diff options
Diffstat (limited to 'src/gallium')
25 files changed, 964 insertions, 662 deletions
diff --git a/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c b/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c index 2ef4293d4d..2f973684f6 100644 --- a/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c +++ b/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c @@ -1,6 +1,6 @@ /************************************************************************** * - * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * Copyright 2007-2009 VMware, Inc. * All Rights Reserved. * * Permission is hereby granted, free of charge, to any person obtaining a @@ -18,7 +18,7 @@ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. - * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. @@ -80,11 +80,27 @@ struct fenced_buffer_list */ struct fenced_buffer { + /* + * Immutable members. + */ + struct pb_buffer base; - struct pb_buffer *buffer; + struct fenced_buffer_list *list; + + /** + * Protected by fenced_buffer_list::mutex + */ + struct list_head head; - /* FIXME: protect access with mutex */ + /** + * Following members are mutable and protected by this mutex. + * + * You may lock this mutex alone, or lock it with fenced_buffer_list::mutex + * held, but in order to prevent deadlocks you must never lock + * fenced_buffer_list::mutex with this mutex held. + */ + pipe_mutex mutex; /** * A bitmask of PIPE_BUFFER_USAGE_CPU/GPU_READ/WRITE describing the current @@ -96,9 +112,6 @@ struct fenced_buffer struct pb_validate *vl; unsigned validation_flags; struct pipe_fence_handle *fence; - - struct list_head head; - struct fenced_buffer_list *list; }; @@ -110,15 +123,24 @@ fenced_buffer(struct pb_buffer *buf) } +/** + * Add the buffer to the fenced list. + * + * fenced_buffer_list::mutex and fenced_buffer::mutex must be held, in this + * order before calling this function. + * + * Reference count should be incremented before calling this function. + */ static INLINE void -_fenced_buffer_add(struct fenced_buffer *fenced_buf) +fenced_buffer_add_locked(struct fenced_buffer_list *fenced_list, + struct fenced_buffer *fenced_buf) { - struct fenced_buffer_list *fenced_list = fenced_buf->list; - assert(pipe_is_referenced(&fenced_buf->base.base.reference)); assert(fenced_buf->flags & PIPE_BUFFER_USAGE_GPU_READ_WRITE); assert(fenced_buf->fence); + /* TODO: Move the reference count increment here */ + #ifdef DEBUG LIST_DEL(&fenced_buf->head); assert(fenced_list->numUnfenced); @@ -130,32 +152,16 @@ _fenced_buffer_add(struct fenced_buffer *fenced_buf) /** - * Actually destroy the buffer. + * Remove the buffer from the fenced list. + * + * fenced_buffer_list::mutex and fenced_buffer::mutex must be held, in this + * order before calling this function. + * + * Reference count should be decremented after calling this function. */ static INLINE void -_fenced_buffer_destroy(struct fenced_buffer *fenced_buf) -{ - struct fenced_buffer_list *fenced_list = fenced_buf->list; - - assert(!pipe_is_referenced(&fenced_buf->base.base.reference)); - assert(!fenced_buf->fence); -#ifdef DEBUG - assert(fenced_buf->head.prev); - assert(fenced_buf->head.next); - LIST_DEL(&fenced_buf->head); - assert(fenced_list->numUnfenced); - --fenced_list->numUnfenced; -#else - (void)fenced_list; -#endif - pb_reference(&fenced_buf->buffer, NULL); - FREE(fenced_buf); -} - - -static INLINE void -_fenced_buffer_remove(struct fenced_buffer_list *fenced_list, - struct fenced_buffer *fenced_buf) +fenced_buffer_remove_locked(struct fenced_buffer_list *fenced_list, + struct fenced_buffer *fenced_buf) { struct pb_fence_ops *ops = fenced_list->ops; @@ -177,37 +183,53 @@ _fenced_buffer_remove(struct fenced_buffer_list *fenced_list, ++fenced_list->numUnfenced; #endif - /** - * FIXME!!! - */ - - if(!pipe_is_referenced(&fenced_buf->base.base.reference)) - _fenced_buffer_destroy(fenced_buf); + /* TODO: Move the reference count decrement and destruction here */ } +/** + * Wait for the fence to expire, and remove it from the fenced list. + * + * fenced_buffer::mutex must be held. fenced_buffer_list::mutex must not be + * held -- it will + */ static INLINE enum pipe_error -_fenced_buffer_finish(struct fenced_buffer *fenced_buf) +fenced_buffer_finish_locked(struct fenced_buffer_list *fenced_list, + struct fenced_buffer *fenced_buf) { - struct fenced_buffer_list *fenced_list = fenced_buf->list; struct pb_fence_ops *ops = fenced_list->ops; + enum pipe_error ret = PIPE_ERROR; #if 0 debug_warning("waiting for GPU"); #endif + assert(pipe_is_referenced(&fenced_buf->base.base.reference)); assert(fenced_buf->fence); + + /* Acquire the global lock */ + pipe_mutex_unlock(fenced_buf->mutex); + pipe_mutex_lock(fenced_list->mutex); + pipe_mutex_lock(fenced_buf->mutex); + if(fenced_buf->fence) { - if(ops->fence_finish(ops, fenced_buf->fence, 0) != 0) { - return PIPE_ERROR; + if(ops->fence_finish(ops, fenced_buf->fence, 0) == 0) { + /* Remove from the fenced list */ + /* TODO: remove consequents */ + fenced_buffer_remove_locked(fenced_list, fenced_buf); + + p_atomic_dec(&fenced_buf->base.base.reference.count); + assert(pipe_is_referenced(&fenced_buf->base.base.reference)); + + fenced_buf->flags &= ~PIPE_BUFFER_USAGE_GPU_READ_WRITE; + + ret = PIPE_OK; } - /* Remove from the fenced list */ - /* TODO: remove consequents */ - _fenced_buffer_remove(fenced_list, fenced_buf); } - fenced_buf->flags &= ~PIPE_BUFFER_USAGE_GPU_READ_WRITE; - return PIPE_OK; + pipe_mutex_unlock(fenced_list->mutex); + + return ret; } @@ -215,7 +237,7 @@ _fenced_buffer_finish(struct fenced_buffer *fenced_buf) * Free as many fenced buffers from the list head as possible. */ static void -_fenced_buffer_list_check_free(struct fenced_buffer_list *fenced_list, +fenced_buffer_list_check_free_locked(struct fenced_buffer_list *fenced_list, int wait) { struct pb_fence_ops *ops = fenced_list->ops; @@ -228,21 +250,28 @@ _fenced_buffer_list_check_free(struct fenced_buffer_list *fenced_list, while(curr != &fenced_list->delayed) { fenced_buf = LIST_ENTRY(struct fenced_buffer, curr, head); + pipe_mutex_lock(fenced_buf->mutex); + if(fenced_buf->fence != prev_fence) { int signaled; if (wait) signaled = ops->fence_finish(ops, fenced_buf->fence, 0); else signaled = ops->fence_signalled(ops, fenced_buf->fence, 0); - if (signaled != 0) + if (signaled != 0) { + pipe_mutex_unlock(fenced_buf->mutex); break; + } prev_fence = fenced_buf->fence; } else { assert(ops->fence_signalled(ops, fenced_buf->fence, 0) == 0); } - _fenced_buffer_remove(fenced_list, fenced_buf); + fenced_buffer_remove_locked(fenced_list, fenced_buf); + pipe_mutex_unlock(fenced_buf->mutex); + + pb_reference((struct pb_buffer **)&fenced_buf, NULL); curr = next; next = curr->next; @@ -256,30 +285,25 @@ fenced_buffer_destroy(struct pb_buffer *buf) struct fenced_buffer *fenced_buf = fenced_buffer(buf); struct fenced_buffer_list *fenced_list = fenced_buf->list; - pipe_mutex_lock(fenced_list->mutex); assert(!pipe_is_referenced(&fenced_buf->base.base.reference)); - if (fenced_buf->fence) { - struct pb_fence_ops *ops = fenced_list->ops; - if(ops->fence_signalled(ops, fenced_buf->fence, 0) == 0) { - struct list_head *curr, *prev; - curr = &fenced_buf->head; - prev = curr->prev; - do { - fenced_buf = LIST_ENTRY(struct fenced_buffer, curr, head); - assert(ops->fence_signalled(ops, fenced_buf->fence, 0) == 0); - _fenced_buffer_remove(fenced_list, fenced_buf); - curr = prev; - prev = curr->prev; - } while (curr != &fenced_list->delayed); - } - else { - /* delay destruction */ - } - } - else { - _fenced_buffer_destroy(fenced_buf); - } + assert(!fenced_buf->fence); + +#ifdef DEBUG + pipe_mutex_lock(fenced_list->mutex); + assert(fenced_buf->head.prev); + assert(fenced_buf->head.next); + LIST_DEL(&fenced_buf->head); + assert(fenced_list->numUnfenced); + --fenced_list->numUnfenced; pipe_mutex_unlock(fenced_list->mutex); +#else + (void)fenced_list; +#endif + + pb_reference(&fenced_buf->buffer, NULL); + + pipe_mutex_destroy(fenced_buf->mutex); + FREE(fenced_buf); } @@ -290,24 +314,23 @@ fenced_buffer_map(struct pb_buffer *buf, struct fenced_buffer *fenced_buf = fenced_buffer(buf); struct fenced_buffer_list *fenced_list = fenced_buf->list; struct pb_fence_ops *ops = fenced_list->ops; - void *map; + void *map = NULL; + + pipe_mutex_lock(fenced_buf->mutex); assert(!(flags & PIPE_BUFFER_USAGE_GPU_READ_WRITE)); /* Serialize writes */ if((fenced_buf->flags & PIPE_BUFFER_USAGE_GPU_WRITE) || ((fenced_buf->flags & PIPE_BUFFER_USAGE_GPU_READ) && (flags & PIPE_BUFFER_USAGE_CPU_WRITE))) { - if(flags & PIPE_BUFFER_USAGE_DONTBLOCK) { + if((flags & PIPE_BUFFER_USAGE_DONTBLOCK) && + ops->fence_signalled(ops, fenced_buf->fence, 0) == 0) { /* Don't wait for the GPU to finish writing */ - if(ops->fence_signalled(ops, fenced_buf->fence, 0) == 0) - _fenced_buffer_remove(fenced_list, fenced_buf); - else - return NULL; - } - else { - /* Wait for the GPU to finish writing */ - _fenced_buffer_finish(fenced_buf); + goto finish; } + + /* Wait for the GPU to finish writing */ + fenced_buffer_finish_locked(fenced_list, fenced_buf); } #if 0 @@ -324,6 +347,9 @@ fenced_buffer_map(struct pb_buffer *buf, fenced_buf->flags |= flags & PIPE_BUFFER_USAGE_CPU_READ_WRITE; } +finish: + pipe_mutex_unlock(fenced_buf->mutex); + return map; } @@ -332,6 +358,9 @@ static void fenced_buffer_unmap(struct pb_buffer *buf) { struct fenced_buffer *fenced_buf = fenced_buffer(buf); + + pipe_mutex_lock(fenced_buf->mutex); + assert(fenced_buf->mapcount); if(fenced_buf->mapcount) { pb_unmap(fenced_buf->buffer); @@ -339,6 +368,8 @@ fenced_buffer_unmap(struct pb_buffer *buf) if(!fenced_buf->mapcount) fenced_buf->flags &= ~PIPE_BUFFER_USAGE_CPU_READ_WRITE; } + + pipe_mutex_unlock(fenced_buf->mutex); } @@ -350,11 +381,14 @@ fenced_buffer_validate(struct pb_buffer *buf, struct fenced_buffer *fenced_buf = fenced_buffer(buf); enum pipe_error ret; + pipe_mutex_lock(fenced_buf->mutex); + if(!vl) { /* invalidate */ fenced_buf->vl = NULL; fenced_buf->validation_flags = 0; - return PIPE_OK; + ret = PIPE_OK; + goto finish; } assert(flags & PIPE_BUFFER_USAGE_GPU_READ_WRITE); @@ -362,14 +396,17 @@ fenced_buffer_validate(struct pb_buffer *buf, flags &= PIPE_BUFFER_USAGE_GPU_READ_WRITE; /* Buffer cannot be validated in two different lists */ - if(fenced_buf->vl && fenced_buf->vl != vl) - return PIPE_ERROR_RETRY; + if(fenced_buf->vl && fenced_buf->vl != vl) { + ret = PIPE_ERROR_RETRY; + goto finish; + } #if 0 /* Do not validate if buffer is still mapped */ if(fenced_buf->flags & PIPE_BUFFER_USAGE_CPU_READ_WRITE) { /* TODO: wait for the thread that mapped the buffer to unmap it */ - return PIPE_ERROR_RETRY; + ret = PIPE_ERROR_RETRY; + goto finish; } /* Final sanity checking */ assert(!