src/gallium/drivers/freedreno/a3xx/fd3_draw.c

   1 /* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */
   2
   3 /*
   4  * Copyright (C) 2013 Rob Clark <robclark@freedesktop.org>
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the "Software"),
   8  * to deal in the Software without restriction, including without limitation
   9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  10  * and/or sell copies of the Software, and to permit persons to whom the
  11  * Software is furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice (including the next
  14  * paragraph) shall be included in all copies or substantial portions of the
  15  * Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  23  * SOFTWARE.
  24  *
  25  * Authors:
  26  *    Rob Clark <robclark@freedesktop.org>
  27  */
  28
  29 #include "pipe/p_state.h"
  30 #include "util/u_string.h"
  31 #include "util/u_memory.h"
  32 #include "util/u_prim.h"
  33 #include "util/u_format.h"
  34
  35 #include "freedreno_state.h"
  36 #include "freedreno_resource.h"
  37
  38 #include "fd3_draw.h"
  39 #include "fd3_context.h"
  40 #include "fd3_emit.h"
  41 #include "fd3_program.h"
  42 #include "fd3_format.h"
  43 #include "fd3_zsa.h"
  44
  45
  46 static void
  47 draw_impl(struct fd_context *ctx, struct fd_ringbuffer *ring,
  48                 struct fd3_emit *emit)
  49 {
  50         const struct pipe_draw_info *info = emit->info;
  51
  52         fd3_emit_state(ctx, ring, emit);
  53
  54         if (emit->dirty & (FD_DIRTY_VTXBUF | FD_DIRTY_VTXSTATE))
  55                 fd3_emit_vertex_bufs(ring, emit);
  56
  57         OUT_PKT0(ring, REG_A3XX_PC_VERTEX_REUSE_BLOCK_CNTL, 1);
  58         OUT_RING(ring, 0x0000000b);             /* PC_VERTEX_REUSE_BLOCK_CNTL */
  59
  60         OUT_PKT0(ring, REG_A3XX_VFD_INDEX_MIN, 4);
  61         OUT_RING(ring, info->min_index);        /* VFD_INDEX_MIN */
  62         OUT_RING(ring, info->max_index);        /* VFD_INDEX_MAX */
  63         OUT_RING(ring, info->start_instance);   /* VFD_INSTANCEID_OFFSET */
  64         OUT_RING(ring, info->start);            /* VFD_INDEX_OFFSET */
  65
  66         OUT_PKT0(ring, REG_A3XX_PC_RESTART_INDEX, 1);
  67         OUT_RING(ring, info->primitive_restart ? /* PC_RESTART_INDEX */
  68                         info->restart_index : 0xffffffff);
  69
  70         fd_draw_emit(ctx, ring,
  71                         emit->key.binning_pass ? IGNORE_VISIBILITY : USE_VISIBILITY,
  72                         info);
  73 }
  74
  75 /* fixup dirty shader state in case some "unrelated" (from the state-
  76  * tracker's perspective) state change causes us to switch to a
  77  * different variant.
  78  */
  79 static void
  80 fixup_shader_state(struct fd_context *ctx, struct ir3_shader_key *key)
  81 {
  82         struct fd3_context *fd3_ctx = fd3_context(ctx);
  83         struct ir3_shader_key *last_key = &fd3_ctx->last_key;
  84
  85         if (!ir3_shader_key_equal(last_key, key)) {
  86                 ctx->dirty |= FD_DIRTY_PROG;
  87
  88                 if (last_key->has_per_samp || key->has_per_samp) {
  89                         if ((last_key->vsaturate_s != key->vsaturate_s) ||
  90                                         (last_key->vsaturate_t != key->vsaturate_t) ||
  91                                         (last_key->vsaturate_r != key->vsaturate_r))
  92                                 ctx->prog.dirty |= FD_SHADER_DIRTY_VP;
  93
  94                         if ((last_key->fsaturate_s != key->fsaturate_s) ||
  95                                         (last_key->fsaturate_t != key->fsaturate_t) ||
  96                                         (last_key->fsaturate_r != key->fsaturate_r))
  97                                 ctx->prog.dirty |= FD_SHADER_DIRTY_FP;
  98                 }
  99
 100                 if (last_key->color_two_side != key->color_two_side)
 101                         ctx->prog.dirty |= FD_SHADER_DIRTY_FP;
 102
 103                 if (last_key->half_precision != key->half_precision)
 104                         ctx->prog.dirty |= FD_SHADER_DIRTY_FP;
 105
 106                 if (last_key->alpha != key->alpha)
 107                         ctx->prog.dirty |= FD_SHADER_DIRTY_FP;
 108
 109                 fd3_ctx->last_key = *key;
 110         }
 111 }
 112
 113 static void
 114 fd3_draw_vbo(struct fd_context *ctx, const struct pipe_draw_info *info)
 115 {
 116         struct fd3_context *fd3_ctx = fd3_context(ctx);
 117         struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
 118         struct fd3_emit emit = {
 119                 .vtx  = &ctx->vtx,
 120                 .prog = &ctx->prog,
 121                 .info = info,
 122                 .key = {
 123                         /* do binning pass first: */
 124                         .binning_pass = true,
 125                         .color_two_side = ctx->rasterizer ? ctx->rasterizer->light_twoside : false,
 126                         .alpha = util_format_is_alpha(pipe_surface_format(pfb->cbufs[0])),
 127                         // TODO set .half_precision based on render target format,
 128                         // ie. float16 and smaller use half, float32 use full..
