src/gallium/drivers/nv30/nv30_fragprog.c

   1 /*
   2  * Copyright 2012 Red Hat Inc.
   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 shall be included in
  12  * all copies or substantial portions of the Software.
  13  *
  14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  17  * THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
  18  * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF
  19  * OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  20  * SOFTWARE.
  21  *
  22  * Authors: Ben Skeggs
  23  *
  24  */
  25
  26 #include "tgsi/tgsi_parse.h"
  27
  28 #include "nouveau/nv_object.xml.h"
  29 #include "nv30-40_3d.xml.h"
  30 #include "nv30_context.h"
  31 #include "nvfx_shader.h"
  32
  33 static void
  34 nv30_fragprog_upload(struct nv30_context *nv30)
  35 {
  36    struct nouveau_context *nv = &nv30->base;
  37    struct nv30_fragprog *fp = nv30->fragprog.program;
  38    struct pipe_context *pipe = &nv30->base.pipe;
  39    struct pipe_transfer *transfer;
  40    uint32_t *map;
  41    int i; (void)i;
  42
  43    if (unlikely(!fp->buffer)) {
  44       fp->buffer = pipe_buffer_create(pipe->screen, 0, 0, fp->insn_len * 4);
  45    }
  46
  47    map = pipe_buffer_map(pipe, fp->buffer, PIPE_TRANSFER_WRITE, &transfer);
  48 #ifndef PIPE_ARCH_BIG_ENDIAN
  49    memcpy(map, fp->insn, fp->insn_len * 4);
  50 #else
  51    for (i = 0; i < fp->insn_len; i++)
  52       *map++ = (fp->insn[i] >> 16) | (fp->insn[i] << 16);
  53 #endif
  54    pipe_buffer_unmap(pipe, transfer);
  55
  56    if (nv04_resource(fp->buffer)->domain != NOUVEAU_BO_VRAM)
  57       nouveau_buffer_migrate(nv, nv04_resource(fp->buffer), NOUVEAU_BO_VRAM);
  58 }
  59
  60 void
  61 nv30_fragprog_validate(struct nv30_context *nv30)
  62 {
  63    struct nouveau_pushbuf *push = nv30->base.pushbuf;
  64    struct nouveau_object *eng3d = nv30->screen->eng3d;
  65    struct nv30_fragprog *fp = nv30->fragprog.program;
  66    boolean upload = FALSE;
  67    int i;
  68
  69    if (!fp->translated) {
  70       _nvfx_fragprog_translate(nv30, fp, FALSE);
  71       if (!fp->translated)
  72          return;
  73
  74       upload = TRUE;
  75    }
  76
  77    /* update constants, also needs to be done on every fp switch as we
  78     * have no idea whether the constbuf changed in the meantime
  79     */
  80    if (nv30->fragprog.constbuf) {
  81       struct pipe_resource *constbuf = nv30->fragprog.constbuf;
  82       uint32_t *cbuf = (uint32_t *)nv04_resource(constbuf)->data;
  83
  84       for (i = 0; i < fp->nr_consts; i++) {
  85          unsigned off = fp->consts[i].offset;
  86          unsigned idx = fp->consts[i].index * 4;
  87
  88          if (!memcmp(&fp->insn[off], &cbuf[idx], 4 * 4))
  89             continue;
  90          memcpy(&fp->insn[off], &cbuf[idx], 4 * 4);
  91          upload = TRUE;
  92       }
  93    }
  94
  95    if (upload)
  96       nv30_fragprog_upload(nv30);
  97
  98    /* FP_ACTIVE_PROGRAM needs to be done again even if only the consts
  99     * were updated.  TEX_CACHE_CTL magic is not enough to convince the
 100     * GPU that it should re-read the fragprog from VRAM... sigh.
 101     */
 102    if (nv30->state.fragprog != fp || upload) {
 103       struct nv04_resource *r = nv04_resource(fp->buffer);
 104
 105       if (!PUSH_SPACE(push, 8))
 106          return;
 107       PUSH_RESET(push, BUFCTX_FRAGPROG);
 108
 109       BEGIN_NV04(push, NV30_3D(FP_ACTIVE_PROGRAM), 1);
 110       PUSH_RESRC(push, NV30_3D(FP_ACTIVE_PROGRAM), BUFCTX_FRAGPROG, r, 0,
 111                        NOUVEAU_BO_LOW | NOUVEAU_BO_RD | NOUVEAU_BO_OR,
 112                        NV30_3D_FP_ACTIVE_PROGRAM_DMA0,
 113                        NV30_3D_FP_ACTIVE_PROGRAM_DMA1);
 114       BEGIN_NV04(push, NV30_3D(FP_CONTROL), 1);
 115       PUSH_DATA (push, fp->fp_control);
 116       if (eng3d->oclass < NV40_3D_CLASS) {
 117          BEGIN_NV04(push, NV30_3D(FP_REG_CONTROL), 1);
 118          PUSH_DATA (push, 0x00010004);
 119          BEGIN_NV04(push, NV30_3D(TEX_UNITS_ENABLE), 1);
 120          PUSH_DATA (push, fp->texcoords);
 121       } else {
 122          BEGIN_NV04(push, SUBC_3D(0x0b40), 1);
 123          PUSH_DATA (push, 0x00000000);
 124       }
 125
 126       nv30->state.fragprog = fp;
 127    }
 128 }
 129
 130 static void *
 131 nv30_fp_state_create(struct pipe_context *pipe,
 132                      const struct pipe_shader_state *cso)
 133 {
 134    struct nv30_fragprog *fp = CALLOC_STRUCT(nv30_fragprog);
 135    if (!fp)
 136       return NULL;
 137
 138    fp->pipe.tokens = tgsi_dup_tokens(cso->tokens);
 139    tgsi_scan_shader(fp->pipe.tokens, &fp->info);
 140    return fp;
 141 }
 142
 143 static void
 144 nv30_fp_state_delete(struct pipe_context *pipe, void *hwcso)
 145 {
 146    struct nv30_fragprog *fp = hwcso;
 147
 148    pipe_resource_reference(&fp->buffer, NULL);
 149
 150    FREE((void *)fp->pipe.tokens);
 151    FREE(fp->insn);
 152    FREE(fp);
 153 }
 154
 155 static void
 156 nv30_fp_state_bind(struct pipe_context *pipe, void *hwcso)
 157 {
 158    struct nv30_context *nv30 = nv30_context(pipe);
 159
 160    nv30->fragprog.program = hwcso;
 161    nv30->dirty |= NV30_NEW_FRAGPROG;
 162 }
 163
 164 void
 165 nv30_fragprog_init(struct pipe_context *pipe)
 166 {
 167    pipe->create_fs_state = nv30_fp_state_create;
 168    pipe->bind_fs_state = nv30_fp_state_bind;
 169    pipe->delete_fs_state = nv30_fp_state_delete;
 170 }