From c4a87c6883ede7bbf486743efe3e9325d96f8e54 Mon Sep 17 00:00:00 2001 From: Roland Scheidegger Date: Wed, 8 Dec 2004 16:43:00 +0000 Subject: (Stephane Marchesin, me) add hyperz support to radeon drm. Only fast z clear and z buffer compression are working correctly, hierarchical-z is not. --- shared-core/radeon_state.c | 176 +++++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 172 insertions(+), 4 deletions(-) (limited to 'shared-core/radeon_state.c') diff --git a/shared-core/radeon_state.c b/shared-core/radeon_state.c index ec85efac..5ec9b35e 100644 --- a/shared-core/radeon_state.c +++ b/shared-core/radeon_state.c @@ -271,6 +271,7 @@ static __inline__ int radeon_check_and_fixup_packets(drm_radeon_private_t * case RADEON_EMIT_PP_TEX_SIZE_1: case RADEON_EMIT_PP_TEX_SIZE_2: case R200_EMIT_RB3D_BLENDCOLOR: + case R200_EMIT_TCL_POINT_SPRITE_CNTL: /* These packets don't contain memory offsets */ break; @@ -646,7 +647,9 @@ static struct { 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_RB3D_BLENDCOLOR, 3, "R200_RB3D_BLENDCOLOR"}, { + R200_SE_TCL_POINT_SPRITE_CNTL, 1, "R200_SE_TCL_POINT_SPRITE_CNTL"}, +}; /* ================================================================ * Performance monitoring functions @@ -858,11 +861,159 @@ static void radeon_cp_dispatch_clear(drm_device_t * dev, } } + /* 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(); + tileoffset += depthpixperline >> 6; + } + } + else if (dev_priv->microcode_version==UCODE_R200) { + /* works for rv250. */ + /* find first macro tile (8x2 4x4 z-pixels on rv250) */ + tileoffset = ((pbox[i].y1 >> 3) * depthpixperline + pbox[i].x1) >> 5; + nrtilesx = (pbox[i].x2 >> 5) - (pbox[i].x1 >> 5); + 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 */ + /* judging by the first tile offset needed, could possibly + directly address/clear 4x4 tiles instead of 8x2 * 4x4 + macro tiles, though would still need clear mask for + right/bottom if truely 4x4 granularity is desired ? */ + OUT_RING( tileoffset * 16 ); + /* the number of tiles to clear */ + OUT_RING( nrtilesx + 1 ); + /* clear mask : chooses the clearing pattern. */ + OUT_RING( clearmask ); + ADVANCE_RING(); + tileoffset += depthpixperline >> 5; + } + } + else { /* rv 100 */ + /* rv100 might not need 64 pix alignment, who knows */ + /* offsets are, hmm, weird */ + tileoffset = ((pbox[i].y1 >> 4) * depthpixperline + pbox[i].x1) >> 6; + nrtilesx = ((pbox[i].x2 & ~63) - (pbox[i].x1 & ~63)) >> 4; + nrtilesy = (pbox[i].y2 >> 4) - (pbox[i].y1 >> 4); + for (j = 0; j <= nrtilesy; j++) { + BEGIN_RING( 4 ); + OUT_RING( CP_PACKET3( RADEON_3D_CLEAR_ZMASK, 2 ) ); + OUT_RING( tileoffset * 128 ); + /* the number of tiles to clear */ + OUT_RING( nrtilesx + 4 ); + /* clear mask : chooses the clearing pattern. */ + OUT_RING( clearmask ); + ADVANCE_RING(); + tileoffset += depthpixperline >> 6; + } + } + } + + /* TODO don't always clear all hi-level z tiles */ + if ((dev_priv->flags & CHIP_HAS_HIERZ) && (dev_priv->microcode_version==UCODE_R200) + && (flags & RADEON_USE_HIERZ)) + /* r100 and cards without hierarchical z-buffer have no high-level z-buffer */ + /* 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? */ + { + BEGIN_RING( 4 ); + OUT_RING( CP_PACKET3( RADEON_3D_CLEAR_HIZ, 2 ) ); + OUT_RING( 0x0 ); /* First tile */ + OUT_RING( 0x3cc0 ); + OUT_RING( (0xff<<22)|(0xff<<6)| 0x003f003f); + ADVANCE_RING(); + } + } + /* We have to clear the depth and/or stencil buffers by * rendering a quad into just those buffers. Thus, we have to * make sure the 3D engine is configured correctly. */ - if ((dev_priv->microcode_version == UCODE_R200) && (flags & (RADEON_DEPTH | RADEON_STENCIL))) { + else if ((dev_priv->microcode_version == UCODE_R200) && + (flags & (RADEON_DEPTH | RADEON_STENCIL))) { int tempPP_CNTL; int tempRE_CNTL; @@ -929,6 +1080,14 @@ static void radeon_cp_dispatch_clear(drm_device_t * dev, tempRB3D_STENCILREFMASK = 0x00000000; } + if (flags & RADEON_USE_COMP_ZBUF) { + tempRB3D_ZSTENCILCNTL |= RADEON_Z_COMPRESSION_ENABLE | + RADEON_Z_DECOMPRESSION_ENABLE; + } + if (flags & RADEON_USE_HIERZ) { + tempRB3D_ZSTENCILCNTL |= RADEON_Z_HIERARCHY_ENABLE; + } + BEGIN_RING(26); RADEON_WAIT_UNTIL_2D_IDLE(); @@ -979,6 +1138,8 @@ static void radeon_cp_dispatch_clear(drm_device_t * dev, } } else if ((flags & (RADEON_DEPTH | RADEON_STENCIL))) { + int tempRB3D_ZSTENCILCNTL = depth_clear->rb3d_zstencilcntl; + rb3d_cntl = depth_clear->rb3d_cntl; if (flags & RADEON_DEPTH) { @@ -995,6 +1156,14 @@ static void radeon_cp_dispatch_clear(drm_device_t * dev, rb3d_stencilrefmask = 0x00000000; } + if (flags & RADEON_USE_COMP_ZBUF) { + tempRB3D_ZSTENCILCNTL |= RADEON_Z_COMPRESSION_ENABLE | + RADEON_Z_DECOMPRESSION_ENABLE; + } + if (flags & RADEON_USE_HIERZ) { + tempRB3D_ZSTENCILCNTL |= RADEON_Z_HIERARCHY_ENABLE; + } + BEGIN_RING(13); RADEON_WAIT_UNTIL_2D_IDLE(); @@ -1002,8 +1171,7 @@ static void radeon_cp_dispatch_clear(drm_device_t * dev, OUT_RING(0x00000000); OUT_RING(rb3d_cntl); - OUT_RING_REG(RADEON_RB3D_ZSTENCILCNTL, - depth_clear->rb3d_zstencilcntl); + OUT_RING_REG(RADEON_RB3D_ZSTENCILCNTL, tempRB3D_ZSTENCILCNTL); OUT_RING_REG(RADEON_RB3D_STENCILREFMASK, rb3d_stencilrefmask); OUT_RING_REG(RADEON_RB3D_PLANEMASK, 0x00000000); OUT_RING_REG(RADEON_SE_CNTL, depth_clear->se_cntl); -- cgit v1.2.3