radeon_state.c

linux-2.6.15.6

	if (dirty & RADEON_UPLOAD_TEX2) {
		if (radeon_check_and_fixup_offset(dev_priv, filp_priv,
						  &tex[2].pp_txoffset)) {
			DRM_ERROR("Invalid texture offset for unit 2\n");
			return DRM_ERR(EINVAL);
		}

		BEGIN_RING(9);
		OUT_RING(CP_PACKET0(RADEON_PP_TXFILTER_2, 5));
		OUT_RING(tex[2].pp_txfilter);
		OUT_RING(tex[2].pp_txformat);
		OUT_RING(tex[2].pp_txoffset);
		OUT_RING(tex[2].pp_txcblend);
		OUT_RING(tex[2].pp_txablend);
		OUT_RING(tex[2].pp_tfactor);
		OUT_RING(CP_PACKET0(RADEON_PP_BORDER_COLOR_2, 0));
		OUT_RING(tex[2].pp_border_color);
		ADVANCE_RING();
	}

	return 0;
}

/* Emit 1.2 state
 */
static int radeon_emit_state2(drm_radeon_private_t * dev_priv,
			      drm_file_t * filp_priv,
			      drm_radeon_state_t * state)
{
	RING_LOCALS;

	if (state->dirty & RADEON_UPLOAD_ZBIAS) {
		BEGIN_RING(3);
		OUT_RING(CP_PACKET0(RADEON_SE_ZBIAS_FACTOR, 1));
		OUT_RING(state->context2.se_zbias_factor);
		OUT_RING(state->context2.se_zbias_constant);
		ADVANCE_RING();
	}

	return radeon_emit_state(dev_priv, filp_priv, &state->context,
				 state->tex, state->dirty);
}

