], 0.0f, 0.0f));
 106
 107          wpos_temp = nir_fadd(b, wpos_input, adj_temp);
 108       } else {
 109          wpos_temp = nir_fadd(b,
 110                               wpos_input,
 111                               nir_imm_vec4(b, adjX, adjY[0], 0.0f, 0.0f));
 112       }
 113       wpos_input = wpos_temp;
 114    } else {
 115       /* MOV wpos_temp, input[wpos]
 116        */
 117       wpos_temp = wpos_input;
 118    }
 119
 120    /* Now the conditional y flip: STATE_FB_WPOS_Y_TRANSFORM.xy/zw will be
 121     * inversion/identity, or the other way around if we're drawing to an FBO.
 122     */
 123    if (invert) {
 124       /* MAD wpos_temp.y, wpos_input, wpostrans.xxxx, wpostrans.yyyy
 125        */
 126       wpos_temp_y = nir_ffma(b,
 127                              nir_channel(b, wpos_temp, 1),
 128                              nir_channel(b, wpostrans, 0),
 129                              nir_channel(b, wpostrans, 1));
 130    } else {
 131       /* MAD wpos_temp.y, wpos_input, wpostrans.zzzz, wpostrans.wwww
 132        */
 133       wpos_temp_y = nir_ffma(b,
 134                              nir_channel(b, wpos_temp, 1),
 135                              nir_channel(b, wpostrans, 2),
 136                              nir_channel(b, wpostrans, 3));
 137    }
 138
 139    wpos_temp = nir_vec4(b,
 140                         nir_channel(b, wpos_temp, 0),
 141                         wpos_temp_y,
 142                         nir_channel(b, wpos_temp, 2),
 143                         nir_channel(b, wpos_temp, 3));
 144
 145    nir_ssa_def_rewrite_uses(&intr->dest.ssa, nir_src_for_ssa(wpos_temp));
 146 }
 147
 148 static void
 149 lower_fragcoord(lower_wpos_ytransform_state *state, nir_intrinsic_instr *intr)
 150 {
 151    const nir_lower_wpos_ytransform_options *options = state->options;
 152    nir_variable *fragcoord = intr->variables[0]->var;
 153    float adjX = 0.0f;
 154    float adjY[2] = { 0.0f, 0.0f };
 155    bool invert = false;
 156
 157    /* Based on logic in emit_wpos():
 158     *
 159     * Query the pixel center conventions supported by the pipe driver and set
 160     * adjX, adjY to help out if it cannot handle the requested one internally.
 161     *
 162     * The bias of the y-coordinate depends on whether y-inversion takes place
 163     * (adjY[1]) or not (adjY[0]), which is in turn dependent on whether we are
 164     * drawing to an FBO (causes additional inversion), and whether the the pipe
 165     * driver origin and the requested origin differ (the latter condition is
 166     * stored in the 'invert' variable).
 167     *
 168     * For height = 100 (i = integer, h = half-integer, l = lower, u = upper):
 169     *
 170     * center shift only:
 171     * i -> h: +0.5
 172     * h -> i: -0.5
 173     *
 174     * inversion only:
 175     * l,i -> u,i: ( 0.0 + 1.0) * -1 + 100 = 99
 176     * l,h -> u,h: ( 0.5 + 0.0) * -1 + 100 = 99.5
 177     * u,i -> l,i: (99.0 + 1.0) * -1 + 100 = 0
 178     * u,h -> l,h: (99.5 + 0.0) * -1 + 100 = 0.5
 179     *
 180     * inversion and center shift:
 181     * l,i -> u,h: ( 0.0 + 0.5) * -1 + 100 = 99.5
 182     * l,h -> u,i: ( 0.5 + 0.5) * -1 + 100 = 99
 183     * u,i -> l,h: (99.0 + 0.5) * -1 + 100 = 0.5
 184     * u,h -> l,i: (99.5 + 0.5) * -1 + 100 = 0
 185     */
 186
 187    if (fragcoord->data.origin_upper_left) {
 188       /* Fragment shader wants origin in upper-left */
 189       if (options->fs_coord_origin_upper_left) {
 190          /* the driver supports upper-left origin */
 191       } else if (options->fs_coord_origin_lower_left) {
 192          /* the driver supports lower-left origin, need to invert Y */
 193          invert = true;
 194       } else {
 195          unreachable("invalid options");
 196       }
 197    } else {
 198       /* Fragment shader wants origin in lower-left */
 199       if (options->fs_coord_origin_lower_left) {
 200          /* the driver supports lower-left origin */
