summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorRoland Scheidegger <rscheidegger_lists@hispeed.ch>2005-02-10 19:22:43 +0000
committerRoland Scheidegger <rscheidegger_lists@hispeed.ch>2005-02-10 19:22:43 +0000
commit732cdc5cef25b1ea518fec9c9028d3a9806b690a (patch)
tree397de7888ecb2a12d143de5cf81969fa90769831
parent08790293b13bb4562307309461400dad22c72eaf (diff)
add support for texture micro tiling on radeon/r200. Add support for r100
cube maps (since it also requires a version bump) at the same time.
-rw-r--r--shared-core/radeon_drm.h8
-rw-r--r--shared-core/radeon_drv.h13
-rw-r--r--shared-core/radeon_state.c159
3 files changed, 156 insertions, 24 deletions
diff --git a/shared-core/radeon_drm.h b/shared-core/radeon_drm.h
index 5b9d35b2..adbd6b6b 100644
--- a/shared-core/radeon_drm.h
+++ b/shared-core/radeon_drm.h
@@ -145,7 +145,13 @@
#define RADEON_EMIT_PP_TEX_SIZE_2 75
#define R200_EMIT_RB3D_BLENDCOLOR 76
#define R200_EMIT_TCL_POINT_SPRITE_CNTL 77
-#define RADEON_MAX_STATE_PACKETS 78
+#define RADEON_EMIT_PP_CUBIC_FACES_0 78
+#define RADEON_EMIT_PP_CUBIC_OFFSETS_T0 79
+#define RADEON_EMIT_PP_CUBIC_FACES_1 80
+#define RADEON_EMIT_PP_CUBIC_OFFSETS_T1 81
+#define RADEON_EMIT_PP_CUBIC_FACES_2 82
+#define RADEON_EMIT_PP_CUBIC_OFFSETS_T2 83
+#define RADEON_MAX_STATE_PACKETS 84
/* Commands understood by cmd_buffer ioctl. More can be added but
* obviously these can't be removed or changed:
diff --git a/shared-core/radeon_drv.h b/shared-core/radeon_drv.h
index b840ae6e..9f896d24 100644
--- a/shared-core/radeon_drv.h
+++ b/shared-core/radeon_drv.h
@@ -42,7 +42,7 @@
#define DRIVER_NAME "radeon"
#define DRIVER_DESC "ATI Radeon"
-#define DRIVER_DATE "20050207"
+#define DRIVER_DATE "20050208"
/* Interface history:
*
@@ -82,10 +82,12 @@
* - Add hyperz support, add hyperz flags to clear ioctl.
* 1.14- Add support for color tiling
* - Add R100/R200 surface allocation/free support
+ * 1.15- Add support for texture micro tiling
+ * - Add support for r100 cube maps
*/
#define DRIVER_MAJOR 1
-#define DRIVER_MINOR 14
+#define DRIVER_MINOR 15
#define DRIVER_PATCHLEVEL 0
enum radeon_family {
@@ -800,6 +802,13 @@ extern void radeon_driver_free_filp_priv(drm_device_t * dev,
#define RADEON_PP_TEX_SIZE_1 0x1d0c
#define RADEON_PP_TEX_SIZE_2 0x1d14
+#define RADEON_PP_CUBIC_FACES_0 0x1d24
+#define RADEON_PP_CUBIC_FACES_1 0x1d28
+#define RADEON_PP_CUBIC_FACES_2 0x1d2c
+#define RADEON_PP_CUBIC_OFFSET_T0_0 0x1dd0 /* bits [31:5] */
+#define RADEON_PP_CUBIC_OFFSET_T1_0 0x1e00
+#define RADEON_PP_CUBIC_OFFSET_T2_0 0x1e14
+
#define SE_VAP_CNTL__TCL_ENA_MASK 0x00000001
#define SE_VAP_CNTL__FORCE_W_TO_ONE_MASK 0x00010000
#define SE_VAP_CNTL__VF_MAX_VTX_NUM__SHIFT 0x00000012
diff --git a/shared-core/radeon_state.c b/shared-core/radeon_state.c
index 40ce3cd8..883d1d0e 100644
--- a/shared-core/radeon_state.c
+++ b/shared-core/radeon_state.c
@@ -129,6 +129,22 @@ static __inline__ int radeon_check_and_fixup_packets(drm_radeon_private_t *
break;
