Optimise RADEONWaitForVLine

Only avoid the vlines we are rendering to, instead of the entire
screen. This way we don't stall the card for longer than we
absolutely have to.

EXA calls fixed by Alex Deucher.

8 files changed, 55 insertions, 45 deletions

diff --git a/src/atombios_crtc.c b/src/atombios_crtc.c
index bbd0c0a9..4e2395f8 100644
--- a/src/atombios_crtc.c
+++ b/src/atombios_crtc.c
@@ -511,11 +511,6 @@ atombios_crtc_mode_set(xf86CrtcPtr crtc,
 	else
 	    OUTREG(AVIVO_D1MODE_DATA_FORMAT + radeon_crtc->crtc_offset,
 		   0);
-
-	OUTREG(AVIVO_D1MODE_VLINE_START_END + radeon_crtc->crtc_offset,
-	       (((0) << AVIVO_D1MODE_VLINE_START_SHIFT) |
-		((mode->VDisplay) << AVIVO_D1MODE_VLINE_END_SHIFT) |
-		AVIVO_D1MODE_VLINE_INV));
     }
 
     atombios_crtc_set_pll(crtc, adjusted_mode, pll_flags);
diff --git a/src/legacy_crtc.c b/src/legacy_crtc.c
index 751b5efc..5ed14d44 100644
--- a/src/legacy_crtc.c
+++ b/src/legacy_crtc.c
@@ -133,8 +133,6 @@ RADEONRestoreCrtcRegisters(ScrnInfoPtr pScrn,
     OUTREG(RADEON_CRTC_V_TOTAL_DISP,    restore->crtc_v_total_disp);
     OUTREG(RADEON_CRTC_V_SYNC_STRT_WID, restore->crtc_v_sync_strt_wid);
 
-    OUTREG(RADEON_CRTC_GUI_TRIG_VLINE,  restore->crtc_gui_trig_vline);
-
     if (IS_R300_VARIANT)
 	OUTREG(R300_CRTC_TILE_X0_Y0,    restore->crtc_tile_x0_y0);
     OUTREG(RADEON_CRTC_OFFSET_CNTL,     restore->crtc_offset_cntl);
@@ -179,8 +177,6 @@ RADEONRestoreCrtc2Registers(ScrnInfoPtr pScrn,
     OUTREG(RADEON_CRTC2_V_TOTAL_DISP,    restore->crtc2_v_total_disp);
     OUTREG(RADEON_CRTC2_V_SYNC_STRT_WID, restore->crtc2_v_sync_strt_wid);
 
-    OUTREG(RADEON_CRTC2_GUI_TRIG_VLINE,  restore->crtc2_gui_trig_vline);
-
     OUTREG(RADEON_FP_H2_SYNC_STRT_WID,   restore->fp_h2_sync_strt_wid);
     OUTREG(RADEON_FP_V2_SYNC_STRT_WID,   restore->fp_v2_sync_strt_wid);
 
@@ -522,8 +518,6 @@ RADEONSaveCrtcRegisters(ScrnInfoPtr pScrn, RADEONSavePtr save)
     save->crtc_v_total_disp    = INREG(RADEON_CRTC_V_TOTAL_DISP);
     save->crtc_v_sync_strt_wid = INREG(RADEON_CRTC_V_SYNC_STRT_WID);
 
-    save->crtc_gui_trig_vline  = INREG(RADEON_CRTC_GUI_TRIG_VLINE);
-
     save->crtc_offset          = INREG(RADEON_CRTC_OFFSET);
     save->crtc_offset_cntl     = INREG(RADEON_CRTC_OFFSET_CNTL);
     save->crtc_pitch           = INREG(RADEON_CRTC_PITCH);
@@ -560,8 +554,6 @@ RADEONSaveCrtc2Registers(ScrnInfoPtr pScrn, RADEONSavePtr save)
     save->crtc2_v_total_disp    = INREG(RADEON_CRTC2_V_TOTAL_DISP);
     save->crtc2_v_sync_strt_wid = INREG(RADEON_CRTC2_V_SYNC_STRT_WID);
 
-    save->crtc2_gui_trig_vline  = INREG(RADEON_CRTC2_GUI_TRIG_VLINE);
-
     save->crtc2_offset          = INREG(RADEON_CRTC2_OFFSET);
     save->crtc2_offset_cntl     = INREG(RADEON_CRTC2_OFFSET_CNTL);
     save->crtc2_pitch           = INREG(RADEON_CRTC2_PITCH);
@@ -955,10 +947,6 @@ RADEONInitCrtcRegisters(xf86CrtcPtr crtc, RADEONSavePtr save,
 			 (pScrn->bitsPerPixel * 8));
     save->crtc_pitch |= save->crtc_pitch << 16;
 
