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 #include "program/prog_instruction.h" 27 28 /* Lower gl_FragCoord (and fddy) to account for driver's requested coordinate- 29 * origin and pixel-center vs. shader. If transformation is required, a 30 * gl_FbWposYTransform uniform is inserted (with the specified state-slots) 31 * and additional instructions are inserted to transform gl_FragCoord (and 32 * fddy src arg). 33 * 34 * This is based on the logic in emit_wpos()/emit_wpos_adjustment() in TGSI 35 * compiler. 36 * 37 * Run before nir_lower_io. 38 */ 39 40 typedef struct { 41 const nir_lower_wpos_ytransform_options *options; 42 nir_shader *shader; 43 nir_builder b; 44 nir_variable *transform; 45 } lower_wpos_ytransform_state; 46 47 static nir_ssa_def * 48 get_transform(lower_wpos_ytransform_state *state) 49 { 50 if (state->transform == NULL) { 51 /* NOTE: name must be prefixed w/ "gl_" to trigger slot based 52 * special handling in uniform setup: 53 */ 54 nir_variable *var = nir_variable_create(state->shader, 55 nir_var_uniform, 56 glsl_vec4_type(), 57 "gl_FbWposYTransform"); 58 59 var->num_state_slots = 1; 60 var->state_slots = ralloc_array(var, nir_state_slot, 1); 61 var->state_slots[0].swizzle = SWIZZLE_XYZW; 62 memcpy(var->state_slots[0].tokens, state->options->state_tokens, 63 sizeof(var->state_slots[0].tokens)); 64 65 state->transform = var; 66 } 67 return nir_load_var(&state->b, state->transform); 68 } 69 70 /* NIR equiv of TGSI CMP instruction: */ 71 static nir_ssa_def * 72 nir_cmp(nir_builder *b, nir_ssa_def *src0, nir_ssa_def *src1, nir_ssa_def *src2) 73 { 74 return nir_bcsel(b, nir_flt(b, src0, nir_imm_float(b, 0.0)), src1, src2); 75 } 76 77 /* see emit_wpos_adjustment() in st_mesa_to_tgsi.c */ 78 static void 79 emit_wpos_adjustment(lower_wpos_ytransform_state *state, 80 nir_intrinsic_instr *intr, 81 bool invert, float adjX, float adjY[2]) 82 { 83 nir_builder *b = &state->b; 84 nir_variable *fragcoord = intr->variables[0]->var; 85 nir_ssa_def *wpostrans, *wpos_temp, *wpos_temp_y, *wpos_input; 86 87 assert(intr->dest.is_ssa); 88 89 b->cursor = nir_before_instr(&intr->instr); 90 91 wpostrans = get_transform(state); 92 wpos_input = nir_load_var(b, fragcoord); 93 94 /* First, apply the coordinate shift: */ 95 if (adjX || adjY[0] || adjY[1]) { 96 if (adjY[0] != adjY[1]) { 97 /* Adjust the y coordinate by adjY[1] or adjY[0] respectively 98 * depending on whether inversion is actually going to be applied 99 * or not, which is determined by testing against the inversion 100 * state variable used below, which will be either +1 or -1. 101 */ 102 nir_ssa_def *adj_temp; 103 104 adj_temp = nir_cmp(b, 105 nir_channel(b, wpostrans, invert ? 2 : 0), 106 nir_imm_vec4(b, adjX, adjY[0], 0.0f, 0.0f), 107 nir_imm_vec4(b, adjX, adjY[1], 0.0f, 0.0f)); 108 109 wpos_temp = nir_fadd(b, wpos_input, adj_temp); 110 } else { 111 wpos_temp = nir_fadd(b, 112 wpos_input, 113 nir_imm_vec4(b, adjX, adjY[0], 0.0f, 0.0f)); 114 } 115 wpos_input = wpos_temp; 116 } else { 117 /* MOV wpos_temp, input[wpos] 118 */ 119 wpos_temp = wpos_input; 120 } 121 122 /* Now the conditional y flip: STATE_FB_WPOS_Y_TRANSFORM.xy/zw will be 123 * inversion/identity, or the other way around if we're drawing to an FBO. 124 */ 125 if (invert) { 126 /* wpos_temp.y = wpos_input * wpostrans.xxxx + wpostrans.yyyy */ 127 wpos_temp_y = nir_fadd(b, nir_fmul(b, nir_channel(b, wpos_temp, 1), 128 nir_channel(b, wpostrans, 0)), 129 nir_channel(b, wpostrans, 1)); 130 } else { 131 /* wpos_temp.y = wpos_input * wpostrans.zzzz + wpostrans.wwww */ 132 wpos_temp_y = nir_fadd(b, nir_fmul(b, nir_channel(b, wpos_temp, 1), 133 nir_channel(b, wpostrans, 2)), 134 nir_channel(b, wpostrans, 3)); 135 } 136 137 wpos_temp = nir_vec4(b, 138 nir_channel(b, wpos_temp, 0), 139 wpos_temp_y, 140 nir_channel(b, wpos_temp, 2), 141 nir_channel(b, wpos_temp, 3)); 142 143 nir_ssa_def_rewrite_uses(&intr->dest.ssa, nir_src_for_ssa(wpos_temp)); 144 } 145 146 static void 147 lower_fragcoord(lower_wpos_ytransform_state *state, nir_intrinsic_instr *intr) 148 { 149 const nir_lower_wpos_ytransform_options *options = state->options; 150 nir_variable *fragcoord = intr->variables[0]->var; 151 float adjX = 0.0f; 152 float adjY[2] = { 0.0f, 0.0f }; 153 bool invert = false; 154 155 /* Based on logic in emit_wpos(): 156 * 157 * Query the pixel center conventions supported by the pipe driver and set 158 * adjX, adjY to help out if it cannot handle the requested one internally. 159 * 160 * The bias of the y-coordinate depends on whether y-inversion takes place 161 * (adjY[1]) or not (adjY[0]), which is in turn dependent on whether we are 162 * drawing to an FBO (causes additional inversion), and whether the pipe 163 * driver origin and the requested origin differ (the latter condition is 164 * stored in the 'invert' variable). 165 * 166 * For height = 100 (i = integer, h = half-integer, l = lower, u = upper): 167 * 168 * center shift only: 169 * i -> h: +0.5 170 * h -> i: -0.5 171 * 172 * inversion only: 173 * l,i -> u,i: ( 0.0 + 1.0) * -1 + 100 = 99 174 * l,h -> u,h: ( 0.5 + 0.0) * -1 + 100 = 99.5 175 * u,i -> l,i: (99.0 + 1.0) * -1 + 100 = 0 176 * u,h -> l,h: (99.5 + 0.0) * -1 + 100 = 0.5 177 * 178 * inversion and center shift: 179 * l,i -> u,h: ( 0.0 + 0.5) * -1 + 100 = 99.5 180 * l,h -> u,i: ( 0.5 + 0.5) * -1 + 100 = 99 181 * u,i -> l,h: (99.0 + 0.5) * -1 + 100 = 0.5 182 * u,h -> l,i: (99.5 + 0.5) * -1 + 100 = 0 183 */ 184 185 if (fragcoord->data.origin_upper_left) { 186 /* Fragment shader wants origin in upper-left */ 187 if (options->fs_coord_origin_upper_left) { 188 /* the driver supports upper-left origin */ 189 } else if (options->fs_coord_origin_lower_left) { 190 /* the driver supports lower-left origin, need to invert Y */ 191 invert = true; 192 } else { 193 unreachable("invalid options"); 194 } 195 } else { 196 /* Fragment shader wants origin in lower-left */ 197 if (options->fs_coord_origin_lower_left) { 198 /* the driver supports lower-left origin */ 199 } else if (options->fs_coord_origin_upper_left) { 200 /* the driver supports upper-left origin, need to invert Y */ 201 invert = true; 202 } else { 203 unreachable("invalid options"); 204 } 205 } 206 207 if (fragcoord->data.pixel_center_integer) { 208 /* Fragment shader wants pixel center integer */ 209 if (options->fs_coord_pixel_center_integer) { 210 /* the driver supports pixel center integer */ 211 adjY[1] = 1.0f; 212 } else if (options->fs_coord_pixel_center_half_integer) { 213 /* the driver supports pixel center half integer, need to bias X,Y */ 214 adjX = -0.5f; 215 adjY[0] = -0.5f; 216 adjY[1] = 0.5f; 217 } else { 218 unreachable("invalid options"); 219 } 220 } else { 221 /* Fragment shader wants pixel center half integer */ 222 if (options->fs_coord_pixel_center_half_integer) { 223 /* the driver supports pixel center half integer */ 224 } else if (options->fs_coord_pixel_center_integer) { 225 /* the driver supports pixel center integer, need to bias X,Y */ 226 adjX = adjY[0] = adjY[1] = 0.5f; 227 } else { 228 unreachable("invalid options"); 229 } 230 } 231 232 emit_wpos_adjustment(state, intr, invert, adjX, adjY); 233 } 234 235 /* turns 'fddy(p)' into 'fddy(fmul(p, transform.x))' */ 236 static void 237 lower_fddy(lower_wpos_ytransform_state *state, nir_alu_instr *fddy) 238 { 239 nir_builder *b = &state->b; 240 nir_ssa_def *p, *pt, *trans; 241 242 b->cursor = nir_before_instr(&fddy->instr); 243 244 p = nir_ssa_for_alu_src(b, fddy, 