 129                         .half_precision = !!(fd_mesa_debug & FD_DBG_FRAGHALF),
 130                         .has_per_samp = fd3_ctx->fsaturate || fd3_ctx->vsaturate,
 131                         .vsaturate_s = fd3_ctx->vsaturate_s,
 132                         .vsaturate_t = fd3_ctx->vsaturate_t,
 133                         .vsaturate_r = fd3_ctx->vsaturate_r,
 134                         .fsaturate_s = fd3_ctx->fsaturate_s,
 135                         .fsaturate_t = fd3_ctx->fsaturate_t,
 136                         .fsaturate_r = fd3_ctx->fsaturate_r,
 137                 },
 138                 .format = pipe_surface_format(pfb->cbufs[0]),
 139                 .rasterflat = ctx->rasterizer && ctx->rasterizer->flatshade,
 140         };
 141         unsigned dirty;
 142
 143         fixup_shader_state(ctx, &emit.key);
 144
 145         dirty = ctx->dirty;
 146         emit.dirty = dirty & ~(FD_DIRTY_BLEND);
 147         draw_impl(ctx, ctx->binning_ring, &emit);
 148
 149         /* and now regular (non-binning) pass: */
 150         emit.key.binning_pass = false;
 151         emit.dirty = dirty;
 152         emit.vp = NULL;   /* we changed key so need to refetch vp */
 153         draw_impl(ctx, ctx->ring, &emit);
 154 }
 155
 156 /* clear operations ignore viewport state, so we need to reset it
 157  * based on framebuffer state:
 158  */
 159 static void
 160 reset_viewport(struct fd_ringbuffer *ring, struct pipe_framebuffer_state *pfb)
 161 {
 162         float half_width = pfb->width * 0.5f;
 163         float half_height = pfb->height * 0.5f;
 164
 165         OUT_PKT0(ring, REG_A3XX_GRAS_CL_VPORT_XOFFSET, 4);
 166         OUT_RING(ring, A3XX_GRAS_CL_VPORT_XOFFSET(half_width - 0.5));
 167         OUT_RING(ring, A3XX_GRAS_CL_VPORT_XSCALE(half_width));
 168         OUT_RING(ring, A3XX_GRAS_CL_VPORT_YOFFSET(half_height - 0.5));
 169         OUT_RING(ring, A3XX_GRAS_CL_VPORT_YSCALE(-half_height));
 170 }
 171
 172 /* binning pass cmds for a clear:
 173  * NOTE: newer blob drivers don't use binning for clear, which is probably
 174  * preferable since it is low vtx count.  However that doesn't seem to
 175  * actually work for me.  Not sure if it is depending on support for
 176  * clear pass (rather than using solid-fill shader), or something else
 177  * that newer blob is doing differently.  Once that is figured out, we
 178  * can remove fd3_clear_binning().
