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 "adreno_common.xml.h"
  44 #include "adreno_pm4.xml.h"
  45
  46 enum adreno_rb_depth_format fd_pipe2depth(enum pipe_format format);
  47 enum pc_di_index_size fd_pipe2index(enum pipe_format format);
  48 enum adreno_rb_blend_factor fd_blend_factor(unsigned factor);
  49 enum adreno_pa_su_sc_draw fd_polygon_mode(unsigned mode);
  50 enum adreno_stencil_op fd_stencil_op(unsigned op);
  51
  52 #define A3XX_MAX_MIP_LEVELS 14
  53 /* TBD if it is same on a2xx, but for now: */
  54 #define MAX_MIP_LEVELS A3XX_MAX_MIP_LEVELS
  55
  56 #define FD_DBG_MSGS     0x0001
  57 #define FD_DBG_DISASM   0x0002
  58 #define FD_DBG_DCLEAR   0x0004
  59 #define FD_DBG_FLUSH    0x0008
  60 #define FD_DBG_NOSCIS   0x0010
  61 #define FD_DBG_DIRECT   0x0020
  62 #define FD_DBG_NOBYPASS 0x0040
  63 #define FD_DBG_FRAGHALF 0x0080
  64 #define FD_DBG_NOBIN    0x0100
  65 #define FD_DBG_OPTMSGS  0x0400
  66 #define FD_DBG_OPTDUMP  0x0800
  67 #define FD_DBG_GLSL120  0x1000
  68 #define FD_DBG_NOCP     0x2000
  69
  70 extern int fd_mesa_debug;
  71 extern bool fd_binning_enabled;
  72
  73 #define DBG(fmt, ...) \
  74                 do { if (fd_mesa_debug & FD_DBG_MSGS) \
  75                         debug_printf("%s:%d: "fmt "\n", \
  76                                 __FUNCTION__, __LINE__, ##__VA_ARGS__); } while (0)
  77
  78 /* for conditionally setting boolean flag(s): */
  79 #define COND(bool, val) ((bool) ? (val) : 0)
  80
  81 #define CP_REG(reg) ((0x4 << 16) | ((unsigned int)((reg) - (0x2000))))
  82
  83 static inline uint32_t DRAW(enum pc_di_primtype prim_type,
  84                 enum pc_di_src_sel source_select, enum pc_di_index_size index_size,
  85                 enum pc_di_vis_cull_mode vis_cull_mode,
  86                 uint8_t instances)
  87 {
  88         return (prim_type         << 0) |
  89                         (source_select     << 6) |
  90                         ((index_size & 1)  << 11) |
  91                         ((index_size >> 1) << 13) |
  92                         (vis_cull_mode     << 9) |
  93                         (1                 << 14) |
  94                         (instances         << 24);
  95 }
  96
  97 /* for tracking cmdstream positions that need to be patched: */
  98 struct fd_cs_patch {
  99         uint32_t *cs;
 100         uint32_t val;
 101 };
 102 #define fd_patch_num_elements(buf) ((buf)->size / sizeof(struct fd_cs_patch))
 103 #define fd_patch_element(buf, i)   util_dynarray_element(buf, struct fd_cs_patch, i)
 104
 105 static inline enum pipe_format
 106 pipe_surface_format(struct pipe_surface *psurf)
 107 {
 108         if (!psurf)
 109                 return PIPE_FORMAT_NONE;
 110         return psurf->format;
 111 }
 112
 113 #define LOG_DWORDS 0
 114
 115 static inline void emit_marker(struct fd_ringbuffer *ring, int scratch_idx);
 116
 117 static inline void
 118 OUT_RING(struct fd_ringbuffer *ring, uint32_t data)
 119 {
 120         if (LOG_DWORDS) {
 121                 DBG("ring[%p]: OUT_RING   %04x:  %08x", ring,
 122                                 (uint32_t)(ring->cur - ring->last_start), data);
 123         }
 124         *(ring->cur++) = data;
 125 }
 126
 127 /* like OUT_RING() but appends a cmdstream patch point to 'buf' */
 128 static inline void
 129 OUT_RINGP(struct fd_ringbuffer *ring, uint32_t data,
 130                 struct util_dynarray *buf)
 131 {
 132         if (LOG_DWORDS) {
 133                 DBG("ring[%p]: OUT_RINGP  %04x:  %08x", ring,
 134                                 (uint32_t)(ring->cur - ring->last_start), data);
 135         }
 136         util_dynarray_append(buf, struct fd_cs_patch, ((struct fd_cs_patch){
 137                 .cs  = ring->cur++,
 138                 .val = data,
 139         }));
 140 }
 141
 142 static inline void
 143 OUT_RELOC(struct fd_ringbuffer *ring, struct fd_bo *bo,
 144                 uint32_t offset, uint32_t or, int32_t shift)
 145 {
 146         if (LOG_DWORDS) {
 147                 DBG("ring[%p]: OUT_RELOC   %04x:  %p+%u << %d", ring,
 148                                 (uint32_t)(ring->cur - ring->last_start), bo, offset, shift);
 149         }
 150         fd_ringbuffer_reloc(ring, &(struct fd_reloc){
 151                 .bo = bo,
 152                 .flags = FD_RELOC_READ,
 153                 .offset = offset,
 154                 .or = or,
 155                 .shift = shift,
 156         });
 157 }
 158
 159 static inline void
 160 OUT_RELOCW(struct fd_ringbuffer *ring, struct fd_bo *bo,
 161                 uint32_t offset, uint32_t or, int32_t shift)
 162 {
 163         if (LOG_DWORDS) {
 164                 DBG("ring[%p]: OUT_RELOCW  %04x:  %p+%u << %d", ring,
 165                                 (uint32_t)(ring->cur - ring->last_start), bo, offset, shift);
 166         }
 167         fd_ringbuffer_reloc(ring, &(struct fd_reloc){
 168                 .bo = bo,
 169                 .flags = FD_RELOC_READ | FD_RELOC_WRITE,
 170                 .offset = offset,
 171                 .or = or,
 172                 .shift = shift,
 173         });
 174 }
 175
 176 static inline void BEGIN_RING(struct fd_ringbuffer *ring, uint32_t ndwords)
 177 {
 178         if ((ring->cur + ndwords) >= ring->end) {
 179                 /* this probably won't really work if we have multiple tiles..