/* New (1.3) state mechanism.  3 commands (packet, scalar, vector) in
 * 1.3 cmdbuffers allow all previous state to be updated as well as
 * the tcl scalar and vector areas.
 */
static struct {
	int start;
	int len;
	const char *name;
} packet[RADEON_MAX_STATE_PACKETS] = {
	{RADEON_PP_MISC, 7, "RADEON_PP_MISC"},
	{RADEON_PP_CNTL, 3, "RADEON_PP_CNTL"},
	{RADEON_RB3D_COLORPITCH, 1, "RADEON_RB3D_COLORPITCH"},
	{RADEON_RE_LINE_PATTERN, 2, "RADEON_RE_LINE_PATTERN"},
	{RADEON_SE_LINE_WIDTH, 1, "RADEON_SE_LINE_WIDTH"},
	{RADEON_PP_LUM_MATRIX, 1, "RADEON_PP_LUM_MATRIX"},
	{RADEON_PP_ROT_MATRIX_0, 2, "RADEON_PP_ROT_MATRIX_0"},
	{RADEON_RB3D_STENCILREFMASK, 3, "RADEON_RB3D_STENCILREFMASK"},
	{RADEON_SE_VPORT_XSCALE, 6, "RADEON_SE_VPORT_XSCALE"},
	{RADEON_SE_CNTL, 2, "RADEON_SE_CNTL"},
	{RADEON_SE_CNTL_STATUS, 1, "RADEON_SE_CNTL_STATUS"},
	{RADEON_RE_MISC, 1, "RADEON_RE_MISC"},
	{RADEON_PP_TXFILTER_0, 6, "RADEON_PP_TXFILTER_0"},
	{RADEON_PP_BORDER_COLOR_0, 1, "RADEON_PP_BORDER_COLOR_0"},
	{RADEON_PP_TXFILTER_1, 6, "RADEON_PP_TXFILTER_1"},
	{RADEON_PP_BORDER_COLOR_1, 1, "RADEON_PP_BORDER_COLOR_1"},
	{RADEON_PP_TXFILTER_2, 6, "RADEON_PP_TXFILTER_2"},
	{RADEON_PP_BORDER_COLOR_2, 1, "RADEON_PP_BORDER_COLOR_2"},
	{RADEON_SE_ZBIAS_FACTOR, 2, "RADEON_SE_ZBIAS_FACTOR"},
	{RADEON_SE_TCL_OUTPUT_VTX_FMT, 11, "RADEON_SE_TCL_OUTPUT_VTX_FMT"},
	{RADEON_SE_TCL_MATERIAL_EMMISSIVE_RED, 17,
		    "RADEON_SE_TCL_MATERIAL_EMMISSIVE_RED"},
	{R200_PP_TXCBLEND_0, 4, "R200_PP_TXCBLEND_0"},
	{R200_PP_TXCBLEND_1, 4, "R200_PP_TXCBLEND_1"},
	{R200_PP_TXCBLEND_2, 4, "R200_PP_TXCBLEND_2"},
	{R200_PP_TXCBLEND_3, 4, "R200_PP_TXCBLEND_3"},
	{R200_PP_TXCBLEND_4, 4, "R200_PP_TXCBLEND_4"},
	{R200_PP_TXCBLEND_5, 4, "R200_PP_TXCBLEND_5"},
	{R200_PP_TXCBLEND_6, 4, "R200_PP_TXCBLEND_6"},
	{R200_PP_TXCBLEND_7, 4, "R200_PP_TXCBLEND_7"},
	{R200_SE_TCL_LIGHT_MODEL_CTL_0, 6, "R200_SE_TCL_LIGHT_MODEL_CTL_0"},
	{R200_PP_TFACTOR_0, 6, "R200_PP_TFACTOR_0"},
	{R200_SE_VTX_FMT_0, 4, "R200_SE_VTX_FMT_0"},
	{R200_SE_VAP_CNTL, 1, "R200_SE_VAP_CNTL"},
	{R200_SE_TCL_MATRIX_SEL_0, 5, "R200_SE_TCL_MATRIX_SEL_0"},
	{R200_SE_TCL_TEX_PROC_CTL_2, 5, "R200_SE_TCL_TEX_PROC_CTL_2"},
	{R200_SE_TCL_UCP_VERT_BLEND_CTL, 1, "R200_SE_TCL_UCP_VERT_BLEND_CTL"},
	{R200_PP_TXFILTER_0, 6, "R200_PP_TXFILTER_0"},
	{R200_PP_TXFILTER_1, 6, "R200_PP_TXFILTER_1"},
	{R200_PP_TXFILTER_2, 6, "R200_PP_TXFILTER_2"},
	{R200_PP_TXFILTER_3, 6, "R200_PP_TXFILTER_3"},
	{R200_PP_TXFILTER_4, 6, "R200_PP_TXFILTER_4"},