 179  */
 180 static void
 181 fd3_clear_binning(struct fd_context *ctx, unsigned dirty)
 182 {
 183         struct fd3_context *fd3_ctx = fd3_context(ctx);
 184         struct fd_ringbuffer *ring = ctx->binning_ring;
 185         struct fd3_emit emit = {
 186                 .vtx  = &fd3_ctx->solid_vbuf_state,
 187                 .prog = &ctx->solid_prog,
 188                 .key = {
 189                         .binning_pass = true,
 190                         .half_precision = true,
 191                 },
 192                 .dirty = dirty,
 193         };
 194
 195         fd3_emit_state(ctx, ring, &emit);
 196         fd3_emit_vertex_bufs(ring, &emit);
 197         reset_viewport(ring, &ctx->framebuffer);
 198
 199         OUT_PKT0(ring, REG_A3XX_PC_PRIM_VTX_CNTL, 1);
 200         OUT_RING(ring, A3XX_PC_PRIM_VTX_CNTL_STRIDE_IN_VPC(0) |
 201                         A3XX_PC_PRIM_VTX_CNTL_POLYMODE_FRONT_PTYPE(PC_DRAW_TRIANGLES) |
 202                         A3XX_PC_PRIM_VTX_CNTL_POLYMODE_BACK_PTYPE(PC_DRAW_TRIANGLES) |
 203                         A3XX_PC_PRIM_VTX_CNTL_PROVOKING_VTX_LAST);
 204         OUT_PKT0(ring, REG_A3XX_VFD_INDEX_MIN, 4);
 205         OUT_RING(ring, 0);            /* VFD_INDEX_MIN */
 206         OUT_RING(ring, 2);            /* VFD_INDEX_MAX */
 207         OUT_RING(ring, 0);            /* VFD_INSTANCEID_OFFSET */
 208         OUT_RING(ring, 0);            /* VFD_INDEX_OFFSET */
 209         OUT_PKT0(ring, REG_A3XX_PC_RESTART_INDEX, 1);
 210         OUT_RING(ring, 0xffffffff);   /* PC_RESTART_INDEX */
 211
 212         fd_event_write(ctx, ring, PERFCOUNTER_STOP);
 213
 214         fd_draw(ctx, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
 215                         DI_SRC_SEL_AUTO_INDEX, 2, INDEX_SIZE_IGN, 0, 0, NULL);
 216 }
 217
 218 static void
 219 fd3_clear(struct fd_context *ctx, unsigned buffers,
 220                 const union pipe_color_union *color, double depth, unsigned stencil)
 221 {
 222         struct fd3_context *fd3_ctx = fd3_context(ctx);
 223         struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
 224         enum pipe_format format = pipe_surface_format(pfb->cbufs[0]);
 225         struct fd_ringbuffer *ring = ctx->ring;
 226         unsigned dirty = ctx->dirty;
 227         unsigned ce, i;
 228         struct fd3_emit emit = {
 229                 .vtx  = &fd3_ctx->solid_vbuf_state,
 230                 .prog = &ctx->solid_prog,
 231                 .key = {
 232                         .half_precision = fd3_half_precision(format),
 233                 },
 234                 .format = format,
 235         };
 236
 237         dirty &= FD_DIRTY_FRAMEBUFFER | FD_DIRTY_SCISSOR;
 238         dirty |= FD_DIRTY_PROG;
 239         emit.dirty = dirty;
 240
 241         fd3_clear_binning(ctx, dirty);
 242
 243         /* emit generic state now: */
 244         fd3_emit_state(ctx, ring, &emit);
 245         reset_viewport(ring, &ctx->framebuffer);
 246
 247         OUT_PKT0(ring, REG_A3XX_RB_BLEND_ALPHA, 1);
 248         OUT_RING(ring, A3XX_RB_BLEND_ALPHA_UINT(0xff) |
 249                         A3XX_RB_BLEND_ALPHA_FLOAT(1.0));
 250
 251         OUT_PKT0(ring, REG_A3XX_RB_RENDER_CONTROL, 1);