}
+ case RADEON_EMIT_PP_CUBIC_OFFSETS_T0:
+ case RADEON_EMIT_PP_CUBIC_OFFSETS_T1:
+ case RADEON_EMIT_PP_CUBIC_OFFSETS_T2:{
+ int i;
+ for (i = 0; i < 5; i++) {
+ if (radeon_check_and_fixup_offset(dev_priv,
+ filp_priv,
+ &data[i])) {
+ DRM_ERROR
+ ("Invalid R100 cubic texture offset\n");
+ return DRM_ERR(EINVAL);
+ }
+ }
+ }
+ break;
+
case RADEON_EMIT_RB3D_COLORPITCH:
case RADEON_EMIT_RE_LINE_PATTERN:
case RADEON_EMIT_SE_LINE_WIDTH:
@@ -190,6 +206,9 @@ static __inline__ int radeon_check_and_fixup_packets(drm_radeon_private_t *
case RADEON_EMIT_PP_TEX_SIZE_2:
case R200_EMIT_RB3D_BLENDCOLOR:
case R200_EMIT_TCL_POINT_SPRITE_CNTL:
+ case RADEON_EMIT_PP_CUBIC_FACES_0:
+ case RADEON_EMIT_PP_CUBIC_FACES_1:
+ case RADEON_EMIT_PP_CUBIC_FACES_2:
/* These packets don't contain memory offsets */
break;
@@ -556,6 +575,13 @@ static struct {
RADEON_PP_TEX_SIZE_2, 2, "RADEON_PP_TEX_SIZE_2"}, {
R200_RB3D_BLENDCOLOR, 3, "R200_RB3D_BLENDCOLOR"}, {
R200_SE_TCL_POINT_SPRITE_CNTL, 1, "R200_SE_TCL_POINT_SPRITE_CNTL"},
+ {
+ RADEON_PP_CUBIC_FACES_0, 1, "RADEON_PP_CUBIC_FACES_0"}, {
+ RADEON_PP_CUBIC_OFFSET_T0_0, 5, "RADEON_PP_CUBIC_OFFSET_T0_0"}, {
+ RADEON_PP_CUBIC_FACES_1, 1, "RADEON_PP_CUBIC_FACES_1"}, {
+ RADEON_PP_CUBIC_OFFSET_T1_0, 5, "RADEON_PP_CUBIC_OFFSET_T1_0"}, {
+ RADEON_PP_CUBIC_FACES_2, 1, "RADEON_PP_CUBIC_FACES_2"}, {
+ RADEON_PP_CUBIC_OFFSET_T2_0, 5, "RADEON_PP_CUBIC_OFFSET_T2_0"},
};
/* ================================================================
@@ -1461,6 +1487,7 @@ static int radeon_cp_dispatch_texture(DRMFILE filp,
int size, dwords, tex_width, blit_width;
u32 height;
int i;
+ u32 texpitch, microtile;
RING_LOCALS;
DRM_GET_PRIV_WITH_RETURN(filp_priv, filp);
@@ -1522,6 +1549,16 @@ static int radeon_cp_dispatch_texture(DRMFILE filp,
DRM_ERROR("invalid texture format %d\n", tex->format);
return DRM_ERR(EINVAL);
}
+ texpitch = tex->pitch;
+ if ((texpitch << 22) & RADEON_DST_TILE_MICRO) {
+ microtile = 1;
+ if (tex_width < 64) {
+ texpitch &= ~(RADEON_DST_TILE_MICRO >> 22);
+ /* we got tiled coordinates, untile them */
+ image->x *= 2;
+ }
+ }
+ else microtile = 0;
DRM_DEBUG("tex=%dx%d blit=%d\n", tex_width, tex->height, blit_width);
@@ -1574,7 +1611,7 @@ static int radeon_cp_dispatch_texture(DRMFILE filp,
RADEON_GMC_CLR_CMP_CNTL_DIS |
RADEON_GMC_WR_MSK_DIS);
- buffer[2] = (tex->pitch << 22) | (tex->offset >> 10);
+ buffer[2] = (texpitch << 22) | (tex->offset >> 10);
buffer[3] = 0xffffffff;
buffer[4] = 0xffffffff;
buffer[5] = (image->y << 16) | image->x;
@@ -1582,29 +1619,109 @@ static int radeon_cp_dispatch_texture(DRMFILE filp,
buffer[7] = dwords;
buffer += 8;
- if (tex_width >= 32) {
- /* Texture image width is larger than the minimum, so we
- * can upload it directly.
- */
- if (DRM_COPY_FROM_USER(buffer, data,
- dwords * sizeof(u32))) {
- DRM_ERROR("EFAULT on data, %d dwords\n",
- dwords);
- return DRM_ERR(EFAULT);
+ if (microtile) {
+ /* texture micro tiling in use, minimum texture width is thus 16 bytes.
