workbench/devs/monitors/IntelGMA/3D_blitter.c

   1 /*
   2     Copyright © 2012-2017, The AROS Development Team. All rights reserved.
   3     $Id$
   4 */
   5
   6 /*
   7  * Copyright © 2011 Intel Corporation
   8  *
   9  * Permission is hereby granted, free of charge, to any person obtaining a
  10  * copy of this software and associated documentation files (the "Software"),
  11  * to deal in the Software without restriction, including without limitation
  12  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  13  * and/or sell copies of the Software, and to permit persons to whom the
  14  * Software is furnished to do so, subject to the following conditions:
  15  *
  16  * The above copyright notice and this permission notice (including the next
  17  * paragraph) shall be included in all copies or substantial portions of the
  18  * Software.
  19  *
  20  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  21  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  23  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  24  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  25  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  26  * IN THE SOFTWARE.
  27  *
  28  * Authors:
  29  *    Chris Wilson (intel-gpu-tools)
  30  *    2012-2017, The AROS Development Team.
  31  */
  32
  33 #include <proto/exec.h>
  34 #include <hidd/gfx.h>
  35 #include <aros/debug.h>
  36
  37 #include "intelgma_hidd.h"
  38 #include "intelG45_regs.h"
  39
  40 //#include "i915/i915_reg.h" // crashes !?
  41 #include "i915_reg.h"
  42
  43 #include "i915_3d.h"
  44
  45
  46 extern struct g45staticdata sd;
  47 #define sd ((struct g45staticdata*)&(sd))
  48
  49
  50 static inline uint32_t pack_float(float f)
  51 {
  52     union {
  53         uint32_t dw;
  54         float f;
  55     } u;
  56     u.f = f;
  57     return u.dw;
  58 }
  59
  60 #define IS_915G(Id) ((Id) == 0x2582 || \
  61                      (Id) == 0x258a)
  62 #define IS_915GM(Id) ((Id) == 0x2592)
  63
  64 #define IS_915(Id) (IS_915G(Id) || \
  65                     IS_915GM(Id))
  66
  67 #define IS_945G(Id)  ((Id) == 0x2772)
  68 #define IS_945GM(Id) ((Id) == 0x27A2 || \
  69                       (Id) == 0x27AE)
  70
  71 #define IS_945(Id) (IS_945G(Id) ||  \
  72                     IS_945GM(Id) || \
  73                     IS_G33(Id) ||   \
  74                     IS_PINEVIEW(Id))
  75
  76 #define IS_G33(Id) ((Id) == 0x29C2 || \
  77                     (Id) == 0x29B2 || \
  78                     (Id) == 0x29D2)
  79
  80 #define IS_PINEVIEW(Id) ((Id) == 0xa001 || \
  81                          (Id) == 0xa011)
  82
  83 #define IS_GEN3(Id) (IS_915(Id) || \
  84                      IS_945(Id) || \
  85                      IS_G33(Id) || \
  86                      IS_PINEVIEW(Id))
  87
  88 BOOL copybox3d_supported()
  89 {
  90     // supported chipsets
  91     if( IS_GEN3( sd->ProductID ) )
  92     {
  93         return TRUE;
  94     }
  95     return FALSE;
  96 }
  97
  98 BOOL copybox3d( GMABitMap_t *bm_dst, GMABitMap_t *bm_src,
  99                ULONG dst_x,ULONG dst_y,ULONG dst_width, ULONG dst_height,
 100                ULONG src_x,ULONG src_y,ULONG src_width, ULONG src_height )
 101 {
 102     uint32_t dst_format;
 103     uint32_t src_format;
 104
 105     if( !copybox3d_supported() )
 106     {
 107         return FALSE;
 108     }
 109
 110     // buffers in gfx memory ?
 111     if( ! (bm_src->fbgfx && bm_src->fbgfx) )
 112     {
 113         return FALSE;
 114     }
 115
 116     // Max texture size, src and dst must be differend (at least if overlaps)
 117     if( bm_src->pitch/4 > 2048 || bm_src->height > 2048 || bm_dst == bm_src )
 118     {
 119         return FALSE;
 120     }
 121
 122     // src pitch must be long aligmented.
