diff options
| -rw-r--r-- | src/radeon.h | 1 | ||||
| -rw-r--r-- | src/radeon_render.c | 103 |
2 files changed, 102 insertions, 2 deletions
diff --git a/src/radeon.h b/src/radeon.h index 523d2b72..636923a2 100644 --- a/src/radeon.h +++ b/src/radeon.h @@ -555,6 +555,7 @@ typedef struct { Bool RenderAccel; Bool RenderInited3D; FBLinearPtr RenderTex; + Bool RenderTexValidR100; void (*RenderCallback)(ScrnInfoPtr); Time RenderTimeout; diff --git a/src/radeon_render.c b/src/radeon_render.c index 145b6538..9a218a39 100644 --- a/src/radeon_render.c +++ b/src/radeon_render.c @@ -7,6 +7,11 @@ #ifndef RENDER_GENERIC_HELPER #define RENDER_GENERIC_HELPER +/* R100 code path constants */ +/* Minimum working values for RV200: 65x8 - your mileage may vary */ +#define MAGIC_R100_MIN_TEX_WIDTH (65) +#define MAGIC_R100_MIN_TEX_HEIGHT (8) + static void RadeonInit3DEngineMMIO(ScrnInfoPtr pScrn); #ifdef XF86DRI static void RadeonInit3DEngineCP(ScrnInfoPtr pScrn); @@ -421,6 +426,8 @@ static Bool FUNC_NAME(R100SetupTexture)( CARD8 *dst; CARD32 tex_size = 0, txformat; int dst_pitch, offset, size, i, tex_bytepp; + int offscreensize = (pScrn->bitsPerPixel >> 3) * + pScrn->displayWidth * MAGIC_R100_MIN_TEX_HEIGHT; ACCEL_PREAMBLE(); if ((width > 2048) || (height > 2048)) @@ -440,7 +447,8 @@ static Bool FUNC_NAME(R100SetupTexture)( dst_pitch = (width * tex_bytepp + 31) & ~31; size = dst_pitch * height; - if (!AllocateLinear(pScrn, size)) + /* Allocate off-screen space for texture pre-rendering as well */ + if (!AllocateLinear(pScrn, size + offscreensize)) return FALSE; if (flags & XAA_RENDER_REPEAT) { @@ -451,7 +459,7 @@ static Bool FUNC_NAME(R100SetupTexture)( txformat |= RADEON_TXFORMAT_NON_POWER2; } - offset = info->RenderTex->offset * pScrn->bitsPerPixel / 8; + offset = info->RenderTex->offset * pScrn->bitsPerPixel / 8 + offscreensize; /* Upload texture to card. Should use ImageWrite to avoid syncing. */ i = height; @@ -481,6 +489,7 @@ static Bool FUNC_NAME(R100SetupTexture)( RADEON_CLAMP_S_WRAP | RADEON_CLAMP_T_WRAP); FINISH_ACCEL(); + info->RenderTexValidR100 = FALSE; return TRUE; } @@ -608,6 +617,95 @@ FUNC_NAME(R100SubsequentCPUToScreenTexture) ( ACCEL_PREAMBLE(); + /* R100 chips seem to have cache problems and do not reload small textures + * from memory for small render areas. Thus we have to render a + * 'sufficiently large' portion of the texture (may contain uninitialized + * data as well) to off-screen memory to invalidate the cache. */ + if (!info->RenderTexValidR100 && + width*height < MAGIC_R100_MIN_TEX_WIDTH*MAGIC_R100_MIN_TEX_HEIGHT) +/* (width<MAGIC_R100_MIN_TEX_WIDTH || height<MAGIC_R100_MIN_TEX_HEIGHT))*/ + { + fboffset = info->fbLocation + pScrn->fbOffset + + info->RenderTex->offset * (pScrn->bitsPerPixel >> 3); + l = 0.0; + t = 0.0; + r = MAGIC_R100_MIN_TEX_WIDTH; + b = MAGIC_R100_MIN_TEX_HEIGHT; + +#ifdef ACCEL_CP + BEGIN_RING(23); + + OUT_ACCEL_REG(RADEON_RB3D_COLOROFFSET, fboffset); + + OUT_RING(CP_PACKET3(RADEON_CP_PACKET3_3D_DRAW_IMMD, 17)); + /* RADEON_SE_VTX_FMT */ + OUT_RING(RADEON_CP_VC_FRMT_XY | + RADEON_CP_VC_FRMT_ST0); + /* SE_VF_CNTL */ + OUT_RING(RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_FAN | + RADEON_CP_VC_CNTL_PRIM_WALK_RING | + RADEON_CP_VC_CNTL_MAOS_ENABLE | + RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE | + (4 << RADEON_CP_VC_CNTL_NUM_SHIFT)); + + OUT_RING(F_TO_DW(l)); + OUT_RING(F_TO_DW(t)); + OUT_RING(F_TO_DW(l)); + OUT_RING(F_TO_DW(t)); + + OUT_RING(F_TO_DW(r)); + OUT_RING(F_TO_DW(t)); + OUT_RING(F_TO_DW(r)); + OUT_RING(F_TO_DW(t)); + + OUT_RING(F_TO_DW(r)); + OUT_RING(F_TO_DW(b)); + OUT_RING(F_TO_DW(r)); + OUT_RING(F_TO_DW(b)); + + OUT_RING(F_TO_DW(l)); + OUT_RING(F_TO_DW(b)); + OUT_RING(F_TO_DW(l)); + OUT_RING(F_TO_DW(b)); + + OUT_ACCEL_REG(RADEON_WAIT_UNTIL, RADEON_WAIT_3D_IDLECLEAN); + + ADVANCE_RING(); +#else + BEGIN_ACCEL(19); + + OUT_ACCEL_REG(RADEON_RB3D_COLOROFFSET, fboffset); + + OUT_ACCEL_REG(RADEON_SE_VF_CNTL, RADEON_VF_PRIM_TYPE_TRIANGLE_FAN | + RADEON_VF_PRIM_WALK_DATA | + RADEON_VF_RADEON_MODE | + (4 << RADEON_VF_NUM_VERTICES_SHIFT)); + + OUT_ACCEL_REG(RADEON_SE_PORT_DATA0, F_TO_DW(l)); + OUT_ACCEL_REG(RADEON_SE_PORT_DATA0, F_TO_DW(t)); + OUT_ACCEL_REG(RADEON_SE_PORT_DATA0, F_TO_DW(l)); + OUT_ACCEL_REG(RADEON_SE_PORT_DATA0, F_TO_DW(t)); + + OUT_ACCEL_REG(RADEON_SE_PORT_DATA0, F_TO_DW(r)); + OUT_ACCEL_REG(RADEON_SE_PORT_DATA0, F_TO_DW(t)); + OUT_ACCEL_REG(RADEON_SE_PORT_DATA0, F_TO_DW(r)); + OUT_ACCEL_REG(RADEON_SE_PORT_DATA0, F_TO_DW(t)); + + OUT_ACCEL_REG(RADEON_SE_PORT_DATA0, F_TO_DW(r)); + OUT_ACCEL_REG(RADEON_SE_PORT_DATA0, F_TO_DW(b)); + OUT_ACCEL_REG(RADEON_SE_PORT_DATA0, F_TO_DW(r)); + OUT_ACCEL_REG(RADEON_SE_PORT_DATA0, F_TO_DW(b)); + + OUT_ACCEL_REG(RADEON_SE_PORT_DATA0, F_TO_DW(l)); + OUT_ACCEL_REG(RADEON_SE_PORT_DATA0, F_TO_DW(b)); + OUT_ACCEL_REG(RADEON_SE_PORT_DATA0, F_TO_DW(l)); + OUT_ACCEL_REG(RADEON_SE_PORT_DATA0, F_TO_DW(b)); + + OUT_ACCEL_REG(RADEON_WAIT_UNTIL, RADEON_WAIT_3D_IDLECLEAN); + FINISH_ACCEL(); +#endif + } + /* Note: we can't simply set up the 3D surface at the same location as the * front buffer, because the 2048x2048 limit on coordinates may be smaller * than the (MergedFB) screen. @@ -696,6 +794,7 @@ FUNC_NAME(R100SubsequentCPUToScreenTexture) ( OUT_ACCEL_REG(RADEON_WAIT_UNTIL, RADEON_WAIT_3D_IDLECLEAN); FINISH_ACCEL(); #endif + info->RenderTexValidR100 = TRUE; } |