src/gallium/drivers/freedreno/freedreno_util.h

   1 /* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */
   2
   3 /*
   4  * Copyright (C) 2012 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 #ifndef FREEDRENO_UTIL_H_
  30 #define FREEDRENO_UTIL_H_
  31
  32 #include <freedreno_drmif.h>
  33 #include <freedreno_ringbuffer.h>
  34
  35 #include "pipe/p_format.h"
  36 #include "pipe/p_state.h"
  37 #include "util/u_debug.h"
  38 #include "util/u_math.h"
  39 #include "util/u_half.h"
  40 #include "util/u_dynarray.h"
  41 #include "util/u_pack_color.h"
  42
  43 #include "disasm.h"
  44 #include "adreno_common.xml.h"
  45 #include "adreno_pm4.xml.h"
  46
  47 enum adreno_rb_depth_format fd_pipe2depth(enum pipe_format format);
  48 enum pc_di_index_size fd_pipe2index(enum pipe_format format);
  49 enum adreno_rb_blend_factor fd_blend_factor(unsigned factor);
  50 enum adreno_pa_su_sc_draw fd_polygon_mode(unsigned mode);
  51 enum adreno_stencil_op fd_stencil_op(unsigned op);
  52
  53 #define A3XX_MAX_MIP_LEVELS 14
  54 /* TBD if it is same on a2xx, but for now: */
  55 #define MAX_MIP_LEVELS A3XX_MAX_MIP_LEVELS
  56
  57 #define A2XX_MAX_RENDER_TARGETS 1
  58 #define A3XX_MAX_RENDER_TARGETS 4
  59 #define A4XX_MAX_RENDER_TARGETS 8
  60
  61 #define MAX_RENDER_TARGETS A4XX_MAX_RENDER_TARGETS
  62
  63 #define FD_DBG_MSGS     0x0001
  64 #define FD_DBG_DISASM   0x0002
  65 #define FD_DBG_DCLEAR   0x0004
  66 #define FD_DBG_DDRAW    0x0008
  67 #define FD_DBG_NOSCIS   0x0010
  68 #define FD_DBG_DIRECT   0x0020
  69 #define FD_DBG_NOBYPASS 0x0040
  70 #define FD_DBG_FRAGHALF 0x0080
  71 #define FD_DBG_NOBIN    0x0100
  72 #define FD_DBG_OPTMSGS  0x0200
  73 #define FD_DBG_GLSL120  0x0400
  74 #define FD_DBG_SHADERDB 0x0800
  75 #define FD_DBG_FLUSH    0x1000
  76 #define FD_DBG_DEQP     0x2000
  77 #define FD_DBG_NIR      0x4000
  78 #define FD_DBG_REORDER  0x8000
  79 #define FD_DBG_BSTAT   0x10000
  80
  81 extern int fd_mesa_debug;
  82 extern bool fd_binning_enabled;
  83
  84 #define DBG(fmt, ...) \
  85                 do { if (fd_mesa_debug & FD_DBG_MSGS) \
  86                         debug_printf("%s:%d: "fmt "\n", \
  87                                 __FUNCTION__, __LINE__, ##__VA_ARGS__); } while (0)
  88
  89 /* for conditionally setting boolean flag(s): */
  90 #define COND(bool, val) ((bool) ? (val) : 0)
  91
  92 #define CP_REG(reg) ((0x4 << 16) | ((unsigned int)((reg) - (0x2000))))
  93
  94 static inline uint32_t DRAW(enum pc_di_primtype prim_type,
  95                 enum pc_di_src_sel source_select, enum pc_di_index_size index_size,
  96                 enum pc_di_vis_cull_mode vis_cull_mode,
  97                 uint8_t instances)
  98 {
  99         return (prim_type         << 0) |
 100                         (source_select     << 6) |
 101                         ((index_size & 1)  << 11) |
 102                         ((index_size >> 1) << 13) |
 103                         (vis_cull_mode     << 9) |
 104                         (1                 << 14) |
 105                         (instances         << 24);
 106 }
 107
 108 /* for tracking cmdstream positions that need to be patched: */
 109 struct fd_cs_patch {
 110         uint32_t *cs;
 111         uint32_t val;
 112 };
 113 #define fd_patch_num_elements(buf) ((buf)->size / sizeof(struct fd_cs_patch))
 114 #define fd_patch_element(buf, i)   util_dynarray_element(buf, struct fd_cs_patch, i)
 115
 116 static inline enum pipe_format
 117 pipe_surface_format(struct pipe_surface *psurf)
 118 {
 119         if (!psurf)
 120                 return PIPE_FORMAT_NONE;
 121         return psurf->format;
 122 }
 123
 124 static inline bool
 125 fd_surface_half_precision(const struct pipe_surface *psurf)
 126 {