 123     if( bm_src->pitch & 0x3 )
 124     {
 125         bug("[GMA] copybox3d: ERROR bm_src->pitch=%d/n",bm_src->pitch);
 126         return FALSE;
 127     }
 128
 129     if(bm_src->bpp == 4)
 130     {
 131         src_format = MAPSURF_32BIT | MT_32BIT_ARGB8888;
 132     }
 133     else if(bm_src->bpp == 2)
 134     {
 135         src_format = MAPSURF_16BIT | MT_16BIT_RGB565;
 136     }
 137     else
 138     {
 139         bug("[GMA] copybox3d: ERROR src_bpp=%d/n",bm_src->bpp);
 140         return FALSE;
 141     }
 142
 143     if(bm_dst->bpp == 4)
 144     {
 145         dst_format = COLR_BUF_ARGB8888;
 146     }
 147     else if(bm_dst->bpp == 2)
 148     {
 149         dst_format = COLR_BUF_RGB565;
 150     }
 151     else
 152     {
 153         bug("[GMA] copybox3d: ERROR dst_bpp=%d/n",bm_dst->bpp);
 154         return FALSE;
 155     }
 156
 157     LOCK_HW
 158     START_RING(72);
 159
 160     /* invariant state */
 161     OUT_RING( _3DSTATE_AA_CMD |
 162         AA_LINE_ECAAR_WIDTH_ENABLE |
 163         AA_LINE_ECAAR_WIDTH_1_0 |
 164         AA_LINE_REGION_WIDTH_ENABLE | AA_LINE_REGION_WIDTH_1_0 );
 165     OUT_RING( _3DSTATE_INDEPENDENT_ALPHA_BLEND_CMD |
 166         IAB_MODIFY_ENABLE |
 167         IAB_MODIFY_FUNC | (BLENDFUNC_ADD << IAB_FUNC_SHIFT) |
 168         IAB_MODIFY_SRC_FACTOR | (BLENDFACT_ONE <<
 169                      IAB_SRC_FACTOR_SHIFT) |
 170         IAB_MODIFY_DST_FACTOR | (BLENDFACT_ZERO <<
 171                      IAB_DST_FACTOR_SHIFT) );
 172     OUT_RING( _3DSTATE_DFLT_DIFFUSE_CMD );
 173     OUT_RING( 0 );
 174     OUT_RING( _3DSTATE_DFLT_SPEC_CMD );
 175     OUT_RING( 0 );
 176     OUT_RING( _3DSTATE_DFLT_Z_CMD );
 177     OUT_RING( 0 );
 178     OUT_RING( _3DSTATE_COORD_SET_BINDINGS |
 179         CSB_TCB(0, 0) |
 180         CSB_TCB(1, 1) |
 181         CSB_TCB(2, 2) |
 182         CSB_TCB(3, 3) |
 183         CSB_TCB(4, 4) |
 184         CSB_TCB(5, 5) | CSB_TCB(6, 6) | CSB_TCB(7, 7) );
 185     OUT_RING( _3DSTATE_RASTER_RULES_CMD |
 186         ENABLE_POINT_RASTER_RULE |
 187         OGL_POINT_RASTER_RULE |
 188         ENABLE_LINE_STRIP_PROVOKE_VRTX |
 189         ENABLE_TRI_FAN_PROVOKE_VRTX |
 190         LINE_STRIP_PROVOKE_VRTX(1) |
 191         TRI_FAN_PROVOKE_VRTX(2) | ENABLE_TEXKILL_3D_4D | TEXKILL_4D );
 192     OUT_RING( _3DSTATE_MODES_4_CMD |
 193         ENABLE_LOGIC_OP_FUNC | LOGIC_OP_FUNC(LOGICOP_COPY) |
 194         ENABLE_STENCIL_WRITE_MASK | STENCIL_WRITE_MASK(0xff) |
 195         ENABLE_STENCIL_TEST_MASK | STENCIL_TEST_MASK(0xff) );
