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
  42 #include "adreno_common.xml.h"
  43 #include "adreno_pm4.xml.h"
  44
  45 enum adreno_rb_depth_format fd_pipe2depth(enum pipe_format format);
  46 enum pc_di_index_size fd_pipe2index(enum pipe_format format);
  47 enum adreno_rb_blend_factor fd_blend_factor(unsigned factor);
  48 enum adreno_pa_su_sc_draw fd_polygon_mode(unsigned mode);
  49 enum adreno_stencil_op fd_stencil_op(unsigned op);
  50
  51 #define A3XX_MAX_MIP_LEVELS 14
  52 /* TBD if it is same on a2xx, but for now: */
  53 #define MAX_MIP_LEVELS A3XX_MAX_MIP_LEVELS
  54
  55 #define FD_DBG_MSGS     0x0001
  56 #define FD_DBG_DISASM   0x0002
  57 #define FD_DBG_DCLEAR   0x0004
  58 #define FD_DBG_DGMEM    0x0008
  59 #define FD_DBG_DSCIS    0x0010
  60 #define FD_DBG_DIRECT   0x0020
  61 #define FD_DBG_DBYPASS  0x0040
  62 #define FD_DBG_FRAGHALF 0x0080
  63 #define FD_DBG_NOBIN    0x0100
  64 #define FD_DBG_NOOPT    0x0200
  65 #define FD_DBG_OPTMSGS  0x0400
  66 #define FD_DBG_OPTDUMP  0x0800
  67 #define FD_DBG_GLSL130  0x1000
  68
  69 extern int fd_mesa_debug;
  70 extern bool fd_binning_enabled;
  71
  72 #define DBG(fmt, ...) \
  73                 do { if (fd_mesa_debug & FD_DBG_MSGS) \
  74                         debug_printf("%s:%d: "fmt "\n", \
  75                                 __FUNCTION__, __LINE__, ##__VA_ARGS__); } while (0)
  76
  77 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof((arr)[0]))
  78
  79 /* for conditionally setting boolean flag(s): */
  80 #define COND(bool, val) ((bool) ? (val) : 0)
  81
  82 #define CP_REG(reg) ((0x4 << 16) | ((unsigned int)((reg) - (0x2000))))
  83
  84 static inline uint32_t DRAW(enum pc_di_primtype prim_type,
  85                 enum pc_di_src_sel source_select, enum pc_di_index_size index_size,
  86                 enum pc_di_vis_cull_mode vis_cull_mode)
  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 }
  95
  96 /* for tracking cmdstream positions that need to be patched: */
  97 struct fd_cs_patch {
  98         uint32_t *cs;
  99         uint32_t val;
 100 };
 101 #define fd_patch_num_elements(buf) ((buf)->size / sizeof(struct fd_cs_patch))
 102 #define fd_patch_element(buf, i)   util_dynarray_element(buf, struct fd_cs_patch, i)
 103
 104 static inline enum pipe_format
 105 pipe_surface_format(struct pipe_surface *psurf)
 106 {
 107         if (!psurf)
 108                 return PIPE_FORMAT_NONE;
 109         return psurf->format;
 110 }
 111
 112 #define LOG_DWORDS 0
 113
 114 static inline void emit_marker(struct fd_ringbuffer *ring, int scratch_idx);
 115
 116 static inline void
 117 OUT_RING(struct fd_ringbuffer *ring, uint32_t data)
 118 {
 119         if (LOG_DWORDS) {
 120                 DBG("ring[%p]: OUT_RING   %04x:  %08x", ring,
 121                                 (uint32_t)(ring->cur - ring->last_start), data);
 122         }
 123         *(ring->cur++) = data;
 124 }
 125
 126 /* like OUT_RING() but appends a cmdstream patch point to 'buf' */
 127 static inline void
 128 OUT_RINGP(struct fd_ringbuffer *ring, uint32_t data,
 129                 struct util_dynarray *buf)
 130 {
 131         if (LOG_DWORDS) {
 132                 DBG("ring[%p]: OUT_RINGP  %04x:  %08x", ring,
 133                                 (uint32_t)(ring->cur - ring->last_start), data);
 134         }
 135         util_dynarray_append(buf, struct fd_cs_patch, ((struct fd_cs_patch){
 136                 .cs  = ring->cur++,
 137                 .val = data,
 138         }));
 139 }
 140
 141 static inline void
 142 OUT_RELOC(struct fd_ringbuffer *ring, struct fd_bo *bo,
 143                 uint32_t offset, uint32_t or, int32_t shift)
 144 {
 145         if (LOG_DWORDS) {
 146                 DBG("ring[%p]: OUT_RELOC   %04x:  %p+%u << %d", ring,
 147                                 (uint32_t)(ring->cur - ring->last_start), bo, offset, shift);