 127         enum pipe_format format;
 128
 129         if (!psurf)
 130                 return true;
 131
 132         format = psurf->format;
 133
 134         /* colors are provided in consts, which go through cov.f32f16, which will
 135          * break these values
 136          */
 137         if (util_format_is_pure_integer(format))
 138                 return false;
 139
 140         /* avoid losing precision on 32-bit float formats */
 141         if (util_format_is_float(format) &&
 142                 util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 0) == 32)
 143                 return false;
 144
 145         return true;
 146 }
 147
 148 static inline unsigned
 149 fd_sampler_first_level(const struct pipe_sampler_view *view)
 150 {
 151         if (view->target == PIPE_BUFFER)
 152                 return 0;
 153         return view->u.tex.first_level;
 154 }
 155
 156 static inline unsigned
 157 fd_sampler_last_level(const struct pipe_sampler_view *view)
 158 {
 159         if (view->target == PIPE_BUFFER)
 160                 return 0;
 161         return view->u.tex.last_level;
 162 }
 163
 164 static inline bool
 165 fd_half_precision(struct pipe_framebuffer_state *pfb)
 166 {
 167         unsigned i;
 168
 169         for (i = 0; i < pfb->nr_cbufs; i++)
 170                 if (!fd_surface_half_precision(pfb->cbufs[i]))
 171                         return false;
 172
 173         return true;
 174 }
 175
 176 #define LOG_DWORDS 0
 177
 178 static inline void emit_marker(struct fd_ringbuffer *ring, int scratch_idx);
 179
 180 static inline void
 181 OUT_RING(struct fd_ringbuffer *ring, uint32_t data)
 182 {
 183         if (LOG_DWORDS) {
 184                 DBG("ring[%p]: OUT_RING   %04x:  %08x", ring,
 185                                 (uint32_t)(ring->cur - ring->last_start), data);
 186         }
 187         fd_ringbuffer_emit(ring, data);
 188 }
 189
 190 /* like OUT_RING() but appends a cmdstream patch point to 'buf' */
 191 static inline void
 192 OUT_RINGP(struct fd_ringbuffer *ring, uint32_t data,
 193                 struct util_dynarray *buf)
 194 {
 195         if (LOG_DWORDS) {
 196                 DBG("ring[%p]: OUT_RINGP  %04x:  %08x", ring,
 197                                 (uint32_t)(ring->cur - ring->last_start), data);
 198         }
 199         util_dynarray_append(buf, struct fd_cs_patch, ((struct fd_cs_patch){
 200                 .cs  = ring->cur++,
 201                 .val = data,
 202         }));
 203 }
 204
 205 static inline void
 206 OUT_RELOC(struct fd_ringbuffer *ring, struct fd_bo *bo,
 207                 uint32_t offset, uint32_t or, int32_t shift)
 208 {
 209         if (LOG_DWORDS) {
 210                 DBG("ring[%p]: OUT_RELOC   %04x:  %p+%u << %d", ring,
 211                                 (uint32_t)(ring->cur - ring->last_start), bo, offset, shift);
 212         }
 213         debug_assert(offset < fd_bo_size(bo));
 214         fd_ringbuffer_reloc(ring, &(struct fd_reloc){
 215                 .bo = bo,
 216                 .flags = FD_RELOC_READ,
 217                 .offset = offset,
 218                 .or = or,
 219                 .shift = shift,
 220         });
 221 }
 222
 223 static inline void
 224 OUT_RELOCW(struct fd_ringbuffer *ring, struct fd_bo *bo,
 225                 uint32_t offset, uint32_t or, int32_t shift)
 226 {
 227         if (LOG_DWORDS) {
 228                 DBG("ring[%p]: OUT_RELOCW  %04x:  %p+%u << %d", ring,
 229                                 (uint32_t)(ring->cur - ring->last_start), bo, offset, shift);
 230         }
 231         debug_assert(offset < fd_bo_size(bo));
 232         fd_ringbuffer_reloc(ring, &(struct fd_reloc){
 233                 .bo = bo,