 196     OUT_RING( _3DSTATE_LOAD_STATE_IMMEDIATE_1 | I1_LOAD_S(3) | I1_LOAD_S(4) | I1_LOAD_S(5) | 2 );
 197     OUT_RING( 0x00000000 );    /* Disable texture coordinate wrap-shortest */
 198     OUT_RING( (1 << S4_POINT_WIDTH_SHIFT) |
 199         S4_LINE_WIDTH_ONE |
 200         S4_CULLMODE_NONE |
 201         S4_VFMT_XY );
 202     OUT_RING( 0x00000000 );    /* Stencil. */
 203     OUT_RING( _3DSTATE_SCISSOR_ENABLE_CMD | DISABLE_SCISSOR_RECT );
 204     OUT_RING( _3DSTATE_SCISSOR_RECT_0_CMD );
 205     OUT_RING( 0 );
 206     OUT_RING( 0 );
 207     OUT_RING( _3DSTATE_DEPTH_SUBRECT_DISABLE );
 208     OUT_RING( _3DSTATE_LOAD_INDIRECT | 0 );    /* disable indirect state */
 209     OUT_RING( 0 );
 210     OUT_RING( _3DSTATE_STIPPLE );
 211     OUT_RING( 0x00000000 );
 212     OUT_RING( _3DSTATE_BACKFACE_STENCIL_OPS | BFO_ENABLE_STENCIL_TWO_SIDE | 0 );
 213
 214     /* samler state */
 215 #define TEX_COUNT 1
 216     OUT_RING( _3DSTATE_MAP_STATE | (3 * TEX_COUNT) );
 217     OUT_RING( (1 << TEX_COUNT) - 1 );
 218
 219     // Source buffer
 220     OUT_RING( bm_src->framebuffer );
 221     OUT_RING( src_format |
 222         (bm_src->height - 1) << MS3_HEIGHT_SHIFT |
 223         (bm_src->pitch/bm_src->bpp - 1)  << MS3_WIDTH_SHIFT );
 224     OUT_RING( (bm_src->pitch/4 -1) << MS4_PITCH_SHIFT );
 225
 226     OUT_RING( _3DSTATE_SAMPLER_STATE | (3 * TEX_COUNT) );
 227     OUT_RING( (1 << TEX_COUNT) - 1 );
 228     OUT_RING( MIPFILTER_NONE << SS2_MIP_FILTER_SHIFT |
 229         FILTER_NEAREST << SS2_MAG_FILTER_SHIFT |
 230         FILTER_NEAREST << SS2_MIN_FILTER_SHIFT );
 231     OUT_RING( TEXCOORDMODE_WRAP << SS3_TCX_ADDR_MODE_SHIFT |
 232         TEXCOORDMODE_WRAP << SS3_TCY_ADDR_MODE_SHIFT |
 233         0 << SS3_TEXTUREMAP_INDEX_SHIFT );
 234     OUT_RING( 0x00000000 );
 235
 236     /* render target state */
 237
 238     // Destination buffer
 239     OUT_RING( _3DSTATE_BUF_INFO_CMD );
 240     OUT_RING( BUF_3D_ID_COLOR_BACK | bm_dst->pitch );
 241     OUT_RING( bm_dst->framebuffer );
 242     OUT_RING( _3DSTATE_DST_BUF_VARS_CMD );
 243     OUT_RING( dst_format |
 244         DSTORG_HORT_BIAS(0x8) |
 245         DSTORG_VERT_BIAS(0x8) );
 246
 247     /* draw rect is unconditional */
 248     OUT_RING( _3DSTATE_DRAW_RECT_CMD );
 249
 250     OUT_RING( 0x00000000 );
 251     OUT_RING( 0x00000000 );    // ymin, xmin
 252     OUT_RING( DRAW_YMAX(dst_y + dst_height - 1) |
 253               DRAW_XMAX(dst_x + dst_width - 1) );
 254