0); 245 trans = get_transform(state); 246 pt = nir_fmul(b, p, nir_channel(b, trans, 0)); 247 248 nir_instr_rewrite_src(&fddy->instr, 249 &fddy->src[0].src, 250 nir_src_for_ssa(pt)); 251 252 for (unsigned i = 0; i < 4; i++) 253 fddy->src[0].swizzle[i] = MIN2(i, pt->num_components - 1); 254 } 255 256 /* Multiply interp_var_at_offset's offset by transform.x to flip it. */ 257 static void 258 lower_interp_var_at_offset(lower_wpos_ytransform_state *state, 259 nir_intrinsic_instr *interp) 260 { 261 nir_builder *b = &state->b; 262 nir_ssa_def *offset; 263 nir_ssa_def *flip_y; 264 265 b->cursor = nir_before_instr(&interp->instr); 266 267 offset = nir_ssa_for_src(b, interp->src[0], 2); 268 flip_y = nir_fmul(b, nir_channel(b, offset, 1), 269 nir_channel(b, get_transform(state), 0)); 270 nir_instr_rewrite_src(&interp->instr, &interp->src[0], 271 nir_src_for_ssa(nir_vec2(b, nir_channel(b, offset, 0), 272 flip_y))); 273 } 274 275 static void 276 lower_load_sample_pos(lower_wpos_ytransform_state *state, 277 nir_intrinsic_instr *intr) 278 { 279 nir_builder *b = &state->b; 280 b->cursor = nir_after_instr(&intr->instr); 281 282 nir_ssa_def *pos = &intr->dest.ssa; 283 nir_ssa_def *scale = nir_channel(b, get_transform(state), 0); 284 nir_ssa_def *neg_scale = nir_channel(b, get_transform(state), 2); 285 /* Either y or 1-y for scale equal to 1 or -1 respectively. */ 286 nir_ssa_def *flipped_y = 287 nir_fadd(b, nir_fmax(b, neg_scale, nir_imm_float(b, 0.0)), 288 nir_fmul(b, nir_channel(b, pos, 1), scale)); 289 nir_ssa_def *flipped_pos = nir_vec2(b, nir_channel(b, pos, 0), flipped_y); 290 291 nir_ssa_def_rewrite_uses_after(&intr->dest.ssa, nir_src_for_ssa(flipped_pos), 292 flipped_pos->parent_instr); 293 } 294 295 static void 296 lower_wpos_ytransform_block(lower_wpos_ytransform_state *state, nir_block *block) 297 { 298 nir_foreach_instr_safe(instr, block) { 299 if (instr->type == nir_instr_type_intrinsic) { 300 nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr); 301 if (intr->intrinsic == nir_intrinsic_load_var) { 302 nir_deref_var *dvar = intr->variables[0]; 303 nir_variable *var = dvar->var; 304 305 if (var->data.mode == nir_var_shader_in && 306 var->data.location == VARYING_SLOT_POS) { 307 /* gl_FragCoord should not have array/struct deref's: */ 308 assert(dvar->deref.child == NULL); 309 lower_fragcoord(state, intr); 310 } else if (var->data.mode == nir_var_system_value && 311 var->data.location == SYSTEM_VALUE_SAMPLE_POS) { 312 assert(dvar->deref.child == NULL); 313 lower_load_sample_pos(state, intr); 314 } 315 } else if (intr->intrinsic == nir_intrinsic_interp_var_at_offset) { 316 lower_interp_var_at_offset(state, intr); 317 } 318 } else if (instr->type == nir_instr_type_alu) { 319 nir_alu_instr *alu = nir_instr_as_alu(instr); 320 if (alu->op == nir_op_fddy || 321 alu->op == nir_op_fddy_fine || 322 alu->op == nir_op_fddy_coarse) 323 lower_fddy(state, alu); 324 } 325 } 326 } 327 328 static void 329 lower_wpos_ytransform_impl(lower_wpos_ytransform_state *state, nir_function_impl *impl) 330 { 331 nir_builder_init(&state->b, impl); 332 333 nir_foreach_block(block, impl) { 334 lower_wpos_ytransform_block(state, block); 335 } 336 nir_metadata_preserve(impl, nir_metadata_block_index | 337 nir_metadata_dominance); 338 } 339 340 bool 341 nir_lower_wpos_ytransform(nir_shader *shader, 342 const nir_lower_wpos_ytransform_options *options) 343 { 344 lower_wpos_ytransform_state state = { 345 .options = options, 346 .shader = shader, 347 }; 348 349 assert(shader->stage == MESA_SHADER_FRAGMENT); 350 351 nir_foreach_function(function, shader) { 352 if (function->impl) 353 lower_wpos_ytransform_impl(&state, function->impl); 354 } 355 356 return state.transform != NULL; 357 } 358