 252         OUT_RINGP(ring, A3XX_RB_RENDER_CONTROL_ALPHA_TEST_FUNC(FUNC_NEVER),
 253                         &fd3_ctx->rbrc_patches);
 254
 255         if (buffers & PIPE_CLEAR_DEPTH) {
 256                 OUT_PKT0(ring, REG_A3XX_RB_DEPTH_CONTROL, 1);
 257                 OUT_RING(ring, A3XX_RB_DEPTH_CONTROL_Z_WRITE_ENABLE |
 258                                 A3XX_RB_DEPTH_CONTROL_Z_ENABLE |
 259                                 A3XX_RB_DEPTH_CONTROL_ZFUNC(FUNC_ALWAYS));
 260
 261                 fd_wfi(ctx, ring);
 262                 OUT_PKT0(ring, REG_A3XX_GRAS_CL_VPORT_ZOFFSET, 2);
 263                 OUT_RING(ring, A3XX_GRAS_CL_VPORT_ZOFFSET(0.0));
 264                 OUT_RING(ring, A3XX_GRAS_CL_VPORT_ZSCALE(depth));
 265                 ctx->dirty |= FD_DIRTY_VIEWPORT;
 266         } else {
 267                 OUT_PKT0(ring, REG_A3XX_RB_DEPTH_CONTROL, 1);
 268                 OUT_RING(ring, A3XX_RB_DEPTH_CONTROL_ZFUNC(FUNC_NEVER));
 269         }
 270
 271         if (buffers & PIPE_CLEAR_STENCIL) {
 272                 OUT_PKT0(ring, REG_A3XX_RB_STENCILREFMASK, 2);
 273                 OUT_RING(ring, A3XX_RB_STENCILREFMASK_STENCILREF(stencil) |
 274                                 A3XX_RB_STENCILREFMASK_STENCILMASK(stencil) |
 275                                 A3XX_RB_STENCILREFMASK_STENCILWRITEMASK(0xff));
 276                 OUT_RING(ring, A3XX_RB_STENCILREFMASK_STENCILREF(0) |
 277                                 A3XX_RB_STENCILREFMASK_STENCILMASK(0) |
 278                                 0xff000000 | // XXX ???
 279                                 A3XX_RB_STENCILREFMASK_STENCILWRITEMASK(0xff));
 280
 281                 OUT_PKT0(ring, REG_A3XX_RB_STENCIL_CONTROL, 1);
 282                 OUT_RING(ring, A3XX_RB_STENCIL_CONTROL_STENCIL_ENABLE |
 283                                 A3XX_RB_STENCIL_CONTROL_FUNC(FUNC_ALWAYS) |
 284                                 A3XX_RB_STENCIL_CONTROL_FAIL(STENCIL_KEEP) |
 285                                 A3XX_RB_STENCIL_CONTROL_ZPASS(STENCIL_REPLACE) |
 286                                 A3XX_RB_STENCIL_CONTROL_ZFAIL(STENCIL_KEEP) |
 287                                 A3XX_RB_STENCIL_CONTROL_FUNC_BF(FUNC_NEVER) |
 288                                 A3XX_RB_STENCIL_CONTROL_FAIL_BF(STENCIL_KEEP) |
 289                                 A3XX_RB_STENCIL_CONTROL_ZPASS_BF(STENCIL_KEEP) |
 290                                 A3XX_RB_STENCIL_CONTROL_ZFAIL_BF(STENCIL_KEEP));
 291         } else {
 292                 OUT_PKT0(ring, REG_A3XX_RB_STENCILREFMASK, 2);
 293                 OUT_RING(ring, A3XX_RB_STENCILREFMASK_STENCILREF(0) |
 294                                 A3XX_RB_STENCILREFMASK_STENCILMASK(0) |
 295                                 A3XX_RB_STENCILREFMASK_STENCILWRITEMASK(0));
 296                 OUT_RING(ring, A3XX_RB_STENCILREFMASK_BF_STENCILREF(0) |
 297                                 A3XX_RB_STENCILREFMASK_BF_STENCILMASK(0) |
 298                                 A3XX_RB_STENCILREFMASK_BF_STENCILWRITEMASK(0));
 299
 300                 OUT_PKT0(ring, REG_A3XX_RB_STENCIL_CONTROL, 1);
 301                 OUT_RING(ring, A3XX_RB_STENCIL_CONTROL_FUNC(FUNC_NEVER) |
 302                                 A3XX_RB_STENCIL_CONTROL_FAIL(STENCIL_KEEP) |