 234                 .flags = FD_RELOC_READ | FD_RELOC_WRITE,
 235                 .offset = offset,
 236                 .or = or,
 237                 .shift = shift,
 238         });
 239 }
 240
 241 static inline void BEGIN_RING(struct fd_ringbuffer *ring, uint32_t ndwords)
 242 {
 243         if (ring->cur + ndwords >= ring->end)
 244                 fd_ringbuffer_grow(ring, ndwords);
 245 }
 246
 247 static inline void
 248 OUT_PKT0(struct fd_ringbuffer *ring, uint16_t regindx, uint16_t cnt)
 249 {
 250         BEGIN_RING(ring, cnt+1);
 251         OUT_RING(ring, CP_TYPE0_PKT | ((cnt-1) << 16) | (regindx & 0x7FFF));
 252 }
 253
 254 static inline void
 255 OUT_PKT2(struct fd_ringbuffer *ring)
 256 {
 257         BEGIN_RING(ring, 1);
 258         OUT_RING(ring, CP_TYPE2_PKT);
 259 }
 260
 261 static inline void
 262 OUT_PKT3(struct fd_ringbuffer *ring, uint8_t opcode, uint16_t cnt)
 263 {
 264         BEGIN_RING(ring, cnt+1);
 265         OUT_RING(ring, CP_TYPE3_PKT | ((cnt-1) << 16) | ((opcode & 0xFF) << 8));
 266 }
 267
 268 static inline void
 269 OUT_WFI(struct fd_ringbuffer *ring)
 270 {
 271         OUT_PKT3(ring, CP_WAIT_FOR_IDLE, 1);
 272         OUT_RING(ring, 0x00000000);
 273 }
 274
 275 static inline void
 276 __OUT_IB(struct fd_ringbuffer *ring, bool prefetch, struct fd_ringbuffer *target)
 277 {
 278         unsigned count = fd_ringbuffer_cmd_count(target);
 279
 280         /* for debug after a lock up, write a unique counter value
 281          * to scratch6 for each IB, to make it easier to match up
 282          * register dumps to cmdstream.  The combination of IB and
 283          * DRAW (scratch7) is enough to "triangulate" the particular
 284          * draw that caused lockup.
 285          */
 286         emit_marker(ring, 6);
 287
 288         for (unsigned i = 0; i < count; i++) {
 289                 uint32_t dwords;
 290                 OUT_PKT3(ring, prefetch ? CP_INDIRECT_BUFFER_PFE : CP_INDIRECT_BUFFER_PFD, 2);
 291                 dwords = fd_ringbuffer_emit_reloc_ring_full(ring, target, i) / 4;
 292                 assert(dwords > 0);
 293                 OUT_RING(ring, dwords);
 294                 OUT_PKT2(ring);
 295         }
 296
 297         emit_marker(ring, 6);
 298 }
 299
 300 /* CP_SCRATCH_REG4 is used to hold base address for query results: */
 301 #define HW_QUERY_BASE_REG REG_AXXX_CP_SCRATCH_REG4
 302
 303 static inline void
 304 emit_marker(struct fd_ringbuffer *ring, int scratch_idx)
 305 {
 306         extern unsigned marker_cnt;
 307         unsigned reg = REG_AXXX_CP_SCRATCH_REG0 + scratch_idx;
 308         assert(reg != HW_QUERY_BASE_REG);
 309         if (reg == HW_QUERY_BASE_REG)
 310                 return;
 311         OUT_PKT0(ring, reg, 1);
 312         OUT_RING(ring, ++marker_cnt);
 313 }
 314
 315 /* helper to get numeric value from environment variable..  mostly
 316  * just leaving this here because it is helpful to brute-force figure
 317  * out unknown formats, etc, which blob driver does not support:
 318  */
 319 static inline uint32_t env2u(const char *envvar)
 320 {
 321         char *str = getenv(envvar);
 322         if (str)
 323                 return strtoul(str, NULL, 0);
 324         return 0;
 325 }
 326
 327 static inline uint32_t
 328 pack_rgba(enum pipe_format format, const float *rgba)
 329 {
 330         union util_color uc;
 331         util_pack_color(rgba, format, &uc);
 332         return uc.ui[0];
 333 }
 334
 335 /*
 336  * swap - swap value of @a and @b
 337  */
 338 #define swap(a, b) \
 339         do { __typeof(a) __tmp = (a); (a) = (b); (b) = __tmp; } while (0)
 340
 341 #define foreach_bit(b, mask) \
 342         for (uint32_t _m = (mask); _m && ({(b) = u_bit_scan(&_m); 1;});)
 343
 344 #endif /* FREEDRENO_UTIL_H_ */