 180                  * but it is ok for 2d..  we might need different behavior
 181                  * depending on 2d or 3d pipe.
 182                  */
 183                 DBG("uh oh..");
 184         }
 185 }
 186
 187 static inline void
 188 OUT_PKT0(struct fd_ringbuffer *ring, uint16_t regindx, uint16_t cnt)
 189 {
 190         BEGIN_RING(ring, cnt+1);
 191         OUT_RING(ring, CP_TYPE0_PKT | ((cnt-1) << 16) | (regindx & 0x7FFF));
 192 }
 193
 194 static inline void
 195 OUT_PKT3(struct fd_ringbuffer *ring, uint8_t opcode, uint16_t cnt)
 196 {
 197         BEGIN_RING(ring, cnt+1);
 198         OUT_RING(ring, CP_TYPE3_PKT | ((cnt-1) << 16) | ((opcode & 0xFF) << 8));
 199 }
 200
 201 static inline void
 202 OUT_WFI(struct fd_ringbuffer *ring)
 203 {
 204         OUT_PKT3(ring, CP_WAIT_FOR_IDLE, 1);
 205         OUT_RING(ring, 0x00000000);
 206 }
 207
 208 static inline void
 209 OUT_IB(struct fd_ringbuffer *ring, struct fd_ringmarker *start,
 210                 struct fd_ringmarker *end)
 211 {
 212         uint32_t dwords = fd_ringmarker_dwords(start, end);
 213
 214         assert(dwords > 0);
 215
 216         /* for debug after a lock up, write a unique counter value
 217          * to scratch6 for each IB, to make it easier to match up
 218          * register dumps to cmdstream.  The combination of IB and
 219          * DRAW (scratch7) is enough to "triangulate" the particular
 220          * draw that caused lockup.
 221          */
 222         emit_marker(ring, 6);
 223
 224         OUT_PKT3(ring, CP_INDIRECT_BUFFER_PFD, 2);
 225         fd_ringbuffer_emit_reloc_ring(ring, start, end);
 226         OUT_RING(ring, dwords);
 227
 228         emit_marker(ring, 6);
 229 }
 230
 231 /* CP_SCRATCH_REG4 is used to hold base address for query results: */
 232 #define HW_QUERY_BASE_REG REG_AXXX_CP_SCRATCH_REG4
 233
 234 static inline void
 235 emit_marker(struct fd_ringbuffer *ring, int scratch_idx)
 236 {
 237         extern unsigned marker_cnt;
 238         unsigned reg = REG_AXXX_CP_SCRATCH_REG0 + scratch_idx;
 239         assert(reg != HW_QUERY_BASE_REG);
 240         if (reg == HW_QUERY_BASE_REG)
 241                 return;
 242         OUT_PKT0(ring, reg, 1);
 243         OUT_RING(ring, ++marker_cnt);
 244 }
 245
 246 /* helper to get numeric value from environment variable..  mostly
 247  * just leaving this here because it is helpful to brute-force figure
 248  * out unknown formats, etc, which blob driver does not support:
 249  */
 250 static inline uint32_t env2u(const char *envvar)
 251 {
 252         char *str = getenv(envvar);
 253         if (str)
 254                 return strtoul(str, NULL, 0);
 255         return 0;
 256 }
 257
 258 static inline uint32_t
 259 pack_rgba(enum pipe_format format, const float *rgba)
 260 {
 261         union util_color uc;
 262         util_pack_color(rgba, format, &uc);
 263         return uc.ui[0];
 264 }
 265
 266 #endif /* FREEDRENO_UTIL_H_ */