	{R200_PP_TXFILTER_5, 6, "R200_PP_TXFILTER_5"},
	{R200_PP_TXOFFSET_0, 1, "R200_PP_TXOFFSET_0"},
	{R200_PP_TXOFFSET_1, 1, "R200_PP_TXOFFSET_1"},
	{R200_PP_TXOFFSET_2, 1, "R200_PP_TXOFFSET_2"},
	{R200_PP_TXOFFSET_3, 1, "R200_PP_TXOFFSET_3"},
	{R200_PP_TXOFFSET_4, 1, "R200_PP_TXOFFSET_4"},
	{R200_PP_TXOFFSET_5, 1, "R200_PP_TXOFFSET_5"},
	{R200_SE_VTE_CNTL, 1, "R200_SE_VTE_CNTL"},
	{R200_SE_TCL_OUTPUT_VTX_COMP_SEL, 1, "R200_SE_TCL_OUTPUT_VTX_COMP_SEL"},
	{R200_PP_TAM_DEBUG3, 1, "R200_PP_TAM_DEBUG3"},
	{R200_PP_CNTL_X, 1, "R200_PP_CNTL_X"},
	{R200_RB3D_DEPTHXY_OFFSET, 1, "R200_RB3D_DEPTHXY_OFFSET"},
	{R200_RE_AUX_SCISSOR_CNTL, 1, "R200_RE_AUX_SCISSOR_CNTL"},
	{R200_RE_SCISSOR_TL_0, 2, "R200_RE_SCISSOR_TL_0"},
	{R200_RE_SCISSOR_TL_1, 2, "R200_RE_SCISSOR_TL_1"},
	{R200_RE_SCISSOR_TL_2, 2, "R200_RE_SCISSOR_TL_2"},
	{R200_SE_VAP_CNTL_STATUS, 1, "R200_SE_VAP_CNTL_STATUS"},
	{R200_SE_VTX_STATE_CNTL, 1, "R200_SE_VTX_STATE_CNTL"},
	{R200_RE_POINTSIZE, 1, "R200_RE_POINTSIZE"},
	{R200_SE_TCL_INPUT_VTX_VECTOR_ADDR_0, 4,
		    "R200_SE_TCL_INPUT_VTX_VECTOR_ADDR_0"},
	{R200_PP_CUBIC_FACES_0, 1, "R200_PP_CUBIC_FACES_0"},	/* 61 */
	{R200_PP_CUBIC_OFFSET_F1_0, 5, "R200_PP_CUBIC_OFFSET_F1_0"},	/* 62 */
	{R200_PP_CUBIC_FACES_1, 1, "R200_PP_CUBIC_FACES_1"},
	{R200_PP_CUBIC_OFFSET_F1_1, 5, "R200_PP_CUBIC_OFFSET_F1_1"},
	{R200_PP_CUBIC_FACES_2, 1, "R200_PP_CUBIC_FACES_2"},
	{R200_PP_CUBIC_OFFSET_F1_2, 5, "R200_PP_CUBIC_OFFSET_F1_2"},
	{R200_PP_CUBIC_FACES_3, 1, "R200_PP_CUBIC_FACES_3"},
	{R200_PP_CUBIC_OFFSET_F1_3, 5, "R200_PP_CUBIC_OFFSET_F1_3"},
	{R200_PP_CUBIC_FACES_4, 1, "R200_PP_CUBIC_FACES_4"},
	{R200_PP_CUBIC_OFFSET_F1_4, 5, "R200_PP_CUBIC_OFFSET_F1_4"},
	{R200_PP_CUBIC_FACES_5, 1, "R200_PP_CUBIC_FACES_5"},
	{R200_PP_CUBIC_OFFSET_F1_5, 5, "R200_PP_CUBIC_OFFSET_F1_5"},
	{RADEON_PP_TEX_SIZE_0, 2, "RADEON_PP_TEX_SIZE_0"},
	{RADEON_PP_TEX_SIZE_1, 2, "RADEON_PP_TEX_SIZE_1"},
	{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"},
	{R200_PP_TRI_PERF, 2, "R200_PP_TRI_PERF"},
	{R200_PP_AFS_0, 32, "R200_PP_AFS_0"},	/* 85 */
	{R200_PP_AFS_1, 32, "R200_PP_AFS_1"},
	{R200_PP_TFACTOR_0, 8, "R200_ATF_TFACTOR"},
	{R200_PP_TXFILTER_0, 8, "R200_PP_TXCTLALL_0"},
	{R200_PP_TXFILTER_1, 8, "R200_PP_TXCTLALL_1"},
	{R200_PP_TXFILTER_2, 8, "R200_PP_TXCTLALL_2"},
	{R200_PP_TXFILTER_3, 8, "R200_PP_TXCTLALL_3"},
	{R200_PP_TXFILTER_4, 8, "R200_PP_TXCTLALL_4"},
	{R200_PP_TXFILTER_5, 8, "R200_PP_TXCTLALL_5"},
};

