diff options
Diffstat (limited to 'src/gallium/winsys/r600')
-rw-r--r-- | src/gallium/winsys/r600/Android.mk | 38 | ||||
-rw-r--r-- | src/gallium/winsys/r600/drm/Makefile | 5 | ||||
-rw-r--r-- | src/gallium/winsys/r600/drm/SConscript | 10 | ||||
-rw-r--r-- | src/gallium/winsys/r600/drm/bof.c | 2 | ||||
-rw-r--r-- | src/gallium/winsys/r600/drm/evergreen_hw_context.c | 76 | ||||
-rw-r--r-- | src/gallium/winsys/r600/drm/r600.c | 175 | ||||
-rw-r--r-- | src/gallium/winsys/r600/drm/r600_bo.c | 174 | ||||
-rw-r--r-- | src/gallium/winsys/r600/drm/r600_bomgr.c | 161 | ||||
-rw-r--r-- | src/gallium/winsys/r600/drm/r600_drm.c | 280 | ||||
-rw-r--r-- | src/gallium/winsys/r600/drm/r600_hw_context.c | 238 | ||||
-rw-r--r-- | src/gallium/winsys/r600/drm/r600_priv.h | 125 | ||||
-rw-r--r-- | src/gallium/winsys/r600/drm/r600d.h | 18 | ||||
-rw-r--r-- | src/gallium/winsys/r600/drm/radeon_bo.c | 31 | ||||
-rw-r--r-- | src/gallium/winsys/r600/drm/radeon_bo_pb.c | 260 | ||||
-rw-r--r-- | src/gallium/winsys/r600/drm/radeon_pciid.c | 114 |
15 files changed, 881 insertions, 826 deletions
diff --git a/src/gallium/winsys/r600/Android.mk b/src/gallium/winsys/r600/Android.mk new file mode 100644 index 0000000000..643a59dc42 --- /dev/null +++ b/src/gallium/winsys/r600/Android.mk @@ -0,0 +1,38 @@ +ifeq ($(strip $(MESA_BUILD_R600G)),true) + +LOCAL_PATH := $(call my-dir) + +# from drm/Makefile +C_SOURCES = \ + bof.c \ + evergreen_hw_context.c \ + radeon_bo.c \ + radeon_pciid.c \ + r600_bo.c \ + r600_drm.c \ + r600_hw_context.c \ + r600_bomgr.c + +include $(CLEAR_VARS) + +LOCAL_SRC_FILES := \ + $(addprefix drm/, $(C_SOURCES)) + +LOCAL_CFLAGS := \ + -std=c99 \ + -fvisibility=hidden \ + -Wno-sign-compare + +LOCAL_C_INCLUDES := \ + external/mesa/include \ + external/mesa/src/gallium/include \ + external/mesa/src/gallium/auxiliary \ + external/mesa/src/gallium/drivers/r600 \ + external/drm \ + external/drm/include/drm + +LOCAL_MODULE := libmesa_winsys_r600 + +include $(BUILD_STATIC_LIBRARY) + +endif # MESA_BUILD_R600G diff --git a/src/gallium/winsys/r600/drm/Makefile b/src/gallium/winsys/r600/drm/Makefile index a396205f89..7310734f05 100644 --- a/src/gallium/winsys/r600/drm/Makefile +++ b/src/gallium/winsys/r600/drm/Makefile @@ -8,12 +8,11 @@ C_SOURCES = \ bof.c \ evergreen_hw_context.c \ radeon_bo.c \ - radeon_bo_pb.c \ radeon_pciid.c \ - r600.c \ r600_bo.c \ r600_drm.c \ - r600_hw_context.c + r600_hw_context.c \ + r600_bomgr.c LIBRARY_INCLUDES = -I$(TOP)/src/gallium/drivers/r600 \ $(shell pkg-config libdrm --cflags-only-I) diff --git a/src/gallium/winsys/r600/drm/SConscript b/src/gallium/winsys/r600/drm/SConscript index cc053c06dd..cc9a06a239 100644 --- a/src/gallium/winsys/r600/drm/SConscript +++ b/src/gallium/winsys/r600/drm/SConscript @@ -6,15 +6,19 @@ r600_sources = [ 'bof.c', 'evergreen_hw_context.c', 'radeon_bo.c', - 'radeon_bo_pb.c', 'radeon_pciid.c', - 'r600.c', 'r600_bo.c', 'r600_drm.c', 'r600_hw_context.c', + 'r600_bomgr.c', ] -env.ParseConfig('pkg-config --cflags libdrm_radeon') +try: + env.ParseConfig('pkg-config --cflags libdrm_radeon') +except OSError: + print 'warning: not building r600g' + Return() + env.Append(CPPPATH = '#/src/gallium/drivers/r600') r600winsys = env.ConvenienceLibrary( diff --git a/src/gallium/winsys/r600/drm/bof.c b/src/gallium/winsys/r600/drm/bof.c index 0598cc6bc0..5c923ad38d 100644 --- a/src/gallium/winsys/r600/drm/bof.c +++ b/src/gallium/winsys/r600/drm/bof.c @@ -46,7 +46,7 @@ static int bof_entry_grow(bof_t *bof) } /* - * object + * object */ bof_t *bof_object(void) { diff --git a/src/gallium/winsys/r600/drm/evergreen_hw_context.c b/src/gallium/winsys/r600/drm/evergreen_hw_context.c index e1f163eab6..66398afa69 100644 --- a/src/gallium/winsys/r600/drm/evergreen_hw_context.c +++ b/src/gallium/winsys/r600/drm/evergreen_hw_context.c @@ -36,7 +36,6 @@ #include "pipe/p_compiler.h" #include "util/u_inlines.h" #include "util/u_memory.h" -#include <pipebuffer/pb_bufmgr.h> #include "r600_priv.h" #define GROUP_FORCE_NEW_BLOCK 0 @@ -622,10 +621,9 @@ int evergreen_context_init(struct r600_context *ctx, struct radeon *radeon) /* save 16dwords space for fence mecanism */ ctx->pm4_ndwords -= 16; - r = r600_context_init_fence(ctx); - if (r) { - goto out_err; - } + ctx->max_db = 8; + + LIST_INITHEAD(&ctx->fenced_bo); /* init dirty list */ LIST_INITHEAD(&ctx->dirty); @@ -836,25 +834,25 @@ void evergreen_context_draw(struct r600_context *ctx, const struct r600_draw *dr } /* draw packet */ - ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_INDEX_TYPE, 0); + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_INDEX_TYPE, 0, ctx->predicate_drawing); ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_index_type; - ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NUM_INSTANCES, 0); + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NUM_INSTANCES, 0, ctx->predicate_drawing); ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_num_instances; if (draw->indices) { - ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_DRAW_INDEX, 3); + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_DRAW_INDEX, 3, ctx->predicate_drawing); ctx->pm4[ctx->pm4_cdwords++] = draw->indices_bo_offset + r600_bo_offset(draw->indices); ctx->pm4[ctx->pm4_cdwords++] = 0; ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_num_indices; ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_draw_initiator; - ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0); + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, ctx->predicate_drawing); ctx->pm4[ctx->pm4_cdwords++] = 0; r600_context_bo_reloc(ctx, &ctx->pm4[ctx->pm4_cdwords - 1], draw->indices); } else { - ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_DRAW_INDEX_AUTO, 1); + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_DRAW_INDEX_AUTO, 1, ctx->predicate_drawing); ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_num_indices; ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_draw_initiator; } - ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0); + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0, ctx->predicate_drawing); ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_EVENT) | EVENT_INDEX(0); /* flush color buffer */ @@ -881,59 +879,3 @@ void evergreen_context_draw(struct r600_context *ctx, const struct r600_draw *dr ctx->pm4_dirty_cdwords = 0; } -static inline void evergreen_resource_set(struct r600_context *ctx, struct r600_pipe_state *state, unsigned offset) -{ - struct r600_range *range; - struct r600_block *block; - - range = &ctx->range[CTX_RANGE_ID(ctx, offset)]; - block = range->blocks[CTX_BLOCK_ID(ctx, offset)]; - block->reg[0] = state->regs[0].value; - block->reg[1] = state->regs[1].value; - block->reg[2] = state->regs[2].value; - block->reg[3] = state->regs[3].value; - block->reg[4] = state->regs[4].value; - block->reg[5] = state->regs[5].value; - block->reg[6] = state->regs[6].value; - block->reg[7] = state->regs[7].value; - r600_bo_reference(ctx->radeon, &block->reloc[1].bo, NULL); - r600_bo_reference(ctx->radeon , &block->reloc[2].bo, NULL); - if (state->regs[0].bo) { - /* VERTEX RESOURCE, we preted there is 2 bo to relocate so - * we have single case btw VERTEX & TEXTURE resource - */ - r600_bo_reference(ctx->radeon, &block->reloc[1].bo, state->regs[0].bo); - r600_bo_reference(ctx->radeon, &block->reloc[2].bo, state->regs[0].bo); - } else { - /* TEXTURE RESOURCE */ - r600_bo_reference(ctx->radeon, &block->reloc[1].bo, state->regs[2].bo); - r600_bo_reference(ctx->radeon, &block->reloc[2].bo, state->regs[3].bo); - } - if (!(block->status & R600_BLOCK_STATUS_DIRTY)) { - block->status |= R600_BLOCK_STATUS_ENABLED; - block->status |= R600_BLOCK_STATUS_DIRTY; - ctx->pm4_dirty_cdwords += block->pm4_ndwords + block->pm4_flush_ndwords; - LIST_ADDTAIL(&block->list,&ctx->dirty); - } -} - -void evergreen_ps_resource_set(struct r600_context *ctx, struct r600_pipe_state *state, unsigned rid) -{ - unsigned offset = R_030000_RESOURCE0_WORD0 + 0x20 * rid; - - evergreen_resource_set(ctx, state, offset); -} - -void evergreen_vs_resource_set(struct r600_context *ctx, struct r600_pipe_state *state, unsigned rid) -{ - unsigned offset = R_030000_RESOURCE0_WORD0 + 0x1600 + 0x20 * rid; - - evergreen_resource_set(ctx, state, offset); -} - -void evergreen_fs_resource_set(struct r600_context *ctx, struct r600_pipe_state *state, unsigned rid) -{ - unsigned offset = R_030000_RESOURCE0_WORD0 + 0x7C00 + 0x20 * rid; - - evergreen_resource_set(ctx, state, offset); -} diff --git a/src/gallium/winsys/r600/drm/r600.c b/src/gallium/winsys/r600/drm/r600.c deleted file mode 100644 index f5e53e21f5..0000000000 --- a/src/gallium/winsys/r600/drm/r600.c +++ /dev/null @@ -1,175 +0,0 @@ -/* - * Copyright 2010 Jerome Glisse <glisse@freedesktop.org> - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * on the rights to use, copy, modify, merge, publish, distribute, sub - * license, and/or sell copies of the Software, and to permit persons to whom - * the Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * 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 - * THE AUTHOR(S) AND/OR THEIR 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. - * - * Authors: - * Jerome Glisse - */ -#include "xf86drm.h" -#include "radeon_drm.h" -#include "pipe/p_compiler.h" -#include "util/u_inlines.h" -#include <pipebuffer/pb_bufmgr.h> -#include "r600_priv.h" - -enum radeon_family r600_get_family(struct radeon *r600) -{ - return r600->family; -} - -enum chip_class r600_get_family_class(struct radeon *radeon) -{ - return radeon->chip_class; -} - -struct r600_tiling_info *r600_get_tiling_info(struct radeon *radeon) -{ - return &radeon->tiling_info; -} - -static int r600_get_device(struct radeon *r600) -{ - struct drm_radeon_info info; - - r600->device = 0; - info.request = RADEON_INFO_DEVICE_ID; - info.value = (uintptr_t)&r600->device; - return drmCommandWriteRead(r600->fd, DRM_RADEON_INFO, &info, sizeof(struct drm_radeon_info)); -} - -struct radeon *r600_new(int fd, unsigned device) -{ - struct radeon *r600; - int r; - - r600 = calloc(1, sizeof(*r600)); - if (r600 == NULL) { - return NULL; - } - r600->fd = fd; - r600->device = device; - if (fd >= 0) { - r = r600_get_device(r600); - if (r) { - R600_ERR("Failed to get device id\n"); - r600_delete(r600); - return NULL; - } - } - r600->family = radeon_family_from_device(r600->device); - if (r600->family == CHIP_UNKNOWN) { - R600_ERR("Unknown chipset 0x%04X\n", r600->device); - r600_delete(r600); - return NULL; - } - switch (r600->family) { - case CHIP_R600: - case CHIP_RV610: - case CHIP_RV630: - case CHIP_RV670: - case CHIP_RV620: - case CHIP_RV635: - case CHIP_RS780: - case CHIP_RS880: - case CHIP_RV770: - case CHIP_RV730: - case CHIP_RV710: - case CHIP_RV740: - case CHIP_CEDAR: - case CHIP_REDWOOD: - case CHIP_JUNIPER: - case CHIP_CYPRESS: - case CHIP_HEMLOCK: - case CHIP_PALM: - break; - case CHIP_R100: - case CHIP_RV100: - case CHIP_RS100: - case CHIP_RV200: - case CHIP_RS200: - case CHIP_R200: - case CHIP_RV250: - case CHIP_RS300: - case CHIP_RV280: - case CHIP_R300: - case CHIP_R350: - case CHIP_RV350: - case CHIP_RV380: - case CHIP_R420: - case CHIP_R423: - case CHIP_RV410: - case CHIP_RS400: - case CHIP_RS480: - case CHIP_RS600: - case CHIP_RS690: - case CHIP_RS740: - case CHIP_RV515: - case CHIP_R520: - case CHIP_RV530: - case CHIP_RV560: - case CHIP_RV570: - case CHIP_R580: - default: - R600_ERR("unknown or unsupported chipset 0x%04X\n", r600->device); - break; - } - - /* setup class */ - switch (r600->family) { - case CHIP_R600: - case CHIP_RV610: - case CHIP_RV630: - case CHIP_RV670: - case CHIP_RV620: - case CHIP_RV635: - case CHIP_RS780: - case CHIP_RS880: - r600->chip_class = R600; - break; - case CHIP_RV770: - case CHIP_RV730: - case CHIP_RV710: - case CHIP_RV740: - r600->chip_class = R700; - break; - case CHIP_CEDAR: - case CHIP_REDWOOD: - case CHIP_JUNIPER: - case CHIP_CYPRESS: - case CHIP_HEMLOCK: - case CHIP_PALM: - r600->chip_class = EVERGREEN; - break; - default: - R600_ERR("unknown or unsupported chipset 0x%04X\n", r600->device); - break; - } - - return r600; -} - -void r600_delete(struct radeon *r600) -{ - if (r600 == NULL) - return; - drmClose(r600->fd); - free(r600); -} diff --git a/src/gallium/winsys/r600/drm/r600_bo.c b/src/gallium/winsys/r600/drm/r600_bo.c index 251f009a6b..122a68884b 100644 --- a/src/gallium/winsys/r600/drm/r600_bo.c +++ b/src/gallium/winsys/r600/drm/r600_bo.c @@ -36,142 +36,152 @@ struct r600_bo *r600_bo(struct radeon *radeon, unsigned size, unsigned alignment, unsigned binding, unsigned usage) { - struct r600_bo *ws_bo = calloc(1, sizeof(struct r600_bo)); - struct pb_desc desc; - struct pb_manager *man; + struct r600_bo *bo; + struct radeon_bo *rbo; - desc.alignment = alignment; - desc.usage = (PB_USAGE_CPU_READ_WRITE | PB_USAGE_GPU_READ_WRITE); - ws_bo->size = size; + if (binding & (PIPE_BIND_CONSTANT_BUFFER | PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER)) { + bo = r600_bomgr_bo_create(radeon->bomgr, size, alignment, *radeon->cfence); + if (bo) { + return bo; + } + } - if (binding & (PIPE_BIND_CONSTANT_BUFFER | PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER)) - man = radeon->cman; - else - man = radeon->kman; + rbo = radeon_bo(radeon, 0, size, alignment); + if (rbo == NULL) { + return NULL; + } + + bo = calloc(1, sizeof(struct r600_bo)); + bo->size = size; + bo->alignment = alignment; + bo->bo = rbo; + if (binding & (PIPE_BIND_CONSTANT_BUFFER | PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER)) { + r600_bomgr_bo_init(radeon->bomgr, bo); + } /* Staging resources particpate in transfers and blits only * and are used for uploads and downloads from regular * resources. We generate them internally for some transfers. */ if (usage == PIPE_USAGE_STAGING) - ws_bo->domains = RADEON_GEM_DOMAIN_CPU | RADEON_GEM_DOMAIN_GTT; - else - ws_bo->domains = (RADEON_GEM_DOMAIN_CPU | - RADEON_GEM_DOMAIN_GTT | - RADEON_GEM_DOMAIN_VRAM); - - - ws_bo->pb = man->create_buffer(man, size, &desc); - if (ws_bo->pb == NULL) { - free(ws_bo); - return NULL; - } + bo->domains = RADEON_GEM_DOMAIN_CPU | RADEON_GEM_DOMAIN_GTT; + else + bo->domains = (RADEON_GEM_DOMAIN_CPU | + RADEON_GEM_DOMAIN_GTT | + RADEON_GEM_DOMAIN_VRAM); - pipe_reference_init(&ws_bo->reference, 1); - return ws_bo; + pipe_reference_init(&bo->reference, 1); + return bo; } struct r600_bo *r600_bo_handle(struct radeon *radeon, unsigned handle, unsigned *array_mode) { - struct r600_bo *ws_bo = calloc(1, sizeof(struct r600_bo)); - struct radeon_bo *bo; + struct r600_bo *bo = calloc(1, sizeof(struct r600_bo)); + struct radeon_bo *rbo; - ws_bo->pb = radeon_bo_pb_create_buffer_from_handle(radeon->kman, handle); - if (!ws_bo->pb) { - free(ws_bo); + rbo = bo->bo = radeon_bo(radeon, handle, 0, 0); + if (rbo == NULL) { + free(bo); return NULL; } - bo = radeon_bo_pb_get_bo(ws_bo->pb); - ws_bo->size = bo->size; - ws_bo->domains = (RADEON_GEM_DOMAIN_CPU | - RADEON_GEM_DOMAIN_GTT | - RADEON_GEM_DOMAIN_VRAM); + bo->size = rbo->size; + bo->domains = (RADEON_GEM_DOMAIN_CPU | + RADEON_GEM_DOMAIN_GTT | + RADEON_GEM_DOMAIN_VRAM); - pipe_reference_init(&ws_bo->reference, 1); + pipe_reference_init(&bo->reference, 1); - radeon_bo_get_tiling_flags(radeon, bo, &ws_bo->tiling_flags, - &ws_bo->kernel_pitch); + radeon_bo_get_tiling_flags(radeon, rbo, &bo->tiling_flags, &bo->kernel_pitch); if (array_mode) { - if (ws_bo->tiling_flags) { - if (ws_bo->tiling_flags & RADEON_TILING_MICRO) - *array_mode = V_0280A0_ARRAY_1D_TILED_THIN1; - if ((ws_bo->tiling_flags & (RADEON_TILING_MICRO | RADEON_TILING_MACRO)) == - (RADEON_TILING_MICRO | RADEON_TILING_MACRO)) + if (bo->tiling_flags) { + if (bo->tiling_flags & RADEON_TILING_MACRO) *array_mode = V_0280A0_ARRAY_2D_TILED_THIN1; + else if (bo->tiling_flags & RADEON_TILING_MICRO) + *array_mode = V_0280A0_ARRAY_1D_TILED_THIN1; } else { *array_mode = 0; } } - return ws_bo; + return bo; } void *r600_bo_map(struct radeon *radeon, struct r600_bo *bo, unsigned usage, void *ctx) { - return pb_map(bo->pb, usage, ctx); + struct pipe_context *pctx = ctx; + + if (usage & PB_USAGE_UNSYNCHRONIZED) { + radeon_bo_map(radeon, bo->bo); + return (uint8_t *) bo->bo->data + bo->offset; + } + + if (p_atomic_read(&bo->bo->reference.count) > 1) { + if (usage & PB_USAGE_DONTBLOCK) { + return NULL; + } + if (ctx) { + pctx->flush(pctx, NULL); + } + } + + if (usage & PB_USAGE_DONTBLOCK) { + uint32_t domain; + + if (radeon_bo_busy(radeon, bo->bo, &domain)) + return NULL; + if (radeon_bo_map(radeon, bo->bo)) { + return NULL; + } + goto out; + } + + radeon_bo_map(radeon, bo->bo); + if (radeon_bo_wait(radeon, bo->bo)) { + radeon_bo_unmap(radeon, bo->bo); + return NULL; + } + +out: + return (uint8_t *) bo->bo->data + bo->offset; } void r600_bo_unmap(struct radeon *radeon, struct r600_bo *bo) { - pb_unmap(bo->pb); + radeon_bo_unmap(radeon, bo->bo); } -static void r600_bo_destroy(struct radeon *radeon, struct r600_bo *bo) +void r600_bo_destroy(struct radeon *radeon, struct r600_bo *bo) { - if (bo->pb) - pb_reference(&bo->pb, NULL); + if (bo->manager_id) { + if (!r600_bomgr_bo_destroy(radeon->bomgr, bo)) { + /* destroy is delayed by buffer manager */ + return; + } + } + radeon_bo_reference(radeon, &bo->bo, NULL); free(bo); } -void r600_bo_reference(struct radeon *radeon, struct r600_bo **dst, - struct r600_bo *src) +void r600_bo_reference(struct radeon *radeon, struct r600_bo **dst, struct r600_bo *src) { struct r600_bo *old = *dst; - + if (pipe_reference(&(*dst)->reference, &src->reference)) { r600_bo_destroy(radeon, old); } *dst = src; } -unsigned r600_bo_get_handle(struct r600_bo *pb_bo) -{ - struct radeon_bo *bo; - - bo = radeon_bo_pb_get_bo(pb_bo->pb); - if (!bo) - return 0; - - return bo->handle; -} - -unsigned r600_bo_get_size(struct r600_bo *pb_bo) -{ - struct radeon_bo *bo; - - bo = radeon_bo_pb_get_bo(pb_bo->pb); - if (!bo) - return 0; - - return bo->size; -} - -boolean r600_bo_get_winsys_handle(struct radeon *radeon, struct r600_bo *pb_bo, +boolean r600_bo_get_winsys_handle(struct radeon *radeon, struct r600_bo *bo, unsigned stride, struct winsys_handle *whandle) { - struct radeon_bo *bo; - - bo = radeon_bo_pb_get_bo(pb_bo->pb); - if (!bo) - return FALSE; - whandle->stride = stride; switch(whandle->type) { case DRM_API_HANDLE_TYPE_KMS: - whandle->handle = r600_bo_get_handle(pb_bo); + whandle->handle = r600_bo_get_handle(bo); break; case DRM_API_HANDLE_TYPE_SHARED: - if (radeon_bo_get_name(radeon, bo, &whandle->handle)) + if (radeon_bo_get_name(radeon, bo->bo, &whandle->handle)) return FALSE; break; default: diff --git a/src/gallium/winsys/r600/drm/r600_bomgr.c b/src/gallium/winsys/r600/drm/r600_bomgr.c new file mode 100644 index 0000000000..446ef0f9cf --- /dev/null +++ b/src/gallium/winsys/r600/drm/r600_bomgr.c @@ -0,0 +1,161 @@ +/* + * Copyright 2010 VMWare. + * Copyright 2010 Red Hat Inc. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * 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 + * THE AUTHOR(S) AND/OR THEIR 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. + * + * Authors: + * Jose Fonseca <jrfonseca-at-vmware-dot-com> + * Thomas Hellström <thomas-at-vmware-dot-com> + * Jerome Glisse <jglisse@redhat.com> + */ +#include <util/u_memory.h> +#include <util/u_double_list.h> +#include <util/u_time.h> +#include <pipebuffer/pb_bufmgr.h> +#include "r600_priv.h" + +static void r600_bomgr_timeout_flush(struct r600_bomgr *mgr) +{ + struct r600_bo *bo, *tmp; + int64_t now; + + now = os_time_get(); + LIST_FOR_EACH_ENTRY_SAFE(bo, tmp, &mgr->delayed, list) { + if(!os_time_timeout(bo->start, bo->end, now)) + break; + + mgr->num_delayed--; + bo->manager_id = 0; + LIST_DEL(&bo->list); + r600_bo_destroy(mgr->radeon, bo); + } +} + +static INLINE int r600_bo_is_compat(struct r600_bomgr *mgr, + struct r600_bo *bo, + unsigned size, + unsigned alignment, + unsigned cfence) +{ + if(bo->size < size) { + return 0; + } + + /* be lenient with size */ + if(bo->size >= 2*size) { + return 0; + } + + if(!pb_check_alignment(alignment, bo->alignment)) { + return 0; + } + + if (!fence_is_after(cfence, bo->fence)) { + return 0; + } + + return 1; +} + +struct r600_bo *r600_bomgr_bo_create(struct r600_bomgr *mgr, + unsigned