-    save->crtc_gui_trig_vline = ((0 << RADEON_CRTC_GUI_TRIG_VLINE_START_SHIFT) |
-				 (mode->CrtcVDisplay << RADEON_CRTC_GUI_TRIG_VLINE_END_SHIFT) |
-				 RADEON_CRTC_GUI_TRIG_VLINE_INV);
-
     if (info->IsDellServer) {
 	save->dac2_cntl = info->SavedReg->dac2_cntl;
 	save->tv_dac_cntl = info->SavedReg->tv_dac_cntl;
@@ -1150,10 +1138,6 @@ RADEONInitCrtc2Registers(xf86CrtcPtr crtc, RADEONSavePtr save,
 			  ((pScrn->bitsPerPixel * 8) -1)) / (pScrn->bitsPerPixel * 8);
     save->crtc2_pitch |= save->crtc2_pitch << 16;
 
-    save->crtc2_gui_trig_vline = ((0 << RADEON_CRTC_GUI_TRIG_VLINE_START_SHIFT) |
-				  (mode->CrtcVDisplay << RADEON_CRTC_GUI_TRIG_VLINE_END_SHIFT) |
-				  RADEON_CRTC_GUI_TRIG_VLINE_INV);
-
     /* check to see if TV DAC is enabled for another crtc and keep it enabled */
     if (save->crtc2_gen_cntl & RADEON_CRTC2_CRT2_ON)
 	save->crtc2_gen_cntl = RADEON_CRTC2_CRT2_ON;
diff --git a/src/radeon.h b/src/radeon.h
index 826ffe38..051b7658 100644
--- a/src/radeon.h
+++ b/src/radeon.h
@@ -941,10 +941,12 @@ extern Bool radeon_card_posted(ScrnInfoPtr pScrn);
 /* radeon_commonfuncs.c */
 #ifdef XF86DRI
 extern void RADEONWaitForIdleCP(ScrnInfoPtr pScrn);
-extern void RADEONWaitForVLineCP(ScrnInfoPtr pScrn, PixmapPtr pPix, int crtc);
+extern void RADEONWaitForVLineCP(ScrnInfoPtr pScrn, PixmapPtr pPix,
+	int crtc, int start, int stop);
 #endif
 extern void RADEONWaitForIdleMMIO(ScrnInfoPtr pScrn);
-extern void RADEONWaitForVLineMMIO(ScrnInfoPtr pScrn, PixmapPtr pPix, int crtc);
+extern void RADEONWaitForVLineMMIO(ScrnInfoPtr pScrn, PixmapPtr pPix,
+	int crtc, int start, int stop);
 
 /* radeon_crtc.c */
 extern void radeon_crtc_dpms(xf86CrtcPtr crtc, int mode);
diff --git a/src/radeon_commonfuncs.c b/src/radeon_commonfuncs.c
index 7f32acde..c2f34511 100644
--- a/src/radeon_commonfuncs.c
+++ b/src/radeon_commonfuncs.c
@@ -636,7 +636,8 @@ static void FUNC_NAME(RADEONInit3DEngine)(ScrnInfoPtr pScrn)
 }
 
 /* inserts a wait for vline in the command stream */
-void FUNC_NAME(RADEONWaitForVLine)(ScrnInfoPtr pScrn, PixmapPtr pPix, int crtc)
+void FUNC_NAME(RADEONWaitForVLine)(ScrnInfoPtr pScrn, PixmapPtr pPix,
+	int crtc, int start, int stop)
 {
     RADEONInfoPtr  info = RADEONPTR(pScrn);
     xf86CrtcConfigPtr  xf86_config = XF86_CRTC_CONFIG_PTR(pScrn);
@@ -646,6 +647,9 @@ void FUNC_NAME(RADEONWaitForVLine)(ScrnInfoPtr pScrn, PixmapPtr pPix, int crtc)
     if ((crtc < 0) || (crtc > 1))
 	return;
 
+    if (stop < start)
+	return;
+
     if (!xf86_config->crtc[crtc]->enabled)
 	return;
 
@@ -657,16 +661,45 @@ void FUNC_NAME(RADEONWaitForVLine)(ScrnInfoPtr pScrn, PixmapPtr pPix, int crtc)
 	offset = pPix->devPrivate.ptr - info->FB;
 
     /* if drawing to front buffer */
-    if (offset == 0) {
-	BEGIN_ACCEL(1);
+    if (offset != 0)
+	return;
+
+    start = max(start, 0);
+    stop = max(stop, xf86_config->crtc[crtc]->mode.VDisplay);
+
+    if (start > xf86_config->crtc[crtc]->mode.VDisplay)
+	return;
+
+    BEGIN_ACCEL(2);
+
+    if (IS_AVIVO_VARIANT) {
+	RADEONCrtcPrivatePtr radeon_crtc = xf86_config->crtc[crtc]->driver_private;
+
+	OUT_ACCEL_REG(AVIVO_D1MODE_VLINE_START_END + radeon_crtc->crtc_offset,
+		      ((start << AVIVO_D1MODE_VLINE_START_SHIFT) |
+		       (stop << AVIVO_D1MODE_VLINE_END_SHIFT) |
+		       AVIVO_D1MODE_VLINE_INV));
+    } else {
 	if (crtc == 0)
-	    OUT_ACCEL_REG(RADEON_WAIT_UNTIL, (RADEON_WAIT_CRTC_VLINE |
-					      RADEON_ENG_DISPLAY_SELECT_CRTC0));
+	    OUT_ACCEL_REG(RADEON_CRTC_GUI_TRIG_VLINE,
+			  ((start << RADEON_CRTC_GUI_TRIG_VLINE_START_SHIFT) |
+			   (stop << RADEON_CRTC_GUI_TRIG_VLINE_END_SHIFT) |
+			   RADEON_CRTC_GUI_TRIG_VLINE_INV));
 	else
-	    OUT_ACCEL_REG(RADEON_WAIT_UNTIL, (RADEON_WAIT_CRTC_VLINE |
-					      RADEON_ENG_DISPLAY_SELECT_CRTC1));
-	FINISH_ACCEL();
+	    OUT_ACCEL_REG(RADEON_CRTC2_GUI_TRIG_VLINE,
+			  ((start << RADEON_CRTC_GUI_TRIG_VLINE_START_SHIFT) |
+			   (stop << RADEON_CRTC_GUI_TRIG_VLINE_END_SHIFT) |
+			   RADEON_CRTC_GUI_TRIG_VLINE_INV));
     }
+
+    if (crtc == 0)
+	OUT_ACCEL_REG(RADEON_WAIT_UNTIL, (RADEON_WAIT_CRTC_VLINE |
+					  RADEON_ENG_DISPLAY_SELECT_CRTC0));
+    else
+	OUT_ACCEL_REG(RADEON_WAIT_UNTIL, (RADEON_WAIT_CRTC_VLINE |
+					  RADEON_ENG_DISPLAY_SELECT_CRTC1));
+
+    FINISH_ACCEL();
 }
 
 /* MMIO:
diff --git a/src/radeon_exa_funcs.c b/src/radeon_exa_funcs.c
index 76d6c413..fe487016 100644
--- a/src/radeon_exa_funcs.c
+++ b/src/radeon_exa_funcs.c
@@ -117,8 +117,6 @@ FUNC_NAME(RADEONPrepareSolid)(PixmapPtr pPix, int alu, Pixel pm, Pixel fg)
     OUT_ACCEL_REG(RADEON_DST_PITCH_OFFSET, dst_pitch_offset);
     FINISH_ACCEL();
 
-    FUNC_NAME(RADEONWaitForVLine)(pScrn, pPix, RADEONBiggerCrtcArea(pPix));
-
     return TRUE;
 }
 
@@ -131,6 +129,8 @@ FUNC_NAME(RADEONSolid)(PixmapPtr pPix, int x1, int y1, int x2, int y2)
 
     TRACE;
 
+    FUNC_NAME(RADEONWaitForVLine)(pScrn, pPix, RADEONBiggerCrtcArea(pPix), y1, y2);
+
     BEGIN_ACCEL(2);
     OUT_ACCEL_REG(RADEON_DST_Y_X, (y1 << 16) | x1);
     OUT_ACCEL_REG(RADEON_DST_HEIGHT_WIDTH, ((y2 - y1) << 16) | (x2 - x1));
@@ -207,8 +207,6 @@ FUNC_NAME(RADEONPrepareCopy)(PixmapPtr pSrc,   PixmapPtr pDst,
     FUNC_NAME(RADEONDoPrepareCopy)(pScrn, src_pitch_offset, dst_pitch_offset,
 				   datatype, rop, planemask);
 
-    FUNC_NAME(RADEONWaitForVLine)(pScrn, pDst, RADEONBiggerCrtcArea(pDst));
-
     return TRUE;
 }
 
@@ -232,6 +230,8 @@ FUNC_NAME(RADEONCopy)(PixmapPtr pDst,
 	dstY += h - 1;
     }
 
+    FUNC_NAME(RADEONWaitForVLine)(pScrn, pDst, RADEONBiggerCrtcArea(pDst), dstY, dstY + h);
+
     BEGIN_ACCEL(3);
 
     OUT_ACCEL_REG(RADEON_SRC_Y_X,	   (srcY << 16) | srcX);
@@ -289,7 +289,7 @@ FUNC_NAME(RADEONUploadToScreen)(PixmapPtr pDst, int x, int y, int w, int h,
 
 	RADEON_SWITCH_TO_2D();
 
-	FUNC_NAME(RADEONWaitForVLine)(pScrn, pDst, RADEONBiggerCrtcArea(pDst));
+	FUNC_NAME(RADEONWaitForVLine)(pScrn, pDst, RADEONBiggerCrtcArea(pDst), y, y + h);
 
 	while ((buf = RADEONHostDataBlit(pScrn,
 					 cpp, w, dst_pitch_off, &buf_pitch,
diff --git a/src/radeon_exa_render.c b/src/radeon_exa_render.c
index 5d3f737e..4b918c4c 100644
--- a/src/radeon_exa_render.c
+++ b/src/radeon_exa_render.c
@@ -624,8 +624,6 @@ static Bool FUNC_NAME(R100PrepareComposite)(int op,
     OUT_ACCEL_REG(RADEON_RB3D_BLENDCNTL, blendcntl);
     FINISH_ACCEL();
 
-    FUNC_NAME(RADEONWaitForVLine)(pScrn, pDst, RADEONBiggerCrtcArea(pDst));
-
     return TRUE;
 }
 
@@ -932,8 +930,6 @@ static Bool FUNC_NAME(R200PrepareComposite)(int op, PicturePtr pSrcPicture,
     OUT_ACCEL_REG(RADEON_RB3D_BLENDCNTL, blendcntl);
     FINISH_ACCEL();
 
-    FUNC_NAME(RADEONWaitForVLine)(pScrn, pDst, RADEONBiggerCrtcArea(pDst));
-
     return TRUE;
 }
 
@@ -1889,8 +1885,6 @@ static Bool FUNC_NAME(R300PrepareComposite)(int op, PicturePtr pSrcPicture,
 	OUT_ACCEL_REG(R300_VAP_VTX_SIZE, 4);
     FINISH_ACCEL();
 
-    FUNC_NAME(RADEONWaitForVLine)(pScrn, pDst, RADEONBiggerCrtcArea(pDst));
-
     return TRUE;
 }
 
@@ -2003,6 +1997,8 @@ static void FUNC_NAME(RadeonCompositeTile)(PixmapPtr pDst,
     else
 	vtx_count = 4;
 
+    FUNC_NAME(RADEONWaitForVLine)(pScrn, pDst, RADEONBiggerCrtcArea(pDst), dstY, dstY + h);
+
 #ifdef ACCEL_CP
     if (info->ChipFamily < CHIP_FAMILY_R200) {
 	BEGIN_RING(3 * vtx_count + 3);
diff --git a/src/radeon_probe.h b/src/radeon_probe.h
index 313daa6b..8c6845f7 100644
--- a/src/radeon_probe.h
+++ b/src/radeon_probe.h
@@ -442,7 +442,6 @@ typedef struct {
     uint32_t          crtc_h_sync_strt_wid;
     uint32_t          crtc_v_total_disp;
     uint32_t          crtc_v_sync_strt_wid;
-    uint32_t          crtc_gui_trig_vline;
     uint32_t          crtc_offset;
     uint32_t          crtc_offset_cntl;
     uint32_t          crtc_pitch;
@@ -464,7 +463,6 @@ typedef struct {
     uint32_t          crtc2_h_sync_strt_wid;
     uint32_t          crtc2_v_total_disp;
     uint32_t          crtc2_v_sync_strt_wid;
-    uint32_t          crtc2_gui_trig_vline;
     uint32_t          crtc2_offset;
     uint32_t          crtc2_offset_cntl;
     uint32_t          crtc2_pitch;
diff --git a/src/radeon_textured_videofuncs.c b/src/radeon_textured_videofuncs.c
index ddf2df63..06387511 100644
--- a/src/radeon_textured_videofuncs.c
+++ b/src/radeon_textured_videofuncs.c
@@ -1519,7 +1519,9 @@ FUNC_NAME(RADEONDisplayTexturedVideo)(ScrnInfoPtr pScrn, RADEONPortPrivPtr pPriv
 							   pPriv->drw_x + pPriv->dst_w,
 							   pPriv->drw_y,
 							   pPriv->drw_y + pPriv->dst_h,
-							   pPriv->desired_crtc));
+							   pPriv->desired_crtc),
+				  pPriv->drw_y,
+				  pPriv->drw_y + pPriv->dst_h);
 
     while (nBox--) {
 	int srcX, srcY, srcw, srch;