/* ================================================================
 * Performance monitoring functions
 */

static void radeon_clear_box(drm_radeon_private_t * dev_priv,
			     int x, int y, int w, int h, int r, int g, int b)
{
	u32 color;
	RING_LOCALS;

	x += dev_priv->sarea_priv->boxes[0].x1;
	y += dev_priv->sarea_priv->boxes[0].y1;

	switch (dev_priv->color_fmt) {
	case RADEON_COLOR_FORMAT_RGB565:
		color = (((r & 0xf8) << 8) |
			 ((g & 0xfc) << 3) | ((b & 0xf8) >> 3));
		break;
	case RADEON_COLOR_FORMAT_ARGB8888:
	default:
		color = (((0xff) << 24) | (r << 16) | (g << 8) | b);
		break;
	}

	BEGIN_RING(4);
	RADEON_WAIT_UNTIL_3D_IDLE();
	OUT_RING(CP_PACKET0(RADEON_DP_WRITE_MASK, 0));
	OUT_RING(0xffffffff);
	ADVANCE_RING();

	BEGIN_RING(6);

	OUT_RING(CP_PACKET3(RADEON_CNTL_PAINT_MULTI, 4));
	OUT_RING(RADEON_GMC_DST_PITCH_OFFSET_CNTL |
		 RADEON_GMC_BRUSH_SOLID_COLOR |
		 (dev_priv->color_fmt << 8) |
		 RADEON_GMC_SRC_DATATYPE_COLOR |
		 RADEON_ROP3_P | RADEON_GMC_CLR_CMP_CNTL_DIS);

	if (dev_priv->page_flipping && dev_priv->current_page == 1) {
		OUT_RING(dev_priv->front_pitch_offset);
	} else {
		OUT_RING(dev_priv->back_pitch_offset);
	}

	OUT_RING(color);

	OUT_RING((x << 16) | y);
	OUT_RING((w << 16) | h);

	ADVANCE_RING();
}

static void radeon_cp_performance_boxes(drm_radeon_private_t * dev_priv)
{
	/* Collapse various things into a wait flag -- trying to
	 * guess if userspase slept -- better just to have them tell us.
	 */
	if (dev_priv->stats.last_frame_reads > 1 ||
	    dev_priv->stats.last_clear_reads > dev_priv->stats.clears) {
		dev_priv->stats.boxes |= RADEON_BOX_WAIT_IDLE;
	}

	if (dev_priv->stats.freelist_loops) {
		dev_priv->stats.boxes |= RADEON_BOX_WAIT_IDLE;
	}

	/* Purple box for page flipping
	 */
	if (dev_priv->stats.boxes & RADEON_BOX_FLIP)
		radeon_clear_box(dev_priv, 4, 4, 8, 8, 255, 0, 255);

	/* Red box if we have to wait for idle at any point
	 */
	if (dev_priv->stats.boxes & RADEON_BOX_WAIT_IDLE)
		radeon_clear_box(dev_priv, 16, 4, 8, 8, 255, 0, 0);

	/* Blue box: lost context?
	 */

	/* Yellow box for texture swaps
	 */
	if (dev_priv->stats.boxes & RADEON_BOX_TEXTURE_LOAD)
		radeon_clear_box(dev_priv, 40, 4, 8, 8, 255, 255, 0);

	/* Green box if hardware never idles (as far as we can tell)
	 */
	if (!(dev_priv->stats.boxes & RADEON_BOX_DMA_IDLE))
		radeon_clear_box(dev_priv, 64, 4, 8, 8, 0, 255, 0);

	/* Draw bars indicating number of buffers allocated
	 * (not a great measure, easily confused)
	 */
	if (dev_priv->stats.requested_bufs) {
		if (dev_priv->stats.requested_bufs > 100)
			dev_priv->stats.requested_bufs = 100;

		radeon_clear_box(dev_priv, 4, 16,
				 dev_priv->stats.requested_bufs, 4,
				 196, 128, 128);
	}

	memset(&dev_priv->stats, 0, sizeof(dev_priv->stats));