size, + unsigned alignment, + unsigned cfence) +{ + struct r600_bo *bo, *tmp; + int64_t now; + + + pipe_mutex_lock(mgr->mutex); + + now = os_time_get(); + LIST_FOR_EACH_ENTRY_SAFE(bo, tmp, &mgr->delayed, list) { + if(r600_bo_is_compat(mgr, bo, size, alignment, cfence)) { + LIST_DEL(&bo->list); + --mgr->num_delayed; + r600_bomgr_timeout_flush(mgr); + pipe_mutex_unlock(mgr->mutex); + LIST_INITHEAD(&bo->list); + pipe_reference_init(&bo->reference, 1); + return bo; + } + + if(os_time_timeout(bo->start, bo->end, now)) { + mgr->num_delayed--; + bo->manager_id = 0; + LIST_DEL(&bo->list); + r600_bo_destroy(mgr->radeon, bo); + } + } + + pipe_mutex_unlock(mgr->mutex); + return NULL; +} + +void r600_bomgr_bo_init(struct r600_bomgr *mgr, struct r600_bo *bo) +{ + LIST_INITHEAD(&bo->list); + bo->manager_id = 1; +} + +bool r600_bomgr_bo_destroy(struct r600_bomgr *mgr, struct r600_bo *bo) +{ + bo->start = os_time_get(); + bo->end = bo->start + mgr->usecs; + pipe_mutex_lock(mgr->mutex); + LIST_ADDTAIL(&bo->list, &mgr->delayed); + ++mgr->num_delayed; + pipe_mutex_unlock(mgr->mutex); + return FALSE; +} + +void r600_bomgr_destroy(struct r600_bomgr *mgr) +{ + struct r600_bo *bo, *tmp; + + pipe_mutex_lock(mgr->mutex); + LIST_FOR_EACH_ENTRY_SAFE(bo, tmp, &mgr->delayed, list) { + mgr->num_delayed--; + bo->manager_id = 0; + LIST_DEL(&bo->list); + r600_bo_destroy(mgr->radeon, bo); + } + pipe_mutex_unlock(mgr->mutex); + + FREE(mgr); +} + +struct r600_bomgr *r600_bomgr_create(struct radeon *radeon, unsigned usecs) +{ + struct r600_bomgr *mgr; + + mgr = CALLOC_STRUCT(r600_bomgr); + if (mgr == NULL) + return NULL; + + mgr->radeon = radeon; + mgr->usecs = usecs; + LIST_INITHEAD(&mgr->delayed); + mgr->num_delayed = 0; + pipe_mutex_init(mgr->mutex); + + return mgr; +} diff --git a/src/gallium/winsys/r600/drm/r600_drm.c b/src/gallium/winsys/r600/drm/r600_drm.c index 8c847122f8..c081abb4dc 100644 --- a/src/gallium/winsys/r600/drm/r600_drm.c +++ b/src/gallium/winsys/r600/drm/r600_drm.c @@ -41,11 +41,48 @@ #define RADEON_INFO_TILING_CONFIG 0x6 #endif -static struct radeon *radeon_new(int fd, unsigned device); +#ifndef RADEON_INFO_CLOCK_CRYSTAL_FREQ +#define RADEON_INFO_CLOCK_CRYSTAL_FREQ 0x9 +#endif + +#ifndef RADEON_INFO_NUM_BACKENDS +#define RADEON_INFO_NUM_BACKENDS 0xa +#endif + +enum radeon_family r600_get_family(struct radeon *r600) +{ + return r600->family; +} + +enum chip_class r600_get_family_class(struct radeon *radeon) +{ + return radeon->chip_class; +} + +struct r600_tiling_info *r600_get_tiling_info(struct radeon *radeon) +{ + return &radeon->tiling_info; +} + +unsigned r600_get_clock_crystal_freq(struct radeon *radeon) +{ + return radeon->clock_crystal_freq; +} + +unsigned r600_get_num_backends(struct radeon *radeon) +{ + return radeon->num_backends; +} + +unsigned r600_get_minor_version(struct radeon *radeon) +{ + return radeon->minor_version; +} + static int radeon_get_device(struct radeon *radeon) { - struct drm_radeon_info info; + struct drm_radeon_info info = {}; int r; radeon->device = 0; @@ -56,20 +93,8 @@ static int radeon_get_device(struct radeon *radeon) return r; } -static int radeon_drm_get_tiling(struct radeon *radeon) +static int r600_interpret_tiling(struct radeon *radeon, uint32_t tiling_config) { - struct drm_radeon_info info; - int r; - uint32_t tiling_config; - - info.request = RADEON_INFO_TILING_CