diff options
Diffstat (limited to 'src/mesa/drivers/dri/radeon/radeon_span.c')
-rw-r--r-- | src/mesa/drivers/dri/radeon/radeon_span.c | 370 |
1 files changed, 253 insertions, 117 deletions
diff --git a/src/mesa/drivers/dri/radeon/radeon_span.c b/src/mesa/drivers/dri/radeon/radeon_span.c index 12051ff1c8..49ec2c378e 100644 --- a/src/mesa/drivers/dri/radeon/radeon_span.c +++ b/src/mesa/drivers/dri/radeon/radeon_span.c @@ -43,37 +43,168 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "main/glheader.h" #include "swrast/swrast.h" -#include "radeon_context.h" -#include "radeon_ioctl.h" -#include "radeon_state.h" +#include "radeon_common.h" +#include "radeon_lock.h" #include "radeon_span.h" -#include "radeon_tex.h" - -#include "drirenderbuffer.h" #define DBG 0 +static GLubyte *radeon_ptr32(const struct radeon_renderbuffer * rrb, + GLint x, GLint y) +{ + GLubyte *ptr = rrb->bo->ptr; + const __DRIdrawablePrivate *dPriv = rrb->dPriv; + uint32_t mask = RADEON_BO_FLAGS_MACRO_TILE | RADEON_BO_FLAGS_MICRO_TILE; + GLint offset; + GLint nmacroblkpl; + GLint nmicroblkpl; + + x += dPriv->x; + y += dPriv->y; + + if (rrb->has_surface || !(rrb->bo->flags & mask)) { + offset = x * rrb->cpp + y * rrb->pitch; + } else { + offset = 0; + if (rrb->bo->flags & RADEON_BO_FLAGS_MACRO_TILE) { + if (rrb->bo->flags & RADEON_BO_FLAGS_MICRO_TILE) { + nmacroblkpl = rrb->pitch >> 5; + offset += ((y >> 4) * nmacroblkpl) << 11; + offset += ((y & 15) >> 1) << 8; + offset += (y & 1) << 4; + offset += (x >> 5) << 11; + offset += ((x & 31) >> 2) << 5; + offset += (x & 3) << 2; + } else { + nmacroblkpl = rrb->pitch >> 6; + offset += ((y >> 3) * nmacroblkpl) << 11; + offset += (y & 7) << 8; + offset += (x >> 6) << 11; + offset += ((x & 63) >> 3) << 5; + offset += (x & 7) << 2; + } + } else { + nmicroblkpl = ((rrb->pitch + 31) & ~31) >> 5; + offset += (y * nmicroblkpl) << 5; + offset += (x >> 3) << 5; + offset += (x & 7) << 2; + } + } + return &ptr[offset]; +} + +static GLubyte *radeon_ptr16(const struct radeon_renderbuffer * rrb, + GLint x, GLint y) +{ + GLubyte *ptr = rrb->bo->ptr; + const __DRIdrawablePrivate *dPriv = rrb->dPriv; + uint32_t mask = RADEON_BO_FLAGS_MACRO_TILE | RADEON_BO_FLAGS_MICRO_TILE; + GLint offset; + GLint nmacroblkpl; + GLint nmicroblkpl; + + x += dPriv->x; + y += dPriv->y; + + if (rrb->has_surface || !(rrb->bo->flags & mask)) { + offset = x * rrb->cpp + y * rrb->pitch; + } else { + offset = 0; + if (rrb->bo->flags & RADEON_BO_FLAGS_MACRO_TILE) { + if (rrb->bo->flags & RADEON_BO_FLAGS_MICRO_TILE) { + nmacroblkpl = rrb->pitch >> 6; + offset += ((y >> 4) * nmacroblkpl) << 11; + offset += ((y & 15) >> 1) << 8; + offset += (y & 1) << 4; + offset += (x >> 6) << 11; + offset += ((x & 63) >> 3) << 5; + offset += (x & 7) << 1; + } else { + nmacroblkpl = rrb->pitch >> 7; + offset += ((y >> 3) * nmacroblkpl) << 11; + offset += (y & 7) << 8; + offset += (x >> 7) << 11; + offset += ((x & 127) >> 4) << 5; + offset += (x & 15) << 2; + } + } else { + nmicroblkpl = ((rrb->pitch + 31) & ~31) >> 5; + offset += (y * nmicroblkpl) << 