}

/* ================================================================
 * CP command dispatch functions
 */

static void radeon_cp_dispatch_clear(drm_device_t * dev,
				     drm_radeon_clear_t * clear,
				     drm_radeon_clear_rect_t * depth_boxes)
{
	drm_radeon_private_t *dev_priv = dev->dev_private;
	drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
	drm_radeon_depth_clear_t *depth_clear = &dev_priv->depth_clear;
	int nbox = sarea_priv->nbox;
	drm_clip_rect_t *pbox = sarea_priv->boxes;
	unsigned int flags = clear->flags;
	u32 rb3d_cntl = 0, rb3d_stencilrefmask = 0;
	int i;
	RING_LOCALS;
	DRM_DEBUG("flags = 0x%x\n", flags);

	dev_priv->stats.clears++;

	if (dev_priv->page_flipping && dev_priv->current_page == 1) {
		unsigned int tmp = flags;

		flags &= ~(RADEON_FRONT | RADEON_BACK);
		if (tmp & RADEON_FRONT)
			flags |= RADEON_BACK;
		if (tmp & RADEON_BACK)
			flags |= RADEON_FRONT;
	}

	if (flags & (RADEON_FRONT | RADEON_BACK)) {

		BEGIN_RING(4);

		/* Ensure the 3D stream is idle before doing a
		 * 2D fill to clear the front or back buffer.
		 */
		RADEON_WAIT_UNTIL_3D_IDLE();

		OUT_RING(CP_PACKET0(RADEON_DP_WRITE_MASK, 0));
		OUT_RING(clear->color_mask);

		ADVANCE_RING();

		/* Make sure we restore the 3D state next time.
		 */
		dev_priv->sarea_priv->ctx_owner = 0;

		for (i = 0; i < nbox; i++) {
			int x = pbox[i].x1;
			int y = pbox[i].y1;
			int w = pbox[i].x2 - x;
			int h = pbox[i].y2 - y;

			DRM_DEBUG("dispatch clear %d,%d-%d,%d flags 0x%x\n",
				  x, y, w, h, flags);

			if (flags & RADEON_FRONT) {
				BEGIN_RING(6);

				OUT_RING(CP_PACKET3
					 (RADEON_CNTL_PAINT_MULTI, 4));
				OUT_RING(RADEON_GMC_DST_PITCH_OFFSET_CNTL |
					 RADEON_GMC_BRUSH_SOLID_COLOR |
					 (dev_priv->
					  color_fmt << 8) |
					 RADEON_GMC_SRC_DATATYPE_COLOR |
					 RADEON_ROP3_P |
					 RADEON_GMC_CLR_CMP_CNTL_DIS);

				OUT_RING(dev_priv->front_pitch_offset);
				OUT_RING(clear->clear_color);

				OUT_RING((x << 16) | y);
				OUT_RING((w << 16) | h);

				ADVANCE_RING();
			}

			if (flags & RADEON_BACK) {
				BEGIN_RING(6);

				OUT_RING(CP_PACKET3
					 (RADEON_CNTL_PAINT_MULTI, 4));
				OUT_RING(RADEON_GMC_DST_PITCH_OFFSET_CNTL |
					 RADEON_GMC_BRUSH_SOLID_COLOR |
					 (dev_priv->
					  color_fmt << 8) |
					 RADEON_GMC_SRC_DATATYPE_COLOR |
					 RADEON_ROP3_P |
					 RADEON_GMC_CLR_CMP_CNTL_DIS);

				OUT_RING(dev_priv->back_pitch_offset);
				OUT_RING(clear->clear_color);

				OUT_RING((x << 16) | y);
				OUT_RING((w << 16) | h);

				ADVANCE_RING();
			}
		}
	}

	/* hyper z clear */
	/* no docs available, based on reverse engeneering by Stephane Marchesin */
	if ((flags & (RADEON_DEPTH | RADEON_STENCIL))
	    && (flags & RADEON_CLEAR_FASTZ)) {

		int i;
		int depthpixperline =
		    dev_priv->depth_fmt ==
		    RADEON_DEPTH_FORMAT_16BIT_INT_Z ? (dev_priv->depth_pitch /
						       2) : (dev_priv->
							     depth_pitch / 4);

		u32 clearmask;

		u32 tempRB3D_DEPTHCLEARVALUE = clear->clear_depth |
		    ((clear->depth_mask & 0xff) << 24);