 255     /* yorig, xorig (relate to color buffer?) */
 256     OUT_RING( 0x00000000 );
 257
 258     /* texfmt */
 259     OUT_RING( _3DSTATE_LOAD_STATE_IMMEDIATE_1 | I1_LOAD_S(1) | I1_LOAD_S(2) | I1_LOAD_S(6) | 2 );
 260     OUT_RING( (4 << S1_VERTEX_WIDTH_SHIFT) | (4 << S1_VERTEX_PITCH_SHIFT) );
 261     OUT_RING( ~S2_TEXCOORD_FMT(0, TEXCOORDFMT_NOT_PRESENT) |
 262         S2_TEXCOORD_FMT(0, TEXCOORDFMT_2D) );
 263     OUT_RING( S6_CBUF_BLEND_ENABLE | S6_COLOR_WRITE_ENABLE |
 264         BLENDFUNC_ADD << S6_CBUF_BLEND_FUNC_SHIFT |
 265         BLENDFACT_ONE << S6_CBUF_SRC_BLEND_FACT_SHIFT |
 266         BLENDFACT_ZERO << S6_CBUF_DST_BLEND_FACT_SHIFT );
 267
 268     /* pixel shader */
 269     OUT_RING( _3DSTATE_PIXEL_SHADER_PROGRAM | (1 + 3*3 - 2) );
 270     /* decl FS_T0 */
 271     OUT_RING( D0_DCL |
 272         REG_TYPE(FS_T0) << D0_TYPE_SHIFT |
 273         REG_NR(FS_T0) << D0_NR_SHIFT |
 274         ((REG_TYPE(FS_T0) != REG_TYPE_S) ? D0_CHANNEL_ALL : 0) );
 275     OUT_RING( 0 );
 276     OUT_RING( 0 );
 277     /* decl FS_S0 */
 278     OUT_RING( D0_DCL |
 279         (REG_TYPE(FS_S0) << D0_TYPE_SHIFT) |
 280         (REG_NR(FS_S0) << D0_NR_SHIFT) |
 281         ((REG_TYPE(FS_S0) != REG_TYPE_S) ? D0_CHANNEL_ALL : 0) );
 282     OUT_RING( 0 );
 283     OUT_RING( 0 );
 284     /* texld(FS_OC, FS_S0, FS_T0 */
 285     OUT_RING( T0_TEXLD |
 286         (REG_TYPE(FS_OC) << T0_DEST_TYPE_SHIFT) |
 287         (REG_NR(FS_OC) << T0_DEST_NR_SHIFT) |
 288         (REG_NR(FS_S0) << T0_SAMPLER_NR_SHIFT) );
 289     OUT_RING( (REG_TYPE(FS_T0) << T1_ADDRESS_REG_TYPE_SHIFT) |
 290         (REG_NR(FS_T0) << T1_ADDRESS_REG_NR_SHIFT) );
 291     OUT_RING( 0 );
 292
 293     // rectangle
 294     // 3--x
 295     // |  |
 296     // 2--1
 297     OUT_RING( PRIM3D_RECTLIST | (3*4 - 1) );
 298     OUT_RING( pack_float( dst_x + dst_width) );
 299     OUT_RING( pack_float( dst_y + dst_height) );
 300     OUT_RING( pack_float(src_x + src_width) );
 301     OUT_RING( pack_float(src_y + src_height) );
 302
 303     OUT_RING( pack_float( dst_x + 0 ) );
 304     OUT_RING( pack_float( dst_y +dst_height) );
 305     OUT_RING( pack_float(src_x + 0) );
 306     OUT_RING( pack_float(src_y + src_height) );
 307
 308     OUT_RING( pack_float( dst_x + 0 ) );
 309     OUT_RING( pack_float( dst_y + 0 ) );
 310     OUT_RING( pack_float(src_x + 0) );
 311     OUT_RING( pack_float(src_y + 0) );
 312
 313     ADVANCE_RING();
 314     DO_FLUSH();
 315     UNLOCK_HW
 316
 317     return TRUE;
 318 }