5; + offset += (x >> 4) << 5; + offset += (x & 15) << 2; + } + } + return &ptr[offset]; +} + +static GLubyte *radeon_ptr(const struct radeon_renderbuffer * rrb, + GLint x, GLint y) +{ + GLubyte *ptr = rrb->bo->ptr; + const __DRIdrawablePrivate *dPriv = rrb->dPriv; + uint32_t mask = RADEON_BO_FLAGS_MACRO_TILE | RADEON_BO_FLAGS_MICRO_TILE; + GLint offset; + GLint microblkxs; + GLint macroblkxs; + GLint nmacroblkpl; + GLint nmicroblkpl; + + x += dPriv->x; + y += dPriv->y; + + if (rrb->has_surface || !(rrb->bo->flags & mask)) { + offset = x * rrb->cpp + y * rrb->pitch; + } else { + offset = 0; + if (rrb->bo->flags & RADEON_BO_FLAGS_MACRO_TILE) { + if (rrb->bo->flags & RADEON_BO_FLAGS_MICRO_TILE) { + microblkxs = 16 / rrb->cpp; + macroblkxs = 128 / rrb->cpp; + nmacroblkpl = rrb->pitch / macroblkxs; + offset += ((y >> 4) * nmacroblkpl) << 11; + offset += ((y & 15) >> 1) << 8; + offset += (y & 1) << 4; + offset += (x / macroblkxs) << 11; + offset += ((x & (macroblkxs - 1)) / microblkxs) << 5; + offset += (x & (microblkxs - 1)) * rrb->cpp; + } else { + microblkxs = 32 / rrb->cpp; + macroblkxs = 256 / rrb->cpp; + nmacroblkpl = rrb->pitch / macroblkxs; + offset += ((y >> 3) * nmacroblkpl) << 11; + offset += (y & 7) << 8; + offset += (x / macroblkxs) << 11; + offset += ((x & (macroblkxs - 1)) / microblkxs) << 5; + offset += (x & (microblkxs - 1)) * rrb->cpp; + } + } else { + microblkxs = 32 / rrb->cpp; + nmicroblkpl = ((rrb->pitch + 31) & ~31) >> 5; + offset += (y * nmicroblkpl) << 5; + offset += (x / microblkxs) << 5; + offset += (x & (microblkxs - 1)) * rrb->cpp; + } + } + return &ptr[offset]; +} + + /* * Note that all information needed to access pixels in a renderbuffer * should be obtained through the gl_renderbuffer parameter, not per-context * information. */ #define LOCAL_VARS \ - driRenderbuffer *drb = (driRenderbuffer *) rb; \ - const __DRIdrawablePrivate *dPriv = drb->dPriv; \ + struct radeon_renderbuffer *rrb = (void *) rb; \ + const __DRIdrawablePrivate *dPriv = rrb->dPriv; \ const GLuint bottom = dPriv->h - 1; \ - GLubyte *buf = (GLubyte *) drb->flippedData \ - + (dPriv->y * drb->flippedPitch + dPriv->x) * drb->cpp; \ - GLuint p; \ - (void) p; + GLuint p; \ + (void)p; #define LOCAL_DEPTH_VARS \ - driRenderbuffer *drb = (driRenderbuffer *) rb; \ - const __DRIdrawablePrivate *dPriv = drb->dPriv; \ - const GLuint bottom = dPriv->h - 1; \ - GLuint xo = dPriv->x; \ - GLuint yo = dPriv->y; \ - GLubyte *buf = (GLubyte *) drb->Base.Data; + struct radeon_renderbuffer *rrb = (void *) rb; \ + const __DRIdrawablePrivate *dPriv = rrb->dPriv; \ + const GLuint bottom = dPriv->h - 1; #define LOCAL_STENCIL_VARS LOCAL_DEPTH_VARS @@ -94,7 +225,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #define TAG(x) radeon##x##_RGB565 #define TAG2(x,y) radeon##x##_RGB565##y -#define GET_PTR(X,Y) (buf + ((Y) * drb->flippedPitch + (X)) * 2) +#define GET_PTR(X,Y) radeon_ptr16(rrb, (X), (Y)) #include "spantmp2.h" /* 32 bit, ARGB8888 color spanline and pixel functions @@ -104,7 +235,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #define TAG(x) radeon##x##_ARGB8888 #define TAG2(x,y) radeon##x##_ARGB8888##y -#define GET_PTR(X,Y) (buf + ((Y) * drb->flippedPitch + (X)) * 4) +#define GET_PTR(X,Y) radeon_ptr32(rrb, (X), (Y)) #include "spantmp2.h" /* ================================================================ @@ -121,65 +252,15 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. * too... */ -static GLuint radeon_mba_z32(const driRenderbuffer * drb, GLint x, GLint y) -{ - GLuint pitch = drb->pitch; - if (drb->depthHasSurface) { - return 4 * (x + y * pitch); - } else { - GLuint ba, address = 0; /* a[0..1] = 0 */ - -#ifdef COMPILE_R300 - ba = (y / 8) * (pitch / 8) + (x / 8); -#else - ba = (y / 16) * (pitch / 16) + (x / 16); -#endif - - address |= (x & 0x7) << 2; /* a[2..4] = x[0..2] */ - address |= (y & 0x3) << 5; /* a[5..6] = y[0..1] */ - address |= (((x & 0x10) >> 2) ^ (y & 0x4)) << 5; /* a[7] = x[4] ^ y[2] */ - address |= (ba & 0x3) << 8; /* a[8..9] = ba[0..1] */ - - address |= (y & 0x8) << 7; /* a[10] = y[3] */ - address |= (((x & 0x8) << 1) ^ (y & 0x10)) << 7; /* a[11] = x[3] ^ y[4] */ - address |= (ba & ~0x3) << 10; /* a[12..] = ba[2..] */ - - return address; - } -} - -static INLINE GLuint -radeon_mba_z16(const driRenderbuffer * drb, GLint x, GLint y) -{ - GLuint pitch = drb->pitch; - if (drb->depthHasSurface) { - return 2 * (x + y * pitch); - } else { - GLuint ba, address = 0; /* a[0] = 0 */ - - ba = (y / 16) * (pitch / 32) + (x / 32); - - address |= (x & 0x7) << 1; /* a[1..3] = x[0..2] */ - address |= (y & 0x7) << 4; /* a[4..6] = y[0..2] */ - address |= (x & 0x8) << 4; /* a[7] = x[3] */ - address |= (ba & 0x3) << 8; /* a[8..9] = ba[0..1] */ - address |= (y & 0x8) << 7; /* a[10] = y[3] */ - address |= ((x & 0x10) ^ (y & 0x10)) << 7; /* a[11] = x[4] ^ y[4] */ - address |= (ba & ~0x3) << 10; /* a[12..] = ba[2..] */ - - return address; - } -} - /* 16-bit depth buffer functions */ #define VALUE_TYPE GLushort #define WRITE_DEPTH( _x, _y, d ) \ - *(GLushort *)(buf + radeon_mba_z16( drb, _x + xo, _y + yo )) = d; + *(GLushort *)radeon_ptr(rrb, _x, _y) = d #define READ_DEPTH( d, _x, _y ) \ - d = *(GLushort *)(buf + radeon_mba_z16( drb, _x + xo, _y + yo )); + d = *(GLushort *)radeon_ptr(rrb, _x, _y) #define TAG(x) radeon##x##_z16 #include "depthtmp.h" @@ -194,35 +275,36 @@ radeon_mba_z16(const driRenderbuffer * drb, GLint x, GLint y) #ifdef COMPILE_R300 #define WRITE_DEPTH( _x, _y, d ) \ do { \ - GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo ); \ - GLuint tmp = *(GLuint *)(buf + offset); \ + GLuint *_ptr = (GLuint*)radeon_ptr32( rrb, _x, _y ); \ + GLuint tmp = *_ptr; \ tmp &= 0x000000ff; \ tmp |= ((d << 8) & 0xffffff00); \ - *(GLuint *)(buf + offset) = tmp; \ + *_ptr = tmp; \ } while (0) #else #define WRITE_DEPTH( _x, _y, d ) \ do { \ - GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo ); \ - GLuint tmp = *(GLuint *)(buf + offset); \ + GLuint *_ptr = (GLuint*)radeon_ptr32( rrb, _x, _y ); \ + GLuint tmp = *_ptr; \ tmp &= 0xff000000; \ tmp |= ((d) & 0x00ffffff); \ - *(GLuint *)(buf + offset) = tmp; \ + *_ptr = tmp; \ } while (0) #endif #ifdef COMPILE_R300 #define READ_DEPTH( d, _x, _y ) \ do { \ - d = (*(GLuint *)(buf + radeon_mba_z32( drb, _x + xo, \ - _y + yo )) & 0xffffff00) >> 8; \ + d = (*(GLuint*)(radeon_ptr32(rrb, _x, _y)) & 0xffffff00) >> 8; \ }while(0) #else #define READ_DEPTH( d, _x, _y ) \ - d = *(GLuint *)(buf + radeon_mba_z32( drb, _x + xo, \ - _y + yo )) & 0x00ffffff; + d = *(GLuint*)(radeon_ptr32(rrb, _x, _y )) & 0x00ffffff; #endif - +/* + fprintf(stderr, "dval(%d, %d, %d, %d)=0x%08X\n", _x, xo, _y, yo, d);\ + d = *(GLuint*)(radeon_ptr(rrb, _x, _y )) & 0x00ffffff; +*/ #define TAG(x) radeon##x##_z24_s8 #include "depthtmp.h" @@ -235,35 +317,35 @@ do { \ #ifdef COMPILE_R300 #define WRITE_STENCIL( _x, _y, d ) \ do { \ - GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo ); \ - GLuint tmp = *(GLuint *)(buf + offset); \ + GLuint *_ptr = (GLuint*)radeon_ptr32(rrb, _x, _y); \ + GLuint tmp = *_ptr; \ tmp &= 0xffffff00; \ tmp |= (d) & 0xff; \ - *(GLuint *)(buf + offset) = tmp; \ + *_ptr = tmp; \ } while (0) #else #define WRITE_STENCIL( _x, _y, d ) \ do { \ - GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo ); \ - GLuint tmp = *(GLuint *)(buf + offset); \ + GLuint *_ptr = (GLuint*)radeon_ptr32(rrb, _x, _y); \ + GLuint tmp = *_ptr; \ tmp &= 0x00ffffff; \ tmp |= (((d) & 0xff) << 24); \ - *(GLuint *)(buf + offset) = tmp; \ + *_ptr = tmp; \ } while (0) #endif #ifdef COMPILE_R300 #define READ_STENCIL( d, _x, _y ) \ do { \ - GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo ); \ - GLuint tmp = *(GLuint *)(buf + offset); \ + GLuint *_ptr = (GLuint*)radeon_ptr32( rrb, _x, _y ); \ + GLuint tmp = *_ptr; \ d = tmp & 0x000000ff; \ } while (0) #else #define READ_STENCIL( d, _x, _y ) \ do { \ - GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo ); \ - GLuint tmp = *(GLuint *)(buf + offset); \ + GLuint *_ptr = (GLuint*)radeon_ptr32( rrb, _x, _y ); \ + GLuint tmp = *_ptr; \ d = (tmp & 0xff000000) >> 24; \ } while (0) #endif @@ -271,20 +353,60 @@ do { \ #define TAG(x) radeon##x##_z24_s8 #include "stenciltmp.h" -/* Move locking out to get reasonable span performance (10x better - * than doing this in HW_LOCK above). WaitForIdle() is the main - * culprit. - */ + +static void map_buffer(struct gl_renderbuffer *rb, GLboolean write) +{ + struct radeon_renderbuffer *rrb = (void*)rb; + int r; + + if (rrb->bo) { + r = radeon_bo_map(rrb->bo, write); + if (r) { + fprintf(stderr, "(%s) error(%d) mapping buffer.\n", + __FUNCTION__, r); + } + } +} + +static void unmap_buffer(struct gl_renderbuffer *rb) +{ + struct radeon_renderbuffer *rrb = (void*)rb; + + if (rrb->bo) { + radeon_bo_unmap(rrb->bo); + } +} static void radeonSpanRenderStart(GLcontext * ctx) { radeonContextPtr rmesa = RADEON_CONTEXT(ctx); -#ifdef COMPILE_R300 - r300ContextPtr r300 = (r300ContextPtr) rmesa; - R300_FIREVERTICES(r300); -#else - RADEON_FIREVERTICES(rmesa); -#endif + int i; + + radeon_firevertices(rmesa); + + for (i = 0; i < ctx->Const.MaxTextureImageUnits; i++) { + if (ctx->Texture.Unit[i]._ReallyEnabled) + ctx->Driver.MapTexture(ctx, ctx->Texture.Unit[i]._Current); + } + + /* color draw buffers */ + for (i = 0; i < ctx->DrawBuffer->_NumColorDrawBuffers; i++) { + map_buffer(ctx->DrawBuffer->_ColorDrawBuffers[i], GL_TRUE); + } + + map_buffer(ctx->ReadBuffer->_ColorReadBuffer, GL_FALSE); + + if (ctx->DrawBuffer->_DepthBuffer) { + map_buffer(ctx->DrawBuffer->_DepthBuffer->Wrapped, GL_TRUE); + } + if (ctx->DrawBuffer->_StencilBuffer) + map_buffer(ctx->DrawBuffer->_StencilBuffer->Wrapped, GL_TRUE); + + /* The locking and wait for idle should really only be needed in classic mode. + * In a future memory manager based implementation, this should become + * unnecessary due to the fact that mapping our buffers, textures, etc. + * should implicitly wait for any previous rendering commands that must + * be waited on. */ LOCK_HARDWARE(rmesa); radeonWaitForIdleLocked(rmesa); } @@ -292,8 +414,25 @@ static void radeonSpanRenderStart(GLcontext * ctx) static void radeonSpanRenderFinish(GLcontext * ctx) { radeonContextPtr rmesa = RADEON_CONTEXT(ctx); + int i; _swrast_flush(ctx); UNLOCK_HARDWARE(rmesa); + + for (i = 0; i < ctx->Const.MaxTextureImageUnits; i++) { + if (ctx->Texture.Unit[i]._ReallyEnabled) + ctx->Driver.UnmapTexture(ctx, ctx->Texture.Unit[i]._Current); + } + + /* color draw buffers */ + for (i = 0; i < ctx->DrawBuffer->_NumColorDrawBuffers; i++) + unmap_buffer(ctx->DrawBuffer->_ColorDrawBuffers[i]); + + unmap_buffer(ctx->ReadBuffer->_ColorReadBuffer); + + if (ctx->DrawBuffer->_DepthBuffer) + unmap_buffer(ctx->DrawBuffer->_DepthBuffer->Wrapped); + if (ctx->DrawBuffer->_StencilBuffer) + unmap_buffer(ctx->DrawBuffer->_StencilBuffer->Wrapped); } void radeonInitSpanFuncs(GLcontext * ctx) @@ -307,20 +446,17 @@ void radeonInitSpanFuncs(GLcontext * ctx) /** * Plug in the Get/Put routines for the given driRenderbuffer. */ -void radeonSetSpanFunctions(driRenderbuffer * drb, const GLvisual * vis) +void radeonSetSpanFunctions(struct radeon_renderbuffer *rrb) { - if (drb->Base.InternalFormat == GL_RGBA) { - if (vis->redBits == 5 && vis->greenBits == 6 - && vis->blueBits == 5) { - radeonInitPointers_RGB565(&drb->Base); - } else { - radeonInitPointers_ARGB8888(&drb->Base); - } - } else if (drb->Base.InternalFormat == GL_DEPTH_COMPONENT16) { - radeonInitDepthPointers_z16(&drb->Base); - } else if (drb->Base.InternalFormat == GL_DEPTH_COMPONENT24) { - radeonInitDepthPointers_z24_s8(&drb->Base); - } else if (drb->Base.InternalFormat == GL_STENCIL_INDEX8_EXT) { - radeonInitStencilPointers_z24_s8(&drb->Base); + if (rrb->base.InternalFormat == GL_RGB5) { + radeonInitPointers_RGB565(&rrb->base); + } else if (rrb->base.InternalFormat == GL_RGBA8) { + radeonInitPointers_ARGB8888(&rrb->base); + } else if (rrb->base.InternalFormat == GL_DEPTH_COMPONENT16) { + radeonInitDepthPointers_z16(&rrb->base); + } else if (rrb->base.InternalFormat == GL_DEPTH_COMPONENT24) { + radeonInitDepthPointers_z24_s8(&rrb->base); + } else if (rrb->base.InternalFormat == GL_STENCIL_INDEX8_EXT) { + radeonInitStencilPointers_z24_s8(&rrb->base); } } |