 303                                 A3XX_RB_STENCIL_CONTROL_ZPASS(STENCIL_KEEP) |
 304                                 A3XX_RB_STENCIL_CONTROL_ZFAIL(STENCIL_KEEP) |
 305                                 A3XX_RB_STENCIL_CONTROL_FUNC_BF(FUNC_NEVER) |
 306                                 A3XX_RB_STENCIL_CONTROL_FAIL_BF(STENCIL_KEEP) |
 307                                 A3XX_RB_STENCIL_CONTROL_ZPASS_BF(STENCIL_KEEP) |
 308                                 A3XX_RB_STENCIL_CONTROL_ZFAIL_BF(STENCIL_KEEP));
 309         }
 310
 311         if (buffers & PIPE_CLEAR_COLOR) {
 312                 ce = 0xf;
 313         } else {
 314                 ce = 0x0;
 315         }
 316
 317         for (i = 0; i < 4; i++) {
 318                 OUT_PKT0(ring, REG_A3XX_RB_MRT_CONTROL(i), 1);
 319                 OUT_RING(ring, A3XX_RB_MRT_CONTROL_ROP_CODE(ROP_COPY) |
 320                                 A3XX_RB_MRT_CONTROL_DITHER_MODE(DITHER_ALWAYS) |
 321                                 A3XX_RB_MRT_CONTROL_COMPONENT_ENABLE(ce));
 322
 323                 OUT_PKT0(ring, REG_A3XX_RB_MRT_BLEND_CONTROL(i), 1);
 324                 OUT_RING(ring, A3XX_RB_MRT_BLEND_CONTROL_RGB_SRC_FACTOR(FACTOR_ONE) |
 325                                 A3XX_RB_MRT_BLEND_CONTROL_RGB_BLEND_OPCODE(BLEND_DST_PLUS_SRC) |
 326                                 A3XX_RB_MRT_BLEND_CONTROL_RGB_DEST_FACTOR(FACTOR_ZERO) |
 327                                 A3XX_RB_MRT_BLEND_CONTROL_ALPHA_SRC_FACTOR(FACTOR_ONE) |
 328                                 A3XX_RB_MRT_BLEND_CONTROL_ALPHA_BLEND_OPCODE(BLEND_DST_PLUS_SRC) |
 329                                 A3XX_RB_MRT_BLEND_CONTROL_ALPHA_DEST_FACTOR(FACTOR_ZERO));
 330         }
 331
 332         OUT_PKT0(ring, REG_A3XX_GRAS_SU_MODE_CONTROL, 1);
 333         OUT_RING(ring, A3XX_GRAS_SU_MODE_CONTROL_LINEHALFWIDTH(0));
 334
 335         fd3_emit_vertex_bufs(ring, &emit);
 336
 337         fd3_emit_constant(ring, SB_FRAG_SHADER, 0, 0, 4, color->ui, NULL);
 338
 339         OUT_PKT0(ring, REG_A3XX_PC_PRIM_VTX_CNTL, 1);
 340         OUT_RING(ring, A3XX_PC_PRIM_VTX_CNTL_STRIDE_IN_VPC(0) |
 341                         A3XX_PC_PRIM_VTX_CNTL_POLYMODE_FRONT_PTYPE(PC_DRAW_TRIANGLES) |
 342                         A3XX_PC_PRIM_VTX_CNTL_POLYMODE_BACK_PTYPE(PC_DRAW_TRIANGLES) |
 343                         A3XX_PC_PRIM_VTX_CNTL_PROVOKING_VTX_LAST);
 344         OUT_PKT0(ring, REG_A3XX_VFD_INDEX_MIN, 4);
 345         OUT_RING(ring, 0);            /* VFD_INDEX_MIN */
 346         OUT_RING(ring, 2);            /* VFD_INDEX_MAX */
 347         OUT_RING(ring, 0);            /* VFD_INSTANCEID_OFFSET */
 348         OUT_RING(ring, 0);            /* VFD_INDEX_OFFSET */
 349         OUT_PKT0(ring, REG_A3XX_PC_RESTART_INDEX, 1);
 350         OUT_RING(ring, 0xffffffff);   /* PC_RESTART_INDEX */
 351
 352         fd_event_write(ctx, ring, PERFCOUNTER_STOP);
 353
 354         fd_draw(ctx, ring, DI_PT_RECTLIST, USE_VISIBILITY,
 355                         DI_SRC_SEL_AUTO_INDEX, 2, INDEX_SIZE_IGN, 0, 0, NULL);
 356 }
 357
 358 void
 359 fd3_draw_init(struct pipe_context *pctx)
 360 {
 361         struct fd_context *ctx = fd_context(pctx);
 362         ctx->draw_vbo = fd3_draw_vbo;
 363         ctx->clear = fd3_clear;
 364 }