 148         }
 149         fd_ringbuffer_reloc(ring, &(struct fd_reloc){
 150                 .bo = bo,
 151                 .flags = FD_RELOC_READ,
 152                 .offset = offset,
 153                 .or = or,
 154                 .shift = shift,
 155         });
 156 }
 157
 158 static inline void
 159 OUT_RELOCW(struct fd_ringbuffer *ring, struct fd_bo *bo,
 160                 uint32_t offset, uint32_t or, int32_t shift)
 161 {
 162         if (LOG_DWORDS) {
 163                 DBG("ring[%p]: OUT_RELOCW  %04x:  %p+%u << %d", ring,
 164                                 (uint32_t)(ring->cur - ring->last_start), bo, offset, shift);
 165         }
 166         fd_ringbuffer_reloc(ring, &(struct fd_reloc){
 167                 .bo = bo,
 168                 .flags = FD_RELOC_READ | FD_RELOC_WRITE,
 169                 .offset = offset,
 170                 .or = or,
 171                 .shift = shift,
 172         });
 173 }
 174
 175 static inline void BEGIN_RING(struct fd_ringbuffer *ring, uint32_t ndwords)
 176 {
 177         if ((ring->cur + ndwords) >= ring->end) {
 178                 /* this probably won't really work if we have multiple tiles..
 179                  * but it is ok for 2d..  we might need different behavior
 180                  * depending on 2d or 3d pipe.
 181                  */
 182                 DBG("uh oh..");
 183         }
 184 }
 185
 186 static inline void
 187 OUT_PKT0(struct fd_ringbuffer *ring, uint16_t regindx, uint16_t cnt)
 188 {
 189         BEGIN_RING(ring, cnt+1);
 190         OUT_RING(ring, CP_TYPE0_PKT | ((cnt-1) << 16) | (regindx & 0x7FFF));
 191 }
 192
 193 static inline void
 194 OUT_PKT3(struct fd_ringbuffer *ring, uint8_t opcode, uint16_t cnt)
 195 {
 196         BEGIN_RING(ring, cnt+1);
 197         OUT_RING(ring, CP_TYPE3_PKT | ((cnt-1) << 16) | ((opcode & 0xFF) << 8));
 198 }
 199
 200 static inline void
 201 OUT_WFI(struct fd_ringbuffer *ring)
 202 {
 203         OUT_PKT3(ring, CP_WAIT_FOR_IDLE, 1);
 204         OUT_RING(ring, 0x00000000);
 205 }
 206
 207 static inline void
 208 OUT_IB(struct fd_ringbuffer *ring, struct fd_ringmarker *start,
 209                 struct fd_ringmarker *end)
 210 {
 211         /* for debug after a lock up, write a unique counter value
 212          * to scratch6 for each IB, to make it easier to match up
 213          * register dumps to cmdstream.  The combination of IB and
 214          * DRAW (scratch7) is enough to "triangulate" the particular
 215          * draw that caused lockup.
 216          */
 217         emit_marker(ring, 6);
 218
 219         OUT_PKT3(ring, CP_INDIRECT_BUFFER_PFD, 2);
 220         fd_ringbuffer_emit_reloc_ring(ring, start, end);
 221         OUT_RING(ring, fd_ringmarker_dwords(start, end));
 222
 223         emit_marker(ring, 6);
 224 }
 225
 226 /* CP_SCRATCH_REG4 is used to hold base address for query results: */
 227 #define HW_QUERY_BASE_REG REG_AXXX_CP_SCRATCH_REG4
 228
 229 static inline void
 230 emit_marker(struct fd_ringbuffer *ring, int scratch_idx)
 231 {
 232         extern unsigned marker_cnt;
 233         unsigned reg = REG_AXXX_CP_SCRATCH_REG0 + scratch_idx;
 234         assert(reg != HW_QUERY_BASE_REG);
 235         if (reg == HW_QUERY_BASE_REG)
 236                 return;
 237         OUT_PKT0(ring, reg, 1);
 238         OUT_RING(ring, ++marker_cnt);
 239 }
 240
 241 /* helper to get numeric value from environment variable..  mostly
 242  * just leaving this here because it is helpful to brute-force figure
 243  * out unknown formats, etc, which blob driver does not support:
 244  */
 245 static inline uint32_t env2u(const char *envvar)
 246 {
 247         char *str = getenv(envvar);
 248         if (str)
 249                 return strtol(str, NULL, 0);
 250         return 0;
 251 }
 252
 253 #endif /* FREEDRENO_UTIL_H_ */