(fenced_buf->flags & PIPE_BUFFER_USAGE_CPU_READ_WRITE)); @@ -379,17 +416,21 @@ fenced_buffer_validate(struct pb_buffer *buf, if(fenced_buf->vl == vl && (fenced_buf->validation_flags & flags) == flags) { /* Nothing to do -- buffer already validated */ - return PIPE_OK; + ret = PIPE_OK; + goto finish; } ret = pb_validate(fenced_buf->buffer, vl, flags); if (ret != PIPE_OK) - return ret; + goto finish; fenced_buf->vl = vl; fenced_buf->validation_flags |= flags; - return PIPE_OK; +finish: + pipe_mutex_unlock(fenced_buf->mutex); + + return ret; } @@ -404,29 +445,36 @@ fenced_buffer_fence(struct pb_buffer *buf, fenced_buf = fenced_buffer(buf); fenced_list = fenced_buf->list; ops = fenced_list->ops; - - if(fence == fenced_buf->fence) { - /* Nothing to do */ - return; - } - assert(fenced_buf->vl); - assert(fenced_buf->validation_flags); - pipe_mutex_lock(fenced_list->mutex); - if (fenced_buf->fence) - _fenced_buffer_remove(fenced_list, fenced_buf); - if (fence) { - ops->fence_reference(ops, &fenced_buf->fence, fence); - fenced_buf->flags |= fenced_buf->validation_flags; - _fenced_buffer_add(fenced_buf); - } - pipe_mutex_unlock(fenced_list->mutex); + pipe_mutex_lock(fenced_buf->mutex); + + assert(pipe_is_referenced(&fenced_buf->base.base.reference)); + + if(fence != fenced_buf->fence) { + assert(fenced_buf->vl); + assert(fenced_buf->validation_flags); + + if (fenced_buf->fence) { + fenced_buffer_remove_locked(fenced_list, fenced_buf); + p_atomic_dec(&fenced_buf->base.base.reference.count); + assert(pipe_is_referenced(&fenced_buf->base.base.reference)); + } + if (fence) { + ops->fence_reference(ops, &fenced_buf->fence, fence); + fenced_buf->flags |= fenced_buf->validation_flags; + p_atomic_inc(&fenced_buf->base.base.reference.count); + fenced_buffer_add_locked(fenced_list, fenced_buf); + } + + pb_fence(fenced_buf->buffer, fence); - pb_fence(fenced_buf->buffer, fence); + fenced_buf->vl = NULL; + fenced_buf->validation_flags = 0; + } - fenced_buf->vl = NULL; - fenced_buf->validation_flags = 0; + pipe_mutex_unlock(fenced_buf->mutex); + pipe_mutex_unlock(fenced_list->mutex); } @@ -436,6 +484,7 @@ fenced_buffer_get_base_buffer(struct pb_buffer *buf, pb_size *offset) { struct fenced_buffer *fenced_buf = fenced_buffer(buf); + /* NOTE: accesses immutable members only -- mutex not necessary */ pb_get_base_buffer(fenced_buf->buffer, base_buf, offset); } @@ -475,6 +524,8 @@ fenced_buffer_create(struct fenced_buffer_list *fenced_list, buf->buffer = buffer; buf->list = fenced_list; + pipe_mutex_init(buf->mutex); + #ifdef DEBUG pipe_mutex_lock(fenced_list->mutex); LIST_ADDTAIL(&buf->head, &fenced_list->unfenced); @@ -516,7 +567,7 @@ fenced_buffer_list_check_free(struct fenced_buffer_list *fenced_list, int wait) { pipe_mutex_lock(fenced_list->mutex); - _fenced_buffer_list_check_free(fenced_list, wait); + fenced_buffer_list_check_free_locked(fenced_list, wait); pipe_mutex_unlock(fenced_list->mutex); } @@ -538,11 +589,13 @@ fenced_buffer_list_dump(struct fenced_buffer_list *fenced_list) next = curr->next; while(curr != &fenced_list->unfenced) { fenced_buf = LIST_ENTRY(struct fenced_buffer, curr, head); + pipe_mutex_lock(fenced_buf->mutex); assert(!fenced_buf->fence); debug_printf("%10p %7u %7u\n", (void *) fenced_buf, fenced_buf->base.base.size, p_atomic_read(&fenced_buf->base.base.reference.count)); + pipe_mutex_unlock(fenced_buf->mutex); curr = next; next = curr->next; } @@ -552,6 +605,7 @@ fenced_buffer_list_dump(struct fenced_buffer_list *fenced_list) while(curr != &fenced_list->delayed) { int signaled; fenced_buf = LIST_ENTRY(struct fenced_buffer, curr, head); + pipe_mutex_lock(fenced_buf->mutex); signaled = ops->fence_signalled(ops, fenced_buf->fence, 0); debug_printf("%10p %7u %7u %10p %s\n", (void *) fenced_buf, @@ -559,6 +613,7 @@ fenced_buffer_list_dump(struct fenced_buffer_list *fenced_list) p_atomic_read(&fenced_buf->base.base.reference.count), (void *) fenced_buf->fence, signaled == 0 ? "y" : "n"); + pipe_mutex_unlock(fenced_buf->mutex); curr = next; next = curr->next; } @@ -579,8 +634,8 @@ fenced_buffer_list_destroy(struct fenced_buffer_list *fenced_list) #if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) sched_yield(); #endif - _fenced_buffer_list_check_free(fenced_list, 1); pipe_mutex_lock(fenced_list->mutex); + fenced_buffer_list_check_free_locked(fenced_list, 1); } #ifdef DEBUG @@ -588,6 +643,7 @@ fenced_buffer_list_destroy(struct fenced_buffer_list *fenced_list) #endif pipe_mutex_unlock(fenced_list->mutex); + pipe_mutex_destroy(fenced_list->mutex); fenced_list->ops->destroy(fenced_list->ops); diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c index b7569e74d4..af914f6d08 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_exec.c +++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c @@ -62,9 +62,6 @@ #define FAST_MATH 1 -/** for tgsi_full_instruction::Flags */ -#define SOA_DEPENDENCY_FLAG 0x1 - #define TILE_TOP_LEFT 0 #define TILE_TOP_RIGHT 1 #define TILE_BOTTOM_LEFT 2 @@ -332,20 +329,6 @@ tgsi_exec_machine_bind_shader( maxInstructions += 10; } - if (tgsi_check_soa_dependencies(&parse.FullToken.FullInstruction)) { - uint opcode = parse.FullToken.FullInstruction.Instruction.Opcode; - parse.FullToken.FullInstruction.Flags = SOA_DEPENDENCY_FLAG; - /* XXX we only handle SOA dependencies properly for MOV/SWZ - * at this time! - */ - if (opcode != TGSI_OPCODE_MOV) { - debug_printf("Warning: SOA dependency in instruction" - " is not handled:\n"); - tgsi_dump_instruction(&parse.FullToken.FullInstruction, - numInstructions); - } - } - memcpy(instructions + numInstructions, &parse.FullToken.FullInstruction, sizeof(instructions[0])); @@ -595,6 +578,24 @@ micro_exp2( dst->f[2] = util_fast_exp2( src->f[2] ); dst->f[3] = util_fast_exp2( src->f[3] ); #else + +#if DEBUG + /* Inf is okay for this instruction, so clamp it to silence assertions. */ + uint i; + union tgsi_exec_channel clamped; + + for (i = 0; i < 4; i++) { + if (src->f[i] > 127.99999f) { + clamped.f[i] = 127.99999f; + } else if (src->f[i] < -126.99999f) { + clamped.f[i] = -126.99999f; + } else { + clamped.f[i] = src->f[i]; + } + } + src = &clamped; +#endif + dst->f[0] = powf( 2.0f, src->f[0] ); dst->f[1] = powf( 2.0f, src->f[1] ); dst->f[2] = powf( 2.0f, src->f[2] ); @@ -1478,6 +1479,13 @@ store_dest( dst = &mach->Addrs[index].xyzw[chan_index]; break; + case TGSI_FILE_LOOP: + assert(reg->DstRegister.Index == 0); + assert(mach->LoopCounterStackTop > 0); + assert(chan_index == CHAN_X); + dst = &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[chan_index]; + break; + case TGSI_FILE_PREDICATE: index = reg->DstRegister.Index; assert(index < TGSI_EXEC_NUM_PREDS); @@ -1715,6 +1723,64 @@ exec_tex(struct tgsi_exec_machine *mach, } } +static void +exec_txd(struct tgsi_exec_machine *mach, + const struct tgsi_full_instruction *inst) +{ + const uint unit = inst->FullSrcRegisters[3].SrcRegister.Index; + union tgsi_exec_channel r[4]; + uint chan_index; + + /* + * XXX: This is fake TXD -- the derivatives are not taken into account, yet. + */ + + switch (inst->InstructionExtTexture.Texture) { + case TGSI_TEXTURE_1D: + case TGSI_TEXTURE_SHADOW1D: + + FETCH(&r[0], 0, CHAN_X); + + fetch_texel(mach->Samplers[unit], + &r[0], &ZeroVec, &ZeroVec, 0.0f, /* S, T, P, BIAS */ + &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */ + break; + + case TGSI_TEXTURE_2D: + case TGSI_TEXTURE_RECT: + case TGSI_TEXTURE_SHADOW2D: + case TGSI_TEXTURE_SHADOWRECT: + + FETCH(&r[0], 0, CHAN_X); + FETCH(&r[1], 0, CHAN_Y); + FETCH(&r[2], 0, CHAN_Z); + + fetch_texel(mach->Samplers[unit], + &r[0], &r[1], &r[2], 0.0f, /* inputs */ + &r[0], &r[1], &r[2], &r[3]); /* outputs */ + break; + + case TGSI_TEXTURE_3D: + case TGSI_TEXTURE_CUBE: + + FETCH(&r[0], 0, CHAN_X); + FETCH(&r[1], 0, CHAN_Y); + FETCH(&r[2], 0, CHAN_Z); + + fetch_texel(mach->Samplers[unit], + &r[0], &r[1], &r[2], 0.0f, + &r[0], &r[1], &r[2], &r[3]); + break; + + default: + assert(0); + } + + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&r[chan_index], 0, chan_index); + } +} + /** * Evaluate a constant-valued coefficient at the position of the @@ -1784,53 +1850,58 @@ typedef void (* eval_coef_func)( unsigned chan ); static void -exec_declaration( - struct tgsi_exec_machine *mach, - const struct tgsi_full_declaration *decl ) +exec_declaration(struct tgsi_exec_machine *mach, + const struct tgsi_full_declaration *decl) { - if( mach->Processor == TGSI_PROCESSOR_FRAGMENT ) { - if( decl->Declaration.File == TGSI_FILE_INPUT ) { - unsigned first, last, mask; - eval_coef_func eval; + if (mach->Processor == TGSI_PROCESSOR_FRAGMENT) { + if (decl->Declaration.File == TGSI_FILE_INPUT) { + uint first, last, mask; first = decl->DeclarationRange.First; last = decl->DeclarationRange.Last; mask = decl->Declaration.UsageMask; - switch( decl->Declaration.Interpolate ) { - case TGSI_INTERPOLATE_CONSTANT: - eval = eval_constant_coef; - break; - - case TGSI_INTERPOLATE_LINEAR: - eval = eval_linear_coef; - break; - - case TGSI_INTERPOLATE_PERSPECTIVE: - eval = eval_perspective_coef; - break; + if (decl->Semantic.SemanticName == TGSI_SEMANTIC_POSITION) { + assert(decl->Semantic.SemanticIndex == 0); + assert(first == last); + assert(mask = TGSI_WRITEMASK_XYZW); - default: - assert( 0 ); - return; - } + mach->Inputs[first] = mach->QuadPos; + } else if (decl->Semantic.SemanticName == TGSI_SEMANTIC_FACE) { + uint i; - if( mask == TGSI_WRITEMASK_XYZW ) { - unsigned i, j; + assert(decl->Semantic.SemanticIndex == 0); + assert(first == last); - for( i = first; i <= last; i++ ) { - for( j = 0; j < NUM_CHANNELS; j++ ) { - eval( mach, i, j ); - } + for (i = 0; i < QUAD_SIZE; i++) { + mach->Inputs[first].xyzw[0].f[i] = mach->Face; + } + } else { + eval_coef_func eval; + uint i, j; + + switch (decl->Declaration.Interpolate) { + case TGSI_INTERPOLATE_CONSTANT: + eval = eval_constant_coef; + break; + + case TGSI_INTERPOLATE_LINEAR: + eval = eval_linear_coef; + break; + + case TGSI_INTERPOLATE_PERSPECTIVE: + eval = eval_perspective_coef; + break; + + default: + assert(0); + return; } - } - else { - unsigned i, j; - for( j = 0; j < NUM_CHANNELS; j++ ) { - if( mask & (1 << j) ) { - for( i = first; i <= last; i++ ) { - eval( mach, i, j ); + for (j = 0; j < NUM_CHANNELS; j++) { + if (mask & (1 << j)) { + for (i = first; i <= last; i++) { + eval(mach, i, j); } } } @@ -1847,6 +1918,7 @@ exec_instruction( { uint chan_index; union tgsi_exec_channel r[10]; + union tgsi_exec_channel d[8]; (*pc)++; @@ -1855,42 +1927,27 @@ exec_instruction( case TGSI_OPCODE_FLR: FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); - micro_flr( &r[0], &r[0] ); - STORE( &r[0], 0, chan_index ); + micro_flr(&d[chan_index], &r[0]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; case TGSI_OPCODE_MOV: - if (inst->Flags & SOA_DEPENDENCY_FLAG) { - /* Do all fetches into temp regs, then do all stores to avoid - * intermediate/accidental clobbering. This could be done all the - * time for MOV but for other instructions we'll need more temps... - */ - FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { - FETCH( &r[chan_index], 0, chan_index ); - } - FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { - STORE( &r[chan_index], 0, chan_index ); - } + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH(&d[chan_index], 0, chan_index); } - else { - FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { - FETCH( &r[0], 0, chan_index ); - STORE( &r[0], 0, chan_index ); - } + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + STORE(&d[chan_index], 0, chan_index); } break; case TGSI_OPCODE_LIT: - if (IS_CHANNEL_ENABLED( *inst, CHAN_X )) { - STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_X ); - } - if (IS_CHANNEL_ENABLED( *inst, CHAN_Y ) || IS_CHANNEL_ENABLED( *inst, CHAN_Z )) { FETCH( &r[0], 0, CHAN_X ); if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) { - micro_max( &r[0], &r[0], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C] ); - STORE( &r[0], 0, CHAN_Y ); + micro_max(&d[CHAN_Y], &r[0], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C]); } if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) { @@ -1901,11 +1958,19 @@ exec_instruction( micro_min( &r[2], &r[2], &mach->Temps[TEMP_128_I].xyzw[TEMP_128_C] ); micro_max( &r[2], &r[2], &mach->Temps[TEMP_M128_I].xyzw[TEMP_M128_C] ); micro_pow( &r[1], &r[1], &r[2] ); - micro_lt( &r[0], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], &r[0], &r[1], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C] ); - STORE( &r[0], 0, CHAN_Z ); + micro_lt(&d[CHAN_Z], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], &r[0], &r[1], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C]); } - } + if (IS_CHANNEL_ENABLED(*inst, CHAN_Y)) { + STORE(&d[CHAN_Y], 0, CHAN_Y); + } + if (IS_CHANNEL_ENABLED(*inst, CHAN_Z)) { + STORE(&d[CHAN_Z], 0, CHAN_Z); + } + } + if (IS_CHANNEL_ENABLED( *inst, CHAN_X )) { + STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_X ); + } if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) { STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W ); } @@ -1973,14 +2038,13 @@ exec_instruction( break; case TGSI_OPCODE_MUL: - FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) - { + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { FETCH(&r[0], 0, chan_index); FETCH(&r[1], 1, chan_index); - - micro_mul( &r[0], &r[0], &r[1] ); - - STORE(&r[0], 0, chan_index); + micro_mul(&d[chan_index], &r[0], &r[1]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -1988,8 +2052,10 @@ exec_instruction( FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); FETCH( &r[1], 1, chan_index ); - micro_add( &r[0], &r[0], &r[1] ); - STORE( &r[0], 0, chan_index ); + micro_add(&d[chan_index], &r[0], &r[1]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -2045,25 +2111,29 @@ exec_instruction( break; case TGSI_OPCODE_DST: - if (IS_CHANNEL_ENABLED( *inst, CHAN_X )) { - STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_X ); - } - if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) { FETCH( &r[0], 0, CHAN_Y ); FETCH( &r[1], 1, CHAN_Y); - micro_mul( &r[0], &r[0], &r[1] ); - STORE( &r[0], 0, CHAN_Y ); + micro_mul(&d[CHAN_Y], &r[0], &r[1]); } - if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) { - FETCH( &r[0], 0, CHAN_Z ); - STORE( &r[0], 0, CHAN_Z ); + FETCH(&d[CHAN_Z], 0, CHAN_Z); } - if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) { - FETCH( &r[0], 1, CHAN_W ); - STORE( &r[0], 0, CHAN_W ); + FETCH(&d[CHAN_W], 1, CHAN_W); + } + + if (IS_CHANNEL_ENABLED(*inst, CHAN_X)) { + STORE(&mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_X); + } + if (IS_CHANNEL_ENABLED(*inst, CHAN_Y)) { + STORE(&d[CHAN_Y], 0, CHAN_Y); + } + if (IS_CHANNEL_ENABLED(*inst, CHAN_Z)) { + STORE(&d[CHAN_Z], 0, CHAN_Z); + } + if (IS_CHANNEL_ENABLED(*inst, CHAN_W)) { + STORE(&d[CHAN_W], 0, CHAN_W); } break; @@ -2073,9 +2143,10 @@ exec_instruction( FETCH(&r[1], 1, chan_index); /* XXX use micro_min()?? */ - micro_lt( &r[0], &r[0], &r[1], &r[0], &r[1] ); - - STORE(&r[0], 0, chan_index); + micro_lt(&d[chan_index], &r[0], &r[1], &r[0], &r[1]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -2085,9 +2156,10 @@ exec_instruction( FETCH(&r[1], 1, chan_index); /* XXX use micro_max()?? */ - micro_lt( &r[0], &r[0], &r[1], &r[1], &r[0] ); - - STORE(&r[0], 0, chan_index ); + micro_lt(&d[chan_index], &r[0], &r[1], &r[1], &r[0] ); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -2096,8 +2168,10 @@ exec_instruction( FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); FETCH( &r[1], 1, chan_index ); - micro_lt( &r[0], &r[0], &r[1], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C] ); - STORE( &r[0], 0, chan_index ); + micro_lt(&d[chan_index], &r[0], &r[1], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -2106,8 +2180,10 @@ exec_instruction( FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); FETCH( &r[1], 1, chan_index ); - micro_le( &r[0], &r[1], &r[0], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C] ); - STORE( &r[0], 0, chan_index ); + micro_le(&d[chan_index], &r[1], &r[0], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -2118,8 +2194,10 @@ exec_instruction( FETCH( &r[1], 1, chan_index ); micro_mul( &r[0], &r[0], &r[1] ); FETCH( &r[1], 2, chan_index ); - micro_add( &r[0], &r[0], &r[1] ); - STORE( &r[0], 0, chan_index ); + micro_add(&d[chan_index], &r[0], &r[1]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -2127,10 +2205,10 @@ exec_instruction( FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH(&r[0], 0, chan_index); FETCH(&r[1], 1, chan_index); - - micro_sub( &r[0], &r[0], &r[1] ); - - STORE(&r[0], 0, chan_index); + micro_sub(&d[chan_index], &r[0], &r[1]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -2139,12 +2217,12 @@ exec_instruction( FETCH(&r[0], 0, chan_index); FETCH(&r[1], 1, chan_index); FETCH(&r[2], 2, chan_index); - micro_sub( &r[1], &r[1], &r[2] ); micro_mul( &r[0], &r[0], &r[1] ); - micro_add( &r[0], &r[0], &r[2] ); - - STORE(&r[0], 0, chan_index); + micro_add(&d[chan_index], &r[0], &r[2]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -2153,8 +2231,10 @@ exec_instruction( FETCH(&r[0], 0, chan_index); FETCH(&r[1], 1, chan_index); FETCH(&r[2], 2, chan_index); - micro_lt(&r[0], &mach->Temps[TEMP_HALF_I].xyzw[TEMP_HALF_C], &r[2], &r[0], &r[1]); - STORE(&r[0], 0, chan_index); + micro_lt(&d[chan_index], &mach->Temps[TEMP_HALF_I].xyzw[TEMP_HALF_C], &r[2], &r[0], &r[1]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -2179,8 +2259,10 @@ exec_instruction( case TGSI_OPCODE_FRC: FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); - micro_frc( &r[0], &r[0] ); - STORE( &r[0], 0, chan_index ); + micro_frc(&d[chan_index], &r[0]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -2190,8 +2272,10 @@ exec_instruction( FETCH(&r[1], 1, chan_index); micro_max(&r[0], &r[0], &r[1]); FETCH(&r[1], 2, chan_index); - micro_min(&r[0], &r[0], &r[1]); - STORE(&r[0], 0, chan_index); + micro_min(&d[chan_index], &r[0], &r[1]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -2199,19 +2283,17 @@ exec_instruction( case TGSI_OPCODE_ARR: FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); - micro_rnd( &r[0], &r[0] ); - STORE( &r[0], 0, chan_index ); + micro_rnd(&d[chan_index], &r[0]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; case TGSI_OPCODE_EX2: FETCH(&r[0], 0, CHAN_X); -#if FAST_MATH micro_exp2( &r[0], &r[0] ); -#else - micro_pow( &r[0], &mach->Temps[TEMP_2_I].xyzw[TEMP_2_C], &r[0] ); -#endif FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { STORE( &r[0], 0, chan_index ); @@ -2247,11 +2329,7 @@ exec_instruction( FETCH(&r[4], 1, CHAN_Y); micro_mul( &r[5], &r[3], &r[4] ); - micro_sub( &r[2], &r[2], &r[5] ); - - if (IS_CHANNEL_ENABLED( *inst, CHAN_X )) { - STORE( &r[2], 0, CHAN_X ); - } + micro_sub(&d[CHAN_X], &r[2], &r[5]); FETCH(&r[2], 1, CHAN_X); @@ -2260,20 +2338,21 @@ exec_instruction( FETCH(&r[5], 0, CHAN_X); micro_mul( &r[1], &r[1], &r[5] ); - micro_sub( &r[3], &r[3], &r[1] ); - - if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) { - STORE( &r[3], 0, CHAN_Y ); - } + micro_sub(&d[CHAN_Y], &r[3], &r[1]); micro_mul( &r[5], &r[5], &r[4] ); micro_mul( &r[0], &r[0], &r[2] ); - micro_sub( &r[5], &r[5], &r[0] ); + micro_sub(&d[CHAN_Z], &r[5], &r[0]); - if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) { - STORE( &r[5], 0, CHAN_Z ); + if (IS_CHANNEL_ENABLED(*inst, CHAN_X)) { + STORE(&d[CHAN_X], 0, CHAN_X); + } + if (IS_CHANNEL_ENABLED(*inst, CHAN_Y)) { + STORE(&d[CHAN_Y], 0, CHAN_Y); + } + if (IS_CHANNEL_ENABLED(*inst, CHAN_Z)) { + STORE(&d[CHAN_Z], 0, CHAN_Z); } - if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) { STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W ); } @@ -2282,11 +2361,11 @@ exec_instruction( case TGSI_OPCODE_ABS: FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH(&r[0], 0, chan_index); - - micro_abs( &r[0], &r[0] ); - - STORE(&r[0], 0, chan_index); + micro_abs(&d[chan_index], &r[0]); } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); + } break; case TGSI_OPCODE_RCC: @@ -2338,16 +2417,20 @@ exec_instruction( case TGSI_OPCODE_DDX: FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); - micro_ddx( &r[0], &r[0] ); - STORE( &r[0], 0, chan_index ); + micro_ddx(&d[chan_index], &r[0]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; case TGSI_OPCODE_DDY: FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); - micro_ddy( &r[0], &r[0] ); - STORE( &r[0], 0, chan_index ); + micro_ddy(&d[chan_index], &r[0]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -2428,10 +2511,10 @@ exec_instruction( FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); FETCH( &r[1], 1, chan_index ); - micro_eq( &r[0], &r[0], &r[1], - &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], - &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C] ); - STORE( &r[0], 0, chan_index ); + micro_eq(&d[chan_index], &r[0], &r[1], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -2445,8 +2528,10 @@ exec_instruction( FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); FETCH( &r[1], 1, chan_index ); - micro_le( &r[0], &r[0], &r[1], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C] ); - STORE( &r[0], 0, chan_index ); + micro_le(&d[chan_index], &r[0], &r[1], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -2462,8 +2547,10 @@ exec_instruction( FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); FETCH( &r[1], 1, chan_index ); - micro_le( &r[0], &r[0], &r[1], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C] ); - STORE( &r[0], 0, chan_index ); + micro_le(&d[chan_index], &r[0], &r[1], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -2471,8 +2558,10 @@ exec_instruction( FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); FETCH( &r[1], 1, chan_index ); - micro_eq( &r[0], &r[0], &r[1], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C] ); - STORE( &r[0], 0, chan_index ); + micro_eq(&d[chan_index], &r[0], &r[1], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -2502,7 +2591,7 @@ exec_instruction( /* src[1] = d[strq]/dx */ /* src[2] = d[strq]/dy */ /* src[3] = sampler unit */ - assert (0); + exec_txd(mach, inst); break; case TGSI_OPCODE_TXL: @@ -2546,13 +2635,8 @@ exec_instruction( micro_mul(&r[3], &r[3], &r[1]); micro_add(&r[2], &r[2], &r[3]); FETCH(&r[3], 0, CHAN_X); - micro_add(&r[2], &r[2], &r[3]); - if (IS_CHANNEL_ENABLED(*inst, CHAN_X)) { - STORE(&r[2], 0, CHAN_X); - } - if (IS_CHANNEL_ENABLED(*inst, CHAN_Z)) { - STORE(&r[2], 0, CHAN_Z); - } + micro_add(&d[CHAN_X], &r[2], &r[3]); + } if (IS_CHANNEL_ENABLED(*inst, CHAN_Y) || IS_CHANNEL_ENABLED(*inst, CHAN_W)) { @@ -2562,13 +2646,20 @@ exec_instruction( micro_mul(&r[3], &r[3], &r[1]); micro_add(&r[2], &r[2], &r[3]); FETCH(&r[3], 0, CHAN_Y); - micro_add(&r[2], &r[2], &r[3]); - if (IS_CHANNEL_ENABLED(*inst, CHAN_Y)) { - STORE(&r[2], 0, CHAN_Y); - } - if (IS_CHANNEL_ENABLED(*inst, CHAN_W)) { - STORE(&r[2], 0, CHAN_W); - } + micro_add(&d[CHAN_Y], &r[2], &r[3]); + + } + if (IS_CHANNEL_ENABLED(*inst, CHAN_X)) { + STORE(&d[CHAN_X], 0, CHAN_X); + } + if (IS_CHANNEL_ENABLED(*inst, CHAN_Y)) { + STORE(&d[CHAN_Y], 0, CHAN_Y); + } + if (IS_CHANNEL_ENABLED(*inst, CHAN_Z)) { + STORE(&d[CHAN_X], 0, CHAN_Z); + } + if (IS_CHANNEL_ENABLED(*inst, CHAN_W)) { + STORE(&d[CHAN_Y], 0, CHAN_W); } break; @@ -2653,8 +2744,10 @@ exec_instruction( /* TGSI_OPCODE_SGN */ FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); - micro_sgn( &r[0], &r[0] ); - STORE( &r[0], 0, chan_index ); + micro_sgn(&d[chan_index], &r[0]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -2663,10 +2756,10 @@ exec_instruction( FETCH(&r[0], 0, chan_index); FETCH(&r[1], 1, chan_index); FETCH(&r[2], 2, chan_index); - - micro_lt( &r[0], &r[0], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], &r[1], &r[2] ); - - STORE(&r[0], 0, chan_index); + micro_lt(&d[chan_index], &r[0], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], &r[1], &r[2]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -2841,32 +2934,40 @@ exec_instruction( case TGSI_OPCODE_CEIL: FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); - micro_ceil( &r[0], &r[0] ); - STORE( &r[0], 0, chan_index ); + micro_ceil(&d[chan_index], &r[0]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; case TGSI_OPCODE_I2F: FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); - micro_i2f( &r[0], &r[0] ); - STORE( &r[0], 0, chan_index ); + micro_i2f(&d[chan_index], &r[0]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; case TGSI_OPCODE_NOT: FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); - micro_not( &r[0], &r[0] ); - STORE( &r[0], 0, chan_index ); + micro_not(&d[chan_index], &r[0]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; case TGSI_OPCODE_TRUNC: FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); - micro_trunc( &r[0], &r[0] ); - STORE( &r[0], 0, chan_index ); + micro_trunc(&d[chan_index], &r[0]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -2874,8 +2975,10 @@ exec_instruction( FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); FETCH( &r[1], 1, chan_index ); - micro_shl( &r[0], &r[0], &r[1] ); - STORE( &r[0], 0, chan_index ); + micro_shl(&d[chan_index], &r[0], &r[1]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -2883,8 +2986,10 @@ exec_instruction( FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); FETCH( &r[1], 1, chan_index ); - micro_ishr( &r[0], &r[0], &r[1] ); - STORE( &r[0], 0, chan_index ); + micro_ishr(&d[chan_index], &r[0], &r[1]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -2892,8 +2997,10 @@ exec_instruction( FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); FETCH( &r[1], 1, chan_index ); - micro_and( &r[0], &r[0], &r[1] ); - STORE( &r[0], 0, chan_index ); + micro_and(&d[chan_index], &r[0], &r[1]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -2901,8 +3008,10 @@ exec_instruction( FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); FETCH( &r[1], 1, chan_index ); - micro_or( &r[0], &r[0], &r[1] ); - STORE( &r[0], 0, chan_index ); + micro_or(&d[chan_index], &r[0], &r[1]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -2914,8 +3023,10 @@ exec_instruction( FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( &r[0], 0, chan_index ); FETCH( &r[1], 1, chan_index ); - micro_xor( &r[0], &r[0], &r[1] ); - STORE( &r[0], 0, chan_index ); + micro_xor(&d[chan_index], &r[0], &r[1]); + } + FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) { + STORE(&d[chan_index], 0, chan_index); } break; @@ -2946,8 +3057,23 @@ exec_instruction( for (chan_index = 0; chan_index < 3; chan_index++) { FETCH( &mach->LoopCounterStack[mach->LoopCounterStackTop].xyzw[chan_index], 0, chan_index ); } - STORE( &mach->LoopCounterStack[mach->LoopCounterStackTop].xyzw[CHAN_Y], 0, CHAN_X ); ++mach->LoopCounterStackTop; + STORE(&mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X], 0, CHAN_X); + /* update LoopMask */ + if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[0] <= 0.0f) { + mach->LoopMask &= ~0x1; + } + if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[1] <= 0.0f) { + mach->LoopMask &= ~0x2; + } + if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[2] <= 0.0f) { + mach->LoopMask &= ~0x4; + } + if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[3] <= 0.0f) { + mach->LoopMask &= ~0x8; + } + /* TODO: if mach->LoopMask == 0, jump to end of loop */ + UPDATE_EXEC_MASK(mach); /* fall-through (for now) */ case TGSI_OPCODE_BGNLOOP: /* push LoopMask and ContMasks */ @@ -2961,28 +3087,28 @@ exec_instruction( case TGSI_OPCODE_ENDFOR: assert(mach->LoopCounterStackTop > 0); - micro_sub( &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X], - &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X], - &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C] ); + micro_sub(&mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y], + &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y], + &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C]); /* update LoopMask */ - if( mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X].f[0] <= 0) { + if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[0] <= 0.0f) { mach->LoopMask &= ~0x1; } - if( mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X].f[1] <= 0 ) { + if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[1] <= 0.0f) { mach->LoopMask &= ~0x2; } - if( mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X].f[2] <= 0 ) { + if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[2] <= 0.0f) { mach->LoopMask &= ~0x4; } - if( mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X].f[3] <= 0 ) { + if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[3] <= 0.0f) { mach->LoopMask &= ~0x8; } - micro_add( &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y], - &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y], - &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Z]); + micro_add(&mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X], + &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X], + &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Z]); assert(mach->LoopLabelStackTop > 0); inst = mach->Instructions + mach->LoopLabelStack[mach->LoopLabelStackTop - 1]; - STORE( &mach->LoopCounterStack[mach->LoopCounterStackTop].xyzw[CHAN_Y], 0, CHAN_X ); + STORE(&mach->LoopCounterStack[mach->LoopCounterStackTop].xyzw[CHAN_X], 0, CHAN_X); /* Restore ContMask, but don't pop */ assert(mach->ContStackTop > 0); mach->ContMask = mach->ContStack[mach->ContStackTop - 1]; @@ -3060,6 +3186,8 @@ exec_instruction( } } +#define DEBUG_EXECUTION 0 + /** * Run TGSI interpreter. @@ -3103,10 +3231,67 @@ tgsi_exec_machine_run( struct tgsi_exec_machine *mach ) exec_declaration( mach, mach->Declarations+i ); } - /* execute instructions, until pc is set to -1 */ - while (pc != -1) { - assert(pc < (int) mach->NumInstructions); - exec_instruction( mach, mach->Instructions + pc, &pc ); + { +#if DEBUG_EXECUTION + struct tgsi_exec_vector temps[TGSI_EXEC_NUM_TEMPS + TGSI_EXEC_NUM_TEMP_EXTRAS]; + struct tgsi_exec_vector outputs[PIPE_MAX_ATTRIBS]; + uint inst = 1; + + memcpy(temps, mach->Temps, sizeof(temps)); + memcpy(outputs, mach->Outputs, sizeof(outputs)); +#endif + + /* execute instructions, until pc is set to -1 */ + while (pc != -1) { + +#if DEBUG_EXECUTION + uint i; + + tgsi_dump_instruction(&mach->Instructions[pc], inst++); +#endif + + assert(pc < (int) mach->NumInstructions); + exec_instruction(mach, mach->Instructions + pc, &pc); + +#if DEBUG_EXECUTION + for (i = 0; i < TGSI_EXEC_NUM_TEMPS + TGSI_EXEC_NUM_TEMP_EXTRAS; i++) { + if (memcmp(&temps[i], &mach->Temps[i], sizeof(temps[i]))) { + uint j; + + memcpy(&temps[i], &mach->Temps[i], sizeof(temps[i])); + debug_printf("TEMP[%2u] = ", i); + for (j = 0; j < 4; j++) { + if (j > 0) { + debug_printf(" "); + } + debug_printf("(%6f, %6f, %6f, %6f)\n", + temps[i].xyzw[0].f[j], + temps[i].xyzw[1].f[j], + temps[i].xyzw[2].f[j], + temps[i].xyzw[3].f[j]); + } + } + } + for (i = 0; i < PIPE_MAX_ATTRIBS; i++) { + if (memcmp(&outputs[i], &mach->Outputs[i], sizeof(outputs[i]))) { + uint j; + + memcpy(&outputs[i], &mach->Outputs[i], sizeof(outputs[i])); + debug_printf("OUT[%2u] = ", i); + for (j = 0; j < 4; j++) { + if (j > 0) { + debug_printf(" "); + } + debug_printf("{%6f, %6f, %6f, %6f}\n", + outputs[i].xyzw[0].f[j], + outputs[i].xyzw[1].f[j], + outputs[i].xyzw[2].f[j], + outputs[i].xyzw[3].f[j]); + } + } + } +#endif + } } #if 0 diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.h b/src/gallium/auxiliary/tgsi/tgsi_exec.h index 471f591dd6..3dff69a505 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_exec.h +++ b/src/gallium/auxiliary/tgsi/tgsi_exec.h @@ -232,6 +232,7 @@ struct tgsi_exec_machine /* FRAGMENT processor only. */ const struct tgsi_interp_coef *InterpCoefs; struct tgsi_exec_vector QuadPos; + float Face; /**< +1 if front facing, -1 if back facing */ /* Conditional execution masks */ uint CondMask; /**< For IF/ELSE/ENDIF */ @@ -252,7 +253,7 @@ struct tgsi_exec_machine uint LoopLabelStack[TGSI_EXEC_MAX_LOOP_NESTING]; int LoopLabelStackTop; - /** Loop counter stack (x = count, y = current, z = step) */ + /** Loop counter stack (x = index, y = counter, z = step) */ struct tgsi_exec_vector LoopCounterStack[TGSI_EXEC_MAX_LOOP_NESTING]; int LoopCounterStackTop; diff --git a/src/gallium/auxiliary/util/u_math.h b/src/gallium/auxiliary/util/u_math.h index 75b075f160..731a11475e 100644 --- a/src/gallium/auxiliary/util/u_math.h +++ b/src/gallium/auxiliary/util/u_math.h @@ -499,6 +499,9 @@ util_next_power_of_two(unsigned x) #define MIN2( A, B ) ( (A)<(B) ? (A) : (B) ) #define MAX2( A, B ) ( (A)>(B) ? (A) : (B) ) +#define MIN3( A, B, C ) MIN2( MIN2( A, B ), C ) +#define MAX3( A, B, C ) MAX2( MAX2( A, B ), C ) + static INLINE int align(int value, int alignment) diff --git a/src/gallium/drivers/identity/id_objects.c b/src/gallium/drivers/identity/id_objects.c index e893e59940..bc9bc7121d 100644 --- a/src/gallium/drivers/identity/id_objects.c +++ b/src/gallium/drivers/identity/id_objects.c @@ -180,3 +180,42 @@ identity_transfer_destroy(struct identity_transfer *id_transfer) screen->tex_transfer_destroy(id_transfer->transfer); FREE(id_transfer); } + +struct pipe_video_surface * +identity_video_surface_create(struct identity_screen *id_screen, + struct pipe_video_surface *video_surface) +{ + struct identity_video_surface *id_video_surface; + + if (!video_surface) { + goto error; + } + + assert(video_surface->screen == id_screen->screen); + + id_video_surface = CALLOC_STRUCT(identity_video_surface); + if (!id_video_surface) { + goto error; + } + + memcpy(&id_video_surface->base, + video_surface, + sizeof(struct pipe_video_surface)); + + pipe_reference_init(&id_video_surface->base.reference, 1); + id_video_surface->base.screen = &id_screen->base; + id_video_surface->video_surface = video_surface; + + return &id_video_surface->base; + +error: + pipe_video_surface_reference(&video_surface, NULL); + return NULL; +} + +void +identity_video_surface_destroy(struct identity_video_surface *id_video_surface) +{ + pipe_video_surface_reference(&id_video_surface->video_surface, NULL); + FREE(id_video_surface); +} diff --git a/src/gallium/drivers/identity/id_objects.h b/src/gallium/drivers/identity/id_objects.h index ce58faa3c7..77cc719079 100644 --- a/src/gallium/drivers/identity/id_objects.h +++ b/src/gallium/drivers/identity/id_objects.h @@ -31,6 +31,7 @@ #include "pipe/p_compiler.h" #include "pipe/p_state.h" +#include "pipe/p_video_state.h" #include "id_screen.h" @@ -67,6 +68,14 @@ struct identity_transfer }; +struct identity_video_surface +{ + struct pipe_video_surface base; + + struct pipe_video_surface *video_surface; +}; + + static INLINE struct identity_buffer * identity_buffer(struct pipe_buffer *_buffer) { @@ -103,6 +112,15 @@ identity_transfer(struct pipe_transfer *_transfer) return (struct identity_transfer *)_transfer; } +static INLINE struct identity_video_surface * +identity_video_surface(struct pipe_video_surface *_video_surface) +{ + if (!_video_surface) { + return NULL; + } + (void)identity_screen(_video_surface->screen); + return (struct identity_video_surface *)_video_surface; +} static INLINE struct pipe_buffer * identity_buffer_unwrap(struct pipe_buffer *_buffer) @@ -165,5 +183,12 @@ identity_transfer_create(struct identity_texture *id_texture, void identity_transfer_destroy(struct identity_transfer *id_transfer); +struct pipe_video_surface * +identity_video_surface_create(struct identity_screen *id_screen, + struct pipe_video_surface *video_surface); + +void +identity_video_surface_destroy(struct identity_video_surface *id_video_surface); + #endif /* ID_OBJECTS_H */ diff --git a/src/gallium/drivers/identity/id_public.h b/src/gallium/drivers/identity/id_public.h index cac14cfd60..3d2862eaa0 100644 --- a/src/gallium/drivers/identity/id_public.h +++ b/src/gallium/drivers/identity/id_public.h @@ -37,4 +37,4 @@ identity_screen_create(struct pipe_screen *screen); struct pipe_context * identity_context_create(struct pipe_screen *screen, struct pipe_context *pipe); -#endif /* PT_PUBLIC_H */ +#endif /* ID_PUBLIC_H */ diff --git a/src/gallium/drivers/identity/id_screen.c b/src/gallium/drivers/identity/id_screen.c index 26439637d0..53eae3ef54 100644 --- a/src/gallium/drivers/identity/id_screen.c +++ b/src/gallium/drivers/identity/id_screen.c @@ -379,6 +379,33 @@ identity_screen_buffer_destroy(struct pipe_buffer *_buffer) identity_buffer_destroy(identity_buffer(_buffer)); } +static struct pipe_video_surface * +identity_screen_video_surface_create(struct pipe_screen *_screen, + enum pipe_video_chroma_format chroma_format, + unsigned width, + unsigned height) +{ + struct identity_screen *id_screen = identity_screen(_screen); + struct pipe_screen *screen = id_screen->screen; + struct pipe_video_surface *result; + + result = screen->video_surface_create(screen, + chroma_format, + width, + height); + + if (result) { + return identity_video_surface_create(id_screen, result); + } + return NULL; +} + +static void +identity_screen_video_surface_destroy(struct pipe_video_surface *_vsfc) +{ + identity_video_surface_destroy(identity_video_surface(_vsfc)); +} + static void identity_screen_flush_frontbuffer(struct pipe_screen *_screen, struct pipe_surface *_surface, @@ -472,6 +499,12 @@ identity_screen_create(struct pipe_screen *screen) if (screen->buffer_unmap) id_screen->base.buffer_unmap = identity_screen_buffer_unmap; id_screen->base.buffer_destroy = identity_screen_buffer_destroy; + if (screen->video_surface_create) { + id_screen->base.video_surface_create = identity_screen_video_surface_create; + } + if (screen->video_surface_destroy) { + id_screen->base.video_surface_destroy = identity_screen_video_surface_destroy; + } id_screen->base.flush_frontbuffer = identity_screen_flush_frontbuffer; id_screen->base.fence_reference = identity_screen_fence_reference; id_screen->base.fence_signalled = identity_screen_fence_signalled; diff --git a/src/gallium/drivers/r300/r300_context.c b/src/gallium/drivers/r300/r300_context.c index ae23329b83..769733b6dd 100644 --- a/src/gallium/drivers/r300/r300_context.c +++ b/src/gallium/drivers/r300/r300_context.c @@ -69,6 +69,7 @@ static void r300_destroy_context(struct pipe_context* context) FREE(r300->blend_color_state); FREE(r300->rs_block); FREE(r300->scissor_state); + FREE(r300->vertex_info); FREE(r300->viewport_state); FREE(r300); } @@ -123,15 +124,24 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen, r300->context.clear = r300_clear; - if (r300screen->caps->has_tcl) - { + if (r300screen->caps->has_tcl) { r300->context.draw_arrays = r300_draw_arrays; r300->context.draw_elements = r300_draw_elements; r300->context.draw_range_elements = r300_draw_range_elements; - } - else - { - assert(0); + } else { + r300->context.draw_arrays = r300_swtcl_draw_arrays; + r300->context.draw_elements = r300_draw_elements; + r300->context.draw_range_elements = r300_swtcl_draw_range_elements; + + /* Create a Draw. This is used for SW TCL. */ + r300->draw = draw_create(); + /* Enable our renderer. */ + draw_set_rasterize_stage(r300->draw, r300_draw_stage(r300)); + /* Enable Draw's clipping. */ + draw_set_driver_clipping(r300->draw, FALSE); + /* Force Draw to never do viewport transform, since we can do + * transform in hardware, always. */ + draw_set_viewport_state(r300->draw, &r300_viewport_identity); } r300->context.is_texture_referenced = r300_is_texture_referenced; @@ -143,18 +153,9 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen, r300->blend_color_state = CALLOC_STRUCT(r300_blend_color_state); r300->rs_block = CALLOC_STRUCT(r300_rs_block); r300->scissor_state = CALLOC_STRUCT(r300_scissor_state); + r300->vertex_info = CALLOC_STRUCT(r300_vertex_info); r300->viewport_state = CALLOC_STRUCT(r300_viewport_state); - /* Create a Draw. This is used for vert collation and SW TCL. */ - r300->draw = draw_create(); - /* Enable our renderer. */ - draw_set_rasterize_stage(r300->draw, r300_draw_stage(r300)); - /* Disable Draw's clipping if TCL is present. */ - draw_set_driver_clipping(r300->draw, r300_screen(screen)->caps->has_tcl); - /* Force Draw to never do viewport transform, since (again) we can do - * transform in hardware, always. */ - draw_set_viewport_state(r300->draw, &r300_viewport_identity); - /* Open up the OQ BO. */ r300->oqbo = screen->buffer_create(screen, 4096, PIPE_BUFFER_USAGE_VERTEX, 4096); diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h index f954ba7f9a..39c0914cff 100644 --- a/src/gallium/drivers/r300/r300_context.h +++ b/src/gallium/drivers/r300/r300_context.h @@ -92,6 +92,10 @@ struct r300_sampler_state { uint32_t filter0; /* R300_TX_FILTER0: 0x4400 */ uint32_t filter1; /* R300_TX_FILTER1: 0x4440 */ uint32_t border_color; /* R300_TX_BORDER_COLOR: 0x45c0 */ + + /* Min/max LOD must be clamped to [0, last_level], thus + * it's dependent on a currently bound texture */ + unsigned min_lod, max_lod; }; struct r300_scissor_state { @@ -219,11 +223,6 @@ struct r300_texture { struct r300_vertex_info { /* Parent class */ struct vertex_info vinfo; - /* Map of vertex attributes into PVS memory for HW TCL, - * or GA memory for SW TCL. */ - int vs_tab[16]; - /* Map of rasterizer attributes from GB through RS to US. */ - int fs_tab[16]; /* R300_VAP_PROG_STREAK_CNTL_[0-7] */ uint32_t vap_prog_stream_cntl[8]; diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c index eeb97a2d37..ad7dff36be 100644 --- a/src/gallium/drivers/r300/r300_emit.c +++ b/src/gallium/drivers/r300/r300_emit.c @@ -129,7 +129,9 @@ static const float * get_shader_constant( struct rc_constant * constant, struct r300_constant_buffer * externals) { - static const float zero[4] = { 0.0, 0.0, 0.0, 0.0 }; + static float vec[4] = { 0.0, 0.0, 0.0, 1.0 }; + struct pipe_texture *tex; + switch(constant->Type) { case RC_CONSTANT_EXTERNAL: return externals->constants[constant->u.External]; @@ -137,11 +139,31 @@ static const float * get_shader_constant( case RC_CONSTANT_IMMEDIATE: return constant->u.Immediate; + case RC_CONSTANT_STATE: + switch (constant->u.State[0]) { + /* Factor for converting rectangle coords to + * normalized coords. Should only show up on non-r500. */ + case RC_STATE_R300_TEXRECT_FACTOR: + tex = &r300->textures[constant->u.State[1]]->tex; + vec[0] = 1.0 / tex->width[0]; + vec[1] = 1.0 / tex->height[0]; + break; + + default: + debug_printf("r300: Implementation error: " + "Unknown RC_CONSTANT type %d\n", constant->u.State[0]); + } + break; + default: - debug_printf("r300: Implementation error: Unhandled constant type %i\n", - constant->Type); - return zero; + debug_printf("r300: Implementation error: " + "Unhandled constant type %d\n", constant->Type); } + + /* This should either be (0, 0, 0, 1), which should be a relatively safe + * RGBA or STRQ value, or it could be one of the RC_CONSTANT_STATE + * state factors. */ + return vec; } /* Convert a normal single-precision float into the 7.16 format @@ -561,6 +583,8 @@ void r300_emit_texture(struct r300_context* r300, unsigned offset) { uint32_t filter0 = sampler->filter0; + uint32_t format0 = tex->state.format0; + unsigned min_level, max_level; CS_LOCALS(r300); /* to emulate 1D textures through 2D ones correctly */ @@ -569,13 +593,20 @@ void r300_emit_texture(struct r300_context* r300, filter0 |= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE); } + /* determine min/max levels */ + /* the MAX_MIP level is the largest (finest) one */ + max_level = MIN2(sampler->max_lod, tex->tex.last_level); + min_level = MIN2(sampler->min_lod, max_level); + format0 |= R300_TX_NUM_LEVELS(max_level); + filter0 |= R300_TX_MAX_MIP_LEVEL(min_level); + BEGIN_CS(16); OUT_CS_REG(R300_TX_FILTER0_0 + (offset * 4), filter0 | (offset << 28)); OUT_CS_REG(R300_TX_FILTER1_0 + (offset * 4), sampler->filter1); OUT_CS_REG(R300_TX_BORDER_COLOR_0 + (offset * 4), sampler->border_color); - OUT_CS_REG(R300_TX_FORMAT0_0 + (offset * 4), tex->state.format0); + OUT_CS_REG(R300_TX_FORMAT0_0 + (offset * 4), format0); OUT_CS_REG(R300_TX_FORMAT1_0 + (offset * 4), tex->state.format1); OUT_CS_REG(R300_TX_FORMAT2_0 + (offset * 4), tex->state.format2); OUT_CS_REG_SEQ(R300_TX_OFFSET_0 + (offset * 4), 1); @@ -690,12 +721,37 @@ void r300_emit_vertex_format_state(struct r300_context* r300) END_CS; } +/* XXX This should go to util ... */ +/* Return the number of bits set in the given number. */ +static unsigned bitcount(unsigned n) +{ + unsigned bits = 0; + + while (n) { + if (n & 1) { + bits++; + } + n >>= 1; + } + + return bits; +} + void r300_emit_vertex_program_code(struct r300_context* r300, struct r300_vertex_program_code* code) { int i; struct r300_screen* r300screen = r300_screen(r300->context.screen); unsigned instruction_count = code->length / 4; + + int vtx_mem_size = r300screen->caps->is_r500 ? 128 : 72; + int input_count = MAX2(bitcount(code->InputsRead), 1); + int output_count = MAX2(bitcount(code->OutputsWritten), 1); + int temp_count = MAX2(code->num_temporaries, 1); + int pvs_num_slots = MIN3(vtx_mem_size / input_count, + vtx_mem_size / output_count, 10); + int pvs_num_controllers = MIN2(vtx_mem_size / temp_count, 6); + CS_LOCALS(r300); if (!r300screen->caps->has_tcl) { @@ -708,8 +764,7 @@ void r300_emit_vertex_program_code(struct r300_context* r300, /* R300_VAP_PVS_CODE_CNTL_0 * R300_VAP_PVS_CONST_CNTL * R300_VAP_PVS_CODE_CNTL_1 - * See the r5xx docs for instructions on how to use these. - * XXX these could be optimized to select better values... */ + * See the r5xx docs for instructions on how to use these. */ OUT_CS_REG_SEQ(R300_VAP_PVS_CODE_CNTL_0, 3); OUT_CS(R300_PVS_FIRST_INST(0) | R300_PVS_XYZW_VALID_INST(instruction_count - 1) | @@ -722,10 +777,11 @@ void r300_emit_vertex_program_code(struct r300_context* r300, for (i = 0; i < code->length; i++) OUT_CS(code->body.d[i]); - OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(10) | - R300_PVS_NUM_CNTLRS(5) | + OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(pvs_num_slots) | + R300_PVS_NUM_CNTLRS(pvs_num_controllers) | R300_PVS_NUM_FPUS(r300screen->caps->num_vert_fpus) | - R300_PVS_VF_MAX_VTX_NUM(12)); + R300_PVS_VF_MAX_VTX_NUM(12) | + (r300screen->caps->is_r500 ? R500_TCL_STATE_OPTIMIZATION : 0)); END_CS; } @@ -790,13 +846,22 @@ void r300_emit_viewport_state(struct r300_context* r300, END_CS; } +void r300_emit_texture_count(struct r300_context* r300) +{ + CS_LOCALS(r300); + + BEGIN_CS(2); + OUT_CS_REG(R300_TX_ENABLE, (1 << r300->texture_count) - 1); + END_CS; + +} + void r300_flush_textures(struct r300_context* r300) { CS_LOCALS(r300); - BEGIN_CS(4); + BEGIN_CS(2); OUT_CS_REG(R300_TX_INVALTAGS, 0); - OUT_CS_REG(R300_TX_ENABLE, (1 << r300->texture_count) - 1); END_CS; } @@ -950,6 +1015,8 @@ validate: /* Samplers and textures are tracked separately but emitted together. */ if (r300->dirty_state & (R300_ANY_NEW_SAMPLERS | R300_ANY_NEW_TEXTURES)) { + r300_emit_texture_count(r300); + for (i = 0; i < MIN2(r300->sampler_count, r300->texture_count); i++) { if (r300->dirty_state & ((R300_NEW_SAMPLER << i) | (R300_NEW_TEXTURE << i))) { diff --git a/src/gallium/drivers/r300/r300_emit.h b/src/gallium/drivers/r300/r300_emit.h index 7c83c5166d..3797d3d332 100644 --- a/src/gallium/drivers/r300/r300_emit.h +++ b/src/gallium/drivers/r300/r300_emit.h @@ -92,6 +92,8 @@ void r300_emit_vertex_shader(struct r300_context* r300, void r300_emit_viewport_state(struct r300_context* r300, struct r300_viewport_state* viewport); +void r300_emit_texture_count(struct r300_context* r300); + void r300_flush_textures(struct r300_context* r300); /* Emit all dirty state. */ diff --git a/src/gallium/drivers/r300/r300_reg.h b/src/gallium/drivers/r300/r300_reg.h index 8ca785cb58..3a419b24b0 100644 --- a/src/gallium/drivers/r300/r300_reg.h +++ b/src/gallium/drivers/r300/r300_reg.h @@ -1293,7 +1293,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R500_RS_INST_TEX_ID(x) ((x) << 0) #define R500_RS_INST_TEX_CN_WRITE (1 << 4) #define R500_RS_INST_TEX_ADDR_SHIFT 5 -# define R500_RS_INST_TEX_ADDR(x) ((x) << 0) +# define R500_RS_INST_TEX_ADDR(x) ((x) << 5) #define R500_RS_INST_COL_ID_SHIFT 12 # define R500_RS_INST_COL_ID(x) ((x) << 12) #define R500_RS_INST_COL_CN_NO_WRITE (0 << 16) @@ -1463,6 +1463,8 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R300_TX_MIN_FILTER_MIP_NEAREST (1 << 13) # define R300_TX_MIN_FILTER_MIP_LINEAR (2 << 13) # define R300_TX_MIN_FILTER_MIP_MASK (3 << 13) +# define R300_TX_MAX_MIP_LEVEL_SHIFT 17 +# define R300_TX_MAX_MIP_LEVEL_MASK (0xf << 17) # define R300_TX_MAX_ANISO_1_TO_1 (0 << 21) # define R300_TX_MAX_ANISO_2_TO_1 (1 << 21) # define R300_TX_MAX_ANISO_4_TO_1 (2 << 21) @@ -1471,6 +1473,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R300_TX_MAX_ANISO_MASK (7 << 21) # define R300_TX_WRAP_S(x) ((x) << 0) # define R300_TX_WRAP_T(x) ((x) << 3) +# define R300_TX_MAX_MIP_LEVEL(x) ((x) << 17) #define R300_TX_FILTER1_0 0x4440 # define R300_CHROMA_KEY_MODE_DISABLE 0 @@ -1500,8 +1503,6 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R300_TX_HEIGHTMASK_MASK (2047 << 11) # define R300_TX_DEPTHMASK_SHIFT 22 # define R300_TX_DEPTHMASK_MASK (0xf << 22) -# define R300_TX_MAX_MIP_LEVEL_SHIFT 26 -# define R300_TX_MAX_MIP_LEVEL_MASK (0xf << 26) # define R300_TX_SIZE_PROJECTED (1 << 30) # define R300_TX_PITCH_EN (1 << 31) # define R300_TX_WIDTH(x) ((x) << 0) diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c index 62e1456ed3..4c5fb405c6 100644 --- a/src/gallium/drivers/r300/r300_render.c +++ b/src/gallium/drivers/r300/r300_render.c @@ -183,8 +183,6 @@ boolean r300_draw_range_elements(struct pipe_context* pipe, return FALSE; } - setup_vertex_attributes(r300); - setup_index_buffer(r300, indexBuffer, indexSize); r300_emit_dirty_state(r300); @@ -226,8 +224,6 @@ boolean r300_draw_arrays(struct pipe_context* pipe, unsigned mode, return FALSE; } - setup_vertex_attributes(r300); - r300_emit_dirty_state(r300); r300_emit_aos(r300, start); diff --git a/src/gallium/drivers/r300/r300_screen.h b/src/gallium/drivers/r300/r300_screen.h index 41df31f670..1ce5ff3904 100644 --- a/src/gallium/drivers/r300/r300_screen.h +++ b/src/gallium/drivers/r300/r300_screen.h @@ -27,6 +27,8 @@ #include "r300_chipset.h" +struct r300_winsys; + struct r300_screen { /* Parent class */ struct pipe_screen screen; diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c index d1eced61db..a88d66db24 100644 --- a/src/gallium/drivers/r300/r300_state.c +++ b/src/gallium/drivers/r300/r300_state.c @@ -523,6 +523,11 @@ static void* state->mag_img_filter, state->min_mip_filter); + /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */ + /* We must pass these to the emit function to clamp them properly. */ + sampler->min_lod = MAX2((unsigned)state->min_lod, 0); + sampler->max_lod = MAX2((unsigned)ceilf(state->max_lod), 0); + lod_bias = CLAMP((int)(state->lod_bias * 32), -(1 << 9), (1 << 9) - 1); sampler->filter1 |= lod_bias << R300_LOD_BIAS_SHIFT; @@ -571,6 +576,7 @@ static void r300_set_sampler_textures(struct pipe_context* pipe, struct pipe_texture** texture) { struct r300_context* r300 = r300_context(pipe); + boolean is_r500 = r300_screen(r300->context.screen)->caps->is_r500; int i; /* XXX magic num */ @@ -585,6 +591,13 @@ static void r300_set_sampler_textures(struct pipe_context* pipe, pipe_texture_reference((struct pipe_texture**)&r300->textures[i], texture[i]); r300->dirty_state |= (R300_NEW_TEXTURE << i); + + /* R300-specific - set the texrect factor in a fragment shader */ + if (!is_r500 && r300->textures[i]->is_npot) { + /* XXX It would be nice to re-emit just 1 constant, + * XXX not all of them */ + r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; + } } } @@ -674,6 +687,8 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe, draw_flush(r300->draw); draw_set_vertex_buffers(r300->draw, count, buffers); } + + r300->dirty_state |= R300_NEW_VERTEX_FORMAT; } static void r300_set_vertex_elements(struct pipe_context* pipe, @@ -706,9 +721,6 @@ static void* r300_create_vs_state(struct pipe_context* pipe, tgsi_scan_shader(shader->tokens, &vs->info); - /* Appease Draw. */ - vs->draw = draw_create_vertex_shader(r300->draw, shader); - return (void*)vs; } else { return draw_create_vertex_shader(r300->draw, shader); @@ -719,8 +731,6 @@ static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) { struct r300_context* r300 = r300_context(pipe); - draw_flush(r300->draw); - if (r300_screen(pipe->screen)->caps->has_tcl) { struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; @@ -731,10 +741,10 @@ static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) r300_translate_vertex_shader(r300, vs); } - draw_bind_vertex_shader(r300->draw, vs->draw); r300->vs = vs; r300->dirty_state |= R300_NEW_VERTEX_SHADER | R300_NEW_VERTEX_SHADER_CONSTANTS; } else { + draw_flush(r300->draw); draw_bind_vertex_shader(r300->draw, (struct draw_vertex_shader*)shader); } @@ -748,7 +758,6 @@ static void r300_delete_vs_state(struct pipe_context* pipe, void* shader) struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; rc_constants_destroy(&vs->code.constants); - draw_delete_vertex_shader(r300->draw, vs->draw); FREE((void*)vs->state.tokens); FREE(shader); } else { diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c index 7166694edf..6fb780cb29 100644 --- a/src/gallium/drivers/r300/r300_state_derived.c +++ b/src/gallium/drivers/r300/r300_state_derived.c @@ -47,8 +47,8 @@ struct r300_shader_derived_value { unsigned r300_shader_key_hash(void* key) { struct r300_shader_key* shader_key = (struct r300_shader_key*)key; - unsigned vs = (unsigned)shader_key->vs; - unsigned fs = (unsigned)shader_key->fs; + unsigned vs = (intptr_t)shader_key->vs; + unsigned fs = (intptr_t)shader_key->fs; return (vs << 16) | (fs & 0xffff); } @@ -61,88 +61,46 @@ int r300_shader_key_compare(void* key1, void* key2) { (shader_key1->fs == shader_key2->fs); } -/* Set up the vs_tab and routes. */ -static void r300_vs_tab_routes(struct r300_context* r300, - struct r300_vertex_info* vformat) +/* Set up the vs_output_tab and routes. */ +static void r300_vs_output_tab_routes(struct r300_context* r300, + int* vs_output_tab) { - struct r300_screen* r300screen = r300_screen(r300->context.screen); - struct vertex_info* vinfo = &vformat->vinfo; - int* tab = vformat->vs_tab; + struct vertex_info* vinfo = &r300->vertex_info->vinfo; boolean pos = FALSE, psize = FALSE, fog = FALSE; int i, texs = 0, cols = 0; - struct tgsi_shader_info* info; - - if (r300screen->caps->has_tcl) { - /* Use vertex shader to determine required routes. */ - info = &r300->vs->info; - } else { - /* Use fragment shader to determine required routes. */ - info = &r300->fs->info; - } + struct tgsi_shader_info* info = &r300->fs->info; - assert(info->num_inputs <= 16); + /* XXX One day we should figure out how to handle a different number of + * VS outputs and FS inputs, as well as a different number of vertex streams + * and VS inputs. It's definitely one of the sources of hardlocks. */ - if (!r300screen->caps->has_tcl || !r300->rs_state->enable_vte) - { - for (i = 0; i < info->num_inputs; i++) { - switch (r300->vs->code.inputs[i]) { - case TGSI_SEMANTIC_POSITION: - pos = TRUE; - tab[i] = 0; - break; - case TGSI_SEMANTIC_COLOR: - tab[i] = 2 + cols; - cols++; - break; - case TGSI_SEMANTIC_PSIZE: - assert(psize == FALSE); - psize = TRUE; - tab[i] = 15; - break; - case TGSI_SEMANTIC_FOG: - assert(fog == FALSE); - fog = TRUE; - /* Fall through */ - case TGSI_SEMANTIC_GENERIC: - tab[i] = 6 + texs; - texs++; - break; - default: - debug_printf("r300: Unknown vertex input %d\n", - info->input_semantic_name[i]); - break; - } - } - } - else - { - /* Just copy vert attribs over as-is. */ - for (i = 0; i < info->num_inputs; i++) { - tab[i] = i; - } - - for (i = 0; i < info->num_outputs; i++) { - switch (info->output_semantic_name[i]) { - case TGSI_SEMANTIC_POSITION: - pos = TRUE; - break; - case TGSI_SEMANTIC_COLOR: - cols++; - break; - case TGSI_SEMANTIC_PSIZE: - psize = TRUE; - break; - case TGSI_SEMANTIC_FOG: - fog = TRUE; - /* Fall through */ - case TGSI_SEMANTIC_GENERIC: - texs++; - break; - default: - debug_printf("r300: Unknown vertex output %d\n", - info->output_semantic_name[i]); - break; - } + for (i = 0; i < info->num_inputs; i++) { + switch (info->input_semantic_name[i]) { + case TGSI_SEMANTIC_POSITION: + pos = TRUE; + vs_output_tab[i] = 0; + break; + case TGSI_SEMANTIC_COLOR: + vs_output_tab[i] = 2 + cols; + cols++; + break; + case TGSI_SEMANTIC_PSIZE: + assert(psize == FALSE); + psize = TRUE; + vs_output_tab[i] = 15; + break; + case TGSI_SEMANTIC_FOG: + assert(fog == FALSE); + fog = TRUE; + /* Fall through */ + case TGSI_SEMANTIC_GENERIC: + vs_output_tab[i] = 6 + texs; + texs++; + break; + default: + debug_printf("r300: Unknown vertex input %d\n", + info->input_semantic_name[i]); + break; } } @@ -161,14 +119,13 @@ static void r300_vs_tab_routes(struct r300_context* r300, /* We need to add vertex position attribute only for SW TCL case, * for HW TCL case it could be generated by vertex shader */ - if (!pos && !r300screen->caps->has_tcl) { - debug_printf("r300: Forcing vertex position attribute emit...\n"); + if (!pos) { /* Make room for the position attribute - * at the beginning of the tab. */ + * at the beginning of the vs_output_tab. */ for (i = 15; i > 0; i--) { - tab[i] = tab[i-1]; + vs_output_tab[i] = vs_output_tab[i-1]; } - tab[0] = 0; + vs_output_tab[0] = 0; } /* Position. */ @@ -227,43 +184,78 @@ static void r300_vs_tab_routes(struct r300_context* r300, } /* Update the PSC tables. */ -static void r300_vertex_psc(struct r300_context* r300, - struct r300_vertex_info* vformat) +static void r300_vertex_psc(struct r300_context* r300) { - struct r300_screen* r300screen = r300_screen(r300->context.screen); - struct vertex_info* vinfo = &vformat->vinfo; - int* tab = vformat->vs_tab; + struct r300_vertex_info *vformat = r300->vertex_info; uint16_t type, swizzle; enum pipe_format format; - unsigned i, attrib_count; + unsigned i; /* Vertex shaders have no semantics on their inputs, - * so PSC should just route stuff based on their info, + * so PSC should just route stuff based on the vertex elements, * and not on attrib information. */ - if (r300screen->caps->has_tcl) { - attrib_count = r300->vs->info.num_inputs; - DBG(r300, DBG_DRAW, "r300: routing %d attribs in psc for vs\n", - attrib_count); - } else { - attrib_count = vinfo->num_attribs; - DBG(r300, DBG_DRAW, "r300: attrib count: %d\n", attrib_count); - for (i = 0; i < attrib_count; i++) { - DBG(r300, DBG_DRAW, "r300: attrib: offset %d, interp %d, size %d," - " tab %d\n", vinfo->attrib[i].src_index, - vinfo->attrib[i].interp_mode, vinfo->attrib[i].emit, - tab[i]); + DBG(r300, DBG_DRAW, "r300: vs expects %d attribs, routing %d elements" + " in psc\n", + r300->vs->info.num_inputs, + r300->vertex_element_count); + + for (i = 0; i < r300->vertex_element_count; i++) { + format = r300->vertex_element[i].src_format; + + type = r300_translate_vertex_data_type(format) | + (i << R300_DST_VEC_LOC_SHIFT); + swizzle = r300_translate_vertex_data_swizzle(format); + + if (i % 2) { + vformat->vap_prog_stream_cntl[i >> 1] |= type << 16; + vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16; + } else { + vformat->vap_prog_stream_cntl[i >> 1] |= type; + vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle; } } + + assert(i <= 15); + + /* Set the last vector in the PSC. */ + if (i) { + i -= 1; + } + vformat->vap_prog_stream_cntl[i >> 1] |= + (R300_LAST_VEC << (i & 1 ? 16 : 0)); +} + +/* Update the PSC tables for SW TCL, using Draw. */ +static void r300_swtcl_vertex_psc(struct r300_context* r300, + int* vs_output_tab) +{ + struct r300_vertex_info *vformat = r300->vertex_info; + struct vertex_info* vinfo = &vformat->vinfo; + uint16_t type, swizzle; + enum pipe_format format; + unsigned i, attrib_count; + + /* For each Draw attribute, route it to the fragment shader according + * to the vs_output_tab. */ + attrib_count = vinfo->num_attribs; + DBG(r300, DBG_DRAW, "r300: attrib count: %d\n", attrib_count); + for (i = 0; i < attrib_count; i++) { + DBG(r300, DBG_DRAW, "r300: attrib: offset %d, interp %d, size %d," + " vs_output_tab %d\n", vinfo->attrib[i].src_index, + vinfo->attrib[i].interp_mode, vinfo->attrib[i].emit, + vs_output_tab[i]); + } + for (i = 0; i < attrib_count; i++) { /* Make sure we have a proper destination for our attribute. */ - assert(tab[i] != -1); + assert(vs_output_tab[i] != -1); format = draw_translate_vinfo_format(vinfo->attrib[i].emit); /* Obtain the type of data in this attribute. */ type = r300_translate_vertex_data_type(format) | - tab[i] << R300_DST_VEC_LOC_SHIFT; + vs_output_tab[i] << R300_DST_VEC_LOC_SHIFT; /* Obtain the swizzle for this attribute. Note that the default * swizzle in the hardware is not XYZW! */ @@ -272,12 +264,10 @@ static void r300_vertex_psc(struct r300_context* r300, /* Add the attribute to the PSC table. */ if (i & 1) { vformat->vap_prog_stream_cntl[i >> 1] |= type << 16; - vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16; } else { - vformat->vap_prog_stream_cntl[i >> 1] |= type << 0; - - vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 0; + vformat->vap_prog_stream_cntl[i >> 1] |= type; + vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle; } } @@ -289,74 +279,12 @@ static void r300_vertex_psc(struct r300_context* r300, (R300_LAST_VEC << (i & 1 ? 16 : 0)); } -/* Set up the mappings from GB to US, for RS block. */ -static void r300_update_fs_tab(struct r300_context* r300, - struct r300_vertex_info* vformat) -{ - struct tgsi_shader_info* info = &r300->fs->info; - int i, cols = 0, texs = 0, cols_emitted = 0; - int* tab = vformat->fs_tab; - - for (i = 0; i < 16; i++) { - tab[i] = -1; - } - - assert(info->num_inputs <= 16); - for (i = 0; i < info->num_inputs; i++) { - switch (info->input_semantic_name[i]) { - case TGSI_SEMANTIC_COLOR: - tab[i] = INTERP_LINEAR; - cols++; - break; - case TGSI_SEMANTIC_POSITION: - case TGSI_SEMANTIC_PSIZE: - debug_printf("r300: Implementation error: Can't use " - "pos attribs in fragshader yet!\n"); - /* Pass through for now */ - case TGSI_SEMANTIC_FOG: - case TGSI_SEMANTIC_GENERIC: - tab[i] = INTERP_PERSPECTIVE; - break; - default: - debug_printf("r300: Unknown vertex input %d\n", - info->input_semantic_name[i]); - break; - } - } - - /* Now that we know where everything is... */ - DBG(r300, DBG_DRAW, "r300: fp input count: %d\n", info->num_inputs); - for (i = 0; i < info->num_inputs; i++) { - switch (tab[i]) { - case INTERP_LINEAR: - DBG(r300, DBG_DRAW, "r300: attrib: " - "stack offset %d, color, tab %d\n", - i, cols_emitted); - tab[i] = cols_emitted; - cols_emitted++; - break; - case INTERP_PERSPECTIVE: - DBG(r300, DBG_DRAW, "r300: attrib: " - "stack offset %d, texcoord, tab %d\n", - i, cols + texs); - tab[i] = cols + texs; - texs++; - break; - case -1: - debug_printf("r300: Implementation error: Bad fp interp!\n"); - default: - break; - } - } - -} - /* Set up the RS block. This is the part of the chipset that actually does * the rasterization of vertices into fragments. This is also the part of the * chipset that locks up if any part of it is even slightly wrong. */ -static void r300_update_rs_block(struct r300_context* r300, - struct r300_rs_block* rs) +static void r300_update_rs_block(struct r300_context* r300) { + struct r300_rs_block* rs = r300->rs_block; struct tgsi_shader_info* info = &r300->fs->info; int col_count = 0, fp_offset = 0, i, tex_count = 0; int rs_tex_comp = 0; @@ -390,17 +318,18 @@ static void r300_update_rs_block(struct r300_context* r300, col_count++; } + for (i = 0; i < col_count; i++) { + rs->inst[i] |= R500_RS_INST_COL_ID(i) | + R500_RS_INST_COL_CN_WRITE | R500_RS_INST_COL_ADDR(fp_offset); + fp_offset++; + } + for (i = 0; i < tex_count; i++) { rs->inst[i] |= R500_RS_INST_TEX_ID(i) | R500_RS_INST_TEX_CN_WRITE | R500_RS_INST_TEX_ADDR(fp_offset); fp_offset++; } - for (i = 0; i < col_count; i++) { - rs->inst[i] |= R500_RS_INST_COL_ID(i) | - R500_RS_INST_COL_CN_WRITE | R500_RS_INST_COL_ADDR(fp_offset); - fp_offset++; - } } else { for (i = 0; i < info->num_inputs; i++) { switch (info->input_semantic_name[i]) { @@ -425,8 +354,10 @@ static void r300_update_rs_block(struct r300_context* r300, } } + /* Rasterize at least one color, or bad things happen. */ if (col_count == 0) { rs->ip[0] |= R300_RS_COL_FMT(R300_RS_COL_FMT_0001); + col_count++; } if (tex_count == 0) { @@ -437,9 +368,10 @@ static void r300_update_rs_block(struct r300_context* r300, R300_RS_SEL_Q(R300_RS_SEL_K1); } - /* Rasterize at least one color, or bad things happen. */ - if ((col_count == 0) && (tex_count == 0)) { - col_count++; + for (i = 0; i < col_count; i++) { + rs->inst[i] |= R300_RS_INST_COL_ID(i) | + R300_RS_INST_COL_CN_WRITE | R300_RS_INST_COL_ADDR(fp_offset); + fp_offset++; } for (i = 0; i < tex_count; i++) { @@ -447,28 +379,22 @@ static void r300_update_rs_block(struct r300_context* r300, R300_RS_INST_TEX_CN_WRITE | R300_RS_INST_TEX_ADDR(fp_offset); fp_offset++; } - - for (i = 0; i < col_count; i++) { - rs->inst[i] |= R300_RS_INST_COL_ID(i) | - R300_RS_INST_COL_CN_WRITE | R300_RS_INST_COL_ADDR(fp_offset); - fp_offset++; - } } rs->count = (rs_tex_comp) | (col_count << R300_IC_COUNT_SHIFT) | R300_HIRES_EN; - rs->inst_count = MAX2(MAX2(col_count - 1, tex_count - 1), 0); + rs->inst_count = MAX3(col_count - 1, tex_count - 1, 0); } /* Update the vertex format. */ static void r300_update_derived_shader_state(struct r300_context* r300) { struct r300_screen* r300screen = r300_screen(r300->context.screen); - struct r300_vertex_info* vformat; - struct r300_rs_block* rs_block; + int vs_output_tab[16]; int i; + /* struct r300_shader_key* key; struct r300_shader_derived_value* value; @@ -495,27 +421,26 @@ static void r300_update_derived_shader_state(struct r300_context* r300) (void*)key, (void*)value); } */ - /* XXX This will be refactored ASAP. */ - vformat = CALLOC_STRUCT(r300_vertex_info); - rs_block = CALLOC_STRUCT(r300_rs_block); + /* Reset structures */ + memset(r300->rs_block, 0, sizeof(struct r300_rs_block)); + memset(r300->vertex_info, 0, sizeof(struct r300_vertex_info)); for (i = 0; i < 16; i++) { - vformat->vs_tab[i] = -1; - vformat->fs_tab[i] = -1; + vs_output_tab[i] = -1; } - r300_vs_tab_routes(r300, vformat); - r300_vertex_psc(r300, vformat); - r300_update_fs_tab(r300, vformat); + /* Update states */ + r300_vs_output_tab_routes(r300, vs_output_tab); - r300_update_rs_block(r300, rs_block); + if (r300screen->caps->has_tcl) { + r300_vertex_psc(r300); + } else { + r300_swtcl_vertex_psc(r300, vs_output_tab); + } - FREE(r300->vertex_info); - FREE(r300->rs_block); + r300_update_rs_block(r300); - r300->vertex_info = vformat; - r300->rs_block = rs_block; - r300->dirty_state |= (R300_NEW_VERTEX_FORMAT | R300_NEW_RS_BLOCK); + r300->dirty_state |= R300_NEW_RS_BLOCK; } static void r300_update_ztop(struct r300_context* r300) @@ -553,9 +478,9 @@ static void r300_update_ztop(struct r300_context* r300) void r300_update_derived_state(struct r300_context* r300) { - /* XXX */ - if (TRUE || r300->dirty_state & - (R300_NEW_FRAGMENT_SHADER | R300_NEW_VERTEX_SHADER)) { + if (r300->dirty_state & + (R300_NEW_FRAGMENT_SHADER | R300_NEW_VERTEX_SHADER | + R300_NEW_VERTEX_FORMAT)) { r300_update_derived_shader_state(r300); } diff --git a/src/gallium/drivers/r300/r300_state_invariant.c b/src/gallium/drivers/r300/r300_state_invariant.c index c07e6ae676..46d1cb39b5 100644 --- a/src/gallium/drivers/r300/r300_state_invariant.c +++ b/src/gallium/drivers/r300/r300_state_invariant.c @@ -84,7 +84,7 @@ void r300_emit_invariant_state(struct r300_context* r300) END_CS; /* XXX unsorted stuff from surface_fill */ - BEGIN_CS(60 + (caps->has_tcl ? 5 : 0) + (caps->is_r500 ? 4 : 0)); + BEGIN_CS(56 + (caps->has_tcl ? 5 : 0) + (caps->is_r500 ? 4 : 0)); /* Flush PVS. */ OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0x0); @@ -135,8 +135,6 @@ void r300_emit_invariant_state(struct r300_context* r300) OUT_CS_REG(R300_ZB_DEPTHCLEARVALUE, 0x00000000); OUT_CS_REG(R300_ZB_HIZ_OFFSET, 0x00000000); OUT_CS_REG(R300_ZB_HIZ_PITCH, 0x00000000); - OUT_CS_REG(R300_VAP_VTX_STATE_CNTL, 0x1); - OUT_CS_REG(R300_VAP_VSM_VTX_ASSM, 0x405); OUT_CS_REG(R300_SE_VTE_CNTL, 0x0000043F); /* XXX */ diff --git a/src/gallium/drivers/r300/r300_texture.c b/src/gallium/drivers/r300/r300_texture.c index aea25cf71d..d13aa8f036 100644 --- a/src/gallium/drivers/r300/r300_texture.c +++ b/src/gallium/drivers/r300/r300_texture.c @@ -43,8 +43,7 @@ static void r300_setup_texture_state(struct r300_texture* tex, boolean is_r500) state->format2 = (tex->pitch[0] - 1) & 0x1fff; } else { /* power of two textures (3D, mipmaps, and no pitch) */ - state->format0 |= R300_TX_DEPTH(util_logbase2(pt->depth[0]) & 0xf) | - R300_TX_NUM_LEVELS(pt->last_level & 0xf); + state->format0 |= R300_TX_DEPTH(util_logbase2(pt->depth[0]) & 0xf); } state->format1 = r300_translate_texformat(pt->format); diff --git a/src/gallium/drivers/r300/r300_vbo.c b/src/gallium/drivers/r300/r300_vbo.c index a6a159667a..6ebaf715dc 100644 --- a/src/gallium/drivers/r300/r300_vbo.c +++ b/src/gallium/drivers/r300/r300_vbo.c @@ -34,52 +34,6 @@ #include "r300_reg.h" #include "r300_winsys.h" -static INLINE void setup_vertex_attribute(struct r300_vertex_info *vinfo, - struct pipe_vertex_element *vert_elem, - unsigned attr_num) -{ - uint16_t hw_fmt1, hw_fmt2; - - hw_fmt1 = r300_translate_vertex_data_type(vert_elem->src_format) | - (attr_num << R300_DST_VEC_LOC_SHIFT); - hw_fmt2 = r300_translate_vertex_data_swizzle(vert_elem->src_format); - - if (attr_num % 2 == 0) - { - vinfo->vap_prog_stream_cntl[attr_num >> 1] = hw_fmt1; - vinfo->vap_prog_stream_cntl_ext[attr_num >> 1] = hw_fmt2; - } - else - { - vinfo->vap_prog_stream_cntl[attr_num >> 1] |= hw_fmt1 << 16; - vinfo->vap_prog_stream_cntl_ext[attr_num >> 1] |= hw_fmt2 << 16; - } -} - -static void finish_vertex_attribs_setup(struct r300_vertex_info *vinfo, - unsigned attribs_num) -{ - uint32_t last_vec_bit = (attribs_num % 2 == 0) ? - (R300_LAST_VEC << 16) : R300_LAST_VEC; - - assert(attribs_num > 0 && attribs_num <= 16); - vinfo->vap_prog_stream_cntl[(attribs_num - 1) >> 1] |= last_vec_bit; -} - -void setup_vertex_attributes(struct r300_context *r300) -{ - struct pipe_vertex_element *vert_elem; - int i; - - for (i = 0; i < r300->vertex_element_count; i++) { - vert_elem = &r300->vertex_element[i]; - setup_vertex_attribute(r300->vertex_info, vert_elem, i); - } - - finish_vertex_attribs_setup(r300->vertex_info, - r300->vertex_element_count); -} - static INLINE int get_buffer_offset(struct r300_context *r300, unsigned int buf_nr, unsigned int elem_offset) diff --git a/src/gallium/drivers/r300/r300_vs.h b/src/gallium/drivers/r300/r300_vs.h index 2a4ce315e3..00b02bf510 100644 --- a/src/gallium/drivers/r300/r300_vs.h +++ b/src/gallium/drivers/r300/r300_vs.h @@ -35,9 +35,6 @@ struct r300_vertex_shader { struct pipe_shader_state state; struct tgsi_shader_info info; - /* Fallback shader, because Draw has issues */ - struct draw_vertex_shader* draw; - /* Has this shader been translated yet? */ boolean translated; diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c index 5f60139968..bdbb7fa9b9 100644 --- a/src/gallium/drivers/softpipe/sp_context.c +++ b/src/gallium/drivers/softpipe/sp_context.c @@ -90,14 +90,15 @@ softpipe_destroy( struct pipe_context *pipe ) if (softpipe->draw) draw_destroy( softpipe->draw ); - softpipe->quad.shade->destroy( softpipe->quad.shade ); - softpipe->quad.depth_test->destroy( softpipe->quad.depth_test ); - softpipe->quad.blend->destroy( softpipe->quad.blend ); + softpipe->quad.shade->destroy( softpipe->quad.shade ); + softpipe->quad.depth_test->destroy( softpipe->quad.depth_test ); + softpipe->quad.blend->destroy( softpipe->quad.blend ); for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) { sp_destroy_tile_cache(softpipe->cbuf_cache[i]); pipe_surface_reference(&softpipe->framebuffer.cbufs[i], NULL); } + sp_destroy_tile_cache(softpipe->zsbuf_cache); pipe_surface_reference(&softpipe->framebuffer.zsbuf, NULL); diff --git a/src/gallium/drivers/softpipe/sp_fs_exec.c b/src/gallium/drivers/softpipe/sp_fs_exec.c index 4076114d39..a8999ed347 100644 --- a/src/gallium/drivers/softpipe/sp_fs_exec.c +++ b/src/gallium/drivers/softpipe/sp_fs_exec.c @@ -126,7 +126,13 @@ exec_run( const struct sp_fragment_shader *base, setup_pos_vector(quad->posCoef, (float)quad->input.x0, (float)quad->input.y0, &machine->QuadPos); - + + if (quad->input.facing) { + machine->Face = -1.0f; + } else { + machine->Face = 1.0f; + } + quad->inout.mask &= tgsi_exec_machine_run( machine ); if (quad->inout.mask == 0) return FALSE; diff --git a/src/gallium/state_trackers/xorg/xorg_exa_tgsi.c b/src/gallium/state_trackers/xorg/xorg_exa_tgsi.c index 3bf64b6331..52b97af163 100644 --- a/src/gallium/state_trackers/xorg/xorg_exa_tgsi.c +++ b/src/gallium/state_trackers/xorg/xorg_exa_tgsi.c @@ -47,22 +47,22 @@ static void print_fs_traits(int fs_traits) { const char *strings[] = { - "FS_COMPOSITE", // = 1 << 0, - "FS_MASK", // = 1 << 1, - "FS_SOLID_FILL", // = 1 << 2, - "FS_LINGRAD_FILL", // = 1 << 3, - "FS_RADGRAD_FILL", // = 1 << 4, - "FS_CA_FULL", // = 1 << 5, /* src.rgba * mask.rgba */ - "FS_CA_SRCALPHA", // = 1 << 6, /* src.aaaa * mask.rgba */ - "FS_YUV", // = 1 << 7, - "FS_SRC_REPEAT_NONE", // = 1 << 8, - "FS_MASK_REPEAT_NONE",// = 1 << 9, - "FS_SRC_SWIZZLE_RGB", // = 1 << 10, - "FS_MASK_SWIZZLE_RGB",// = 1 << 11, - "FS_SRC_SET_ALPHA", // = 1 << 12, - "FS_MASK_SET_ALPHA", // = 1 << 13, - "FS_SRC_LUMINANCE", // = 1 << 14, - "FS_MASK_LUMINANCE", // = 1 << 15, + "FS_COMPOSITE", /* = 1 << 0 */ + "FS_MASK", /* = 1 << 1 */ + "FS_SOLID_FILL", /* = 1 << 2 */ + "FS_LINGRAD_FILL", /* = 1 << 3 */ + "FS_RADGRAD_FILL", /* = 1 << 4 */ + "FS_CA_FULL", /* = 1 << 5 - src.rgba * mask.rgba */ + "FS_CA_SRCALPHA", /* = 1 << 6 - src.aaaa * mask.rgba */ + "FS_YUV", /* = 1 << 7 */ + "FS_SRC_REPEAT_NONE", /* = 1 << 8 */ + "FS_MASK_REPEAT_NONE",/* = 1 << 9 */ + "FS_SRC_SWIZZLE_RGB", /* = 1 << 10 */ + "FS_MASK_SWIZZLE_RGB",/* = 1 << 11 */ + "FS_SRC_SET_ALPHA", /* = 1 << 12 */ + "FS_MASK_SET_ALPHA", /* = 1 << 13 */ + "FS_SRC_LUMINANCE", /* = 1 << 14 */ + "FS_MASK_LUMINANCE", /* = 1 << 15 */ }; int i, k; debug_printf("%s: ", __func__); diff --git a/src/gallium/state_trackers/xorg/xorg_xv.c b/src/gallium/state_trackers/xorg/xorg_xv.c index e7f4900528..4a701e93ec 100644 --- a/src/gallium/state_trackers/xorg/xorg_xv.c +++ b/src/gallium/state_trackers/xorg/xorg_xv.c @@ -73,10 +73,11 @@ static XF86VideoEncodingRec DummyEncoding[1] = { } }; -#define NUM_IMAGES 2 +#define NUM_IMAGES 3 static XF86ImageRec Images[NUM_IMAGES] = { XVIMAGE_UYVY, XVIMAGE_YUY2, + XVIMAGE_YV12, }; struct xorg_xv_port_priv { @@ -545,6 +546,7 @@ put_image(ScrnInfoPtr pScrn, switch (id) { case FOURCC_UYVY: case FOURCC_YUY2: + case FOURCC_YV12: default: srcPitch = width << 1; break; @@ -593,6 +595,7 @@ query_image_attributes(ScrnInfoPtr pScrn, switch (id) { case FOURCC_UYVY: case FOURCC_YUY2: + case FOURCC_YV12: default: size = *w << 1; if (pitches) |