		/* Make sure we restore the 3D state next time.
		 * we haven't touched any "normal" state - still need this?
		 */
		dev_priv->sarea_priv->ctx_owner = 0;

		if ((dev_priv->flags & CHIP_HAS_HIERZ)
		    && (flags & RADEON_USE_HIERZ)) {
			/* FIXME : reverse engineer that for Rx00 cards */
			/* FIXME : the mask supposedly contains low-res z values. So can't set
			   just to the max (0xff? or actually 0x3fff?), need to take z clear
			   value into account? */
			/* pattern seems to work for r100, though get slight
			   rendering errors with glxgears. If hierz is not enabled for r100,
			   only 4 bits which indicate clear (15,16,31,32, all zero) matter, the
			   other ones are ignored, and the same clear mask can be used. That's
			   very different behaviour than R200 which needs different clear mask
			   and different number of tiles to clear if hierz is enabled or not !?!
			 */
			clearmask = (0xff << 22) | (0xff << 6) | 0x003f003f;
		} else {
			/* clear mask : chooses the clearing pattern.
			   rv250: could be used to clear only parts of macrotiles
			   (but that would get really complicated...)?
			   bit 0 and 1 (either or both of them ?!?!) are used to
			   not clear tile (or maybe one of the bits indicates if the tile is
			   compressed or not), bit 2 and 3 to not clear tile 1,...,.
			   Pattern is as follows:
			   | 0,1 | 4,5 | 8,9 |12,13|16,17|20,21|24,25|28,29|
			   bits -------------------------------------------------
			   | 2,3 | 6,7 |10,11|14,15|18,19|22,23|26,27|30,31|
			   rv100: clearmask covers 2x8 4x1 tiles, but one clear still
			   covers 256 pixels ?!?
			 */
			clearmask = 0x0;
		}

		BEGIN_RING(8);
		RADEON_WAIT_UNTIL_2D_IDLE();
		OUT_RING_REG(RADEON_RB3D_DEPTHCLEARVALUE,
			     tempRB3D_DEPTHCLEARVALUE);
		/* what offset is this exactly ? */
		OUT_RING_REG(RADEON_RB3D_ZMASKOFFSET, 0);
		/* need ctlstat, otherwise get some strange black flickering */
		OUT_RING_REG(RADEON_RB3D_ZCACHE_CTLSTAT,
			     RADEON_RB3D_ZC_FLUSH_ALL);
		ADVANCE_RING();

		for (i = 0; i < nbox; i++) {
			int tileoffset, nrtilesx, nrtilesy, j;
			/* it looks like r200 needs rv-style clears, at least if hierz is not enabled? */
			if ((dev_priv->flags & CHIP_HAS_HIERZ)
			    && !(dev_priv->microcode_version == UCODE_R200)) {
				/* FIXME : figure this out for r200 (when hierz is enabled). Or
				   maybe r200 actually doesn't need to put the low-res z value into
				   the tile cache like r100, but just needs to clear the hi-level z-buffer?
				   Works for R100, both with hierz and without.
				   R100 seems to operate on 2x1 8x8 tiles, but...
				   odd: offset/nrtiles need to be 64 pix (4 block) aligned? Potentially
				   problematic with resolutions which are not 64 pix aligned? */
				tileoffset =
				    ((pbox[i].y1 >> 3) * depthpixperline +
				     pbox[i].x1) >> 6;
				nrtilesx =
				    ((pbox[i].x2 & ~63) -
				     (pbox[i].x1 & ~63)) >> 4;
				nrtilesy =
				    (pbox[i].y2 >> 3) - (pbox[i].y1 >> 3);
				for (j = 0; j <= nrtilesy; j++) {
					BEGIN_RING(4);
					OUT_RING(CP_PACKET3
						 (RADEON_3D_CLEAR_ZMASK, 2));
					/* first tile */
					OUT_RING(tileoffset * 8);
					/* the number of tiles to clear */
					OUT_RING(nrtilesx + 4);
					/* clear mask : chooses the clearing pattern. */
					OUT_RING(clearmask);
					ADVANCE_RING();