+ however, we cannot use blitter directly for texture width < 64 bytes,
+ since minimum tex pitch is 64 bytes and we need this to match
+ the texture width, otherwise the blitter will tile it wrong.
+ Thus, tiling manually in this case. Additionally, need to special
+ case tex height = 1, since our actual image will have height 2
+ and we need to ensure we don't read beyond the texture size
+ from user space. */
+ if (tex->height == 1) {
+ if (tex_width >= 64 || tex_width <= 16) {
+ if (DRM_COPY_FROM_USER(buffer, data,
+ tex_width * sizeof(u32))) {
+ DRM_ERROR("EFAULT on pad, %d bytes\n",
+ tex_width);
+ return DRM_ERR(EFAULT);
+ }
+ } else if (tex_width == 32) {
+ if (DRM_COPY_FROM_USER(buffer, data, 16)) {
+ DRM_ERROR("EFAULT on pad, %d bytes\n",
+ tex_width);
+ return DRM_ERR(EFAULT);
+ }
+ if (DRM_COPY_FROM_USER(buffer + 8, data + 16, 16)) {
+ DRM_ERROR("EFAULT on pad, %d bytes\n",
+ tex_width);
+ return DRM_ERR(EFAULT);
+ }
+ }
+ } else if (tex_width >= 64 || tex_width == 16) {
+ if (DRM_COPY_FROM_USER(buffer, data,
+ dwords * sizeof(u32))) {
+ DRM_ERROR("EFAULT on data, %d dwords\n",
+ dwords);
+ return DRM_ERR(EFAULT);
+ }
+ } else if (tex_width < 16) {
+ for (i = 0; i < tex->height; i++) {
+ if (DRM_COPY_FROM_USER(buffer, data, tex_width)) {
+ DRM_ERROR("EFAULT on pad, %d bytes\n",
+ tex_width);
+ return DRM_ERR(EFAULT);
+ }
+ buffer += 4;
+ data += tex_width;
+ }
+ } else if (tex_width == 32) {
+ /* TODO: make sure this works when not fitting in one buffer
+ (i.e. 32bytes x 2048...) */
+ for (i = 0; i < tex->height; i += 2) {
+ if (DRM_COPY_FROM_USER(buffer, data, 16)) {
+ DRM_ERROR("EFAULT on pad, %d bytes\n",
+ tex_width);
+ return DRM_ERR(EFAULT);
+ }
+ data += 16;
+ if (DRM_COPY_FROM_USER(buffer + 8, data, 16)) {
+ DRM_ERROR("EFAULT on pad, %d bytes\n",
+ tex_width);
+ return DRM_ERR(EFAULT);
+ }
+ data += 16;
+ if (DRM_COPY_FROM_USER(buffer + 4, data, 16)) {
+ DRM_ERROR("EFAULT on pad, %d bytes\n",
+ tex_width);
+ return DRM_ERR(EFAULT);
+ }
+ data += 16;
+ if (DRM_COPY_FROM_USER(buffer + 12, data, 16)) {
+ DRM_ERROR("EFAULT on pad, %d bytes\n",
+ tex_width);
+ return DRM_ERR(EFAULT);
+ }
+ data += 16;
+ buffer += 16;
+ }
}
- } else {
- /* Texture image width is less than the minimum, so we
- * need to pad out each image scanline to the minimum
- * width.
- */
- for (i = 0; i < tex->height; i++) {
- if (DRM_COPY_FROM_USER(buffer, data, tex_width)) {
- DRM_ERROR("EFAULT on pad, %d bytes\n",
- tex_width);
+ }
+ else {
+ if (tex_width >= 32) {
+ /* Texture image width is larger than the minimum, so we
+ * can upload it directly.
+ */
+ if (DRM_COPY_FROM_USER(buffer, data,
+ dwords * sizeof(u32))) {
+ DRM_ERROR("EFAULT on data, %d dwords\n",
+ dwords);
return DRM_ERR(EFAULT);
}
- buffer += 8;
- data += tex_width;
+ } else {
+ /* Texture image width is less than the minimum, so we
+ * need to pad out each image scanline to the minimum
+ * width.
+ */
+ for (i = 0; i < tex->height; i++) {
+ if (DRM_COPY_FROM_USER(buffer, data, tex_width)) {
+ DRM_ERROR("EFAULT on pad, %d bytes\n",
+ tex_width);
+ return DRM_ERR(EFAULT);
+ }
+ buffer += 8;
+ data += tex_width;
+ }
}
}