 201       } else if (options->fs_coord_origin_upper_left) {
 202          /* the driver supports upper-left origin, need to invert Y */
 203          invert = true;
 204       } else {
 205          unreachable("invalid options");
 206       }
 207    }
 208
 209    if (fragcoord->data.pixel_center_integer) {
 210       /* Fragment shader wants pixel center integer */
 211       if (options->fs_coord_pixel_center_integer) {
 212          /* the driver supports pixel center integer */
 213          adjY[1] = 1.0f;
 214       } else if (options->fs_coord_pixel_center_half_integer) {
 215          /* the driver supports pixel center half integer, need to bias X,Y */
 216          adjX = -0.5f;
 217          adjY[0] = -0.5f;
 218          adjY[1] = 0.5f;
 219       } else {
 220          unreachable("invalid options");
 221       }
 222    } else {
 223       /* Fragment shader wants pixel center half integer */
 224       if (options->fs_coord_pixel_center_half_integer) {
 225          /* the driver supports pixel center half integer */
 226       } else if (options->fs_coord_pixel_center_integer) {
 227          /* the driver supports pixel center integer, need to bias X,Y */
 228          adjX = adjY[0] = adjY[1] = 0.5f;
 229       } else {
 230          unreachable("invalid options");
 231       }
 232    }
 233
 234    emit_wpos_adjustment(state, intr, invert, adjX, adjY);
 235 }
 236
 237 /* turns 'fddy(p)' into 'fddy(fmul(p, transform.x))' */
 238 static void
 239 lower_fddy(lower_wpos_ytransform_state *state, nir_alu_instr *fddy)
 240 {
 241    nir_builder *b = &state->b;
 242    nir_ssa_def *p, *pt, *trans;
 243
 244    b->cursor = nir_before_instr(&fddy->instr);
 245
 246    p = nir_ssa_for_alu_src(b, fddy, 0);
 247    trans = get_transform(state);
 248    pt = nir_fmul(b, p, nir_channel(b, trans, 0));
 249
 250    nir_instr_rewrite_src(&fddy->instr,
 251                          &fddy->src[0].src,
 252                          nir_src_for_ssa(pt));
 253 }
 254
 255 static bool
 256 lower_wpos_ytransform_block(lower_wpos_ytransform_state *state, nir_block *block)
 257 {
 258    nir_foreach_instr_safe(instr, block) {
 259       if (instr->type == nir_instr_type_intrinsic) {
 260          nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
 261          if (intr->intrinsic == nir_intrinsic_load_var) {
 262             nir_deref_var *dvar = intr->variables[0];
 263             nir_variable *var = dvar->var;
 264
 265             if (strcmp(var->name, "gl_FragCoord") == 0) {
 266                /* gl_FragCoord should not have array/struct deref's: */
 267                assert(dvar->deref.child == NULL);
 268                lower_fragcoord(state, intr);
 269             }
 270          }
 271       } else if (instr->type == nir_instr_type_alu) {
 272          nir_alu_instr *alu = nir_instr_as_alu(instr);
 273          if (alu->op == nir_op_fddy)
 274             lower_fddy(state, alu);
 275       }
 276    }
 277
 278    return true;
 279 }
 280
 281 static void
 282 lower_wpos_ytransform_impl(lower_wpos_ytransform_state *state, nir_function_impl *impl)
 283 {
 284    nir_builder_init(&state->b, impl);
 285
 286    nir_foreach_block(block, impl) {
 287       lower_wpos_ytransform_block(state, block);
 288    }
 289    nir_metadata_preserve(impl, nir_metadata_block_index |
 290                                nir_metadata_dominance);
 291 }
 292
 293 bool
 294 nir_lower_wpos_ytransform(nir_shader *shader,
 295                           const nir_lower_wpos_ytransform_options *options)
 296 {
 297    lower_wpos_ytransform_state state = {
 298       .options = options,
 299       .shader = shader,
 300    };
 301
 302    assert(shader->stage == MESA_SHADER_FRAGMENT);
 303
 304    nir_foreach_function(function, shader) {
 305       if (function->impl)
 306          lower_wpos_ytransform_impl(&state, function->impl);
 307    }
 308
 309    return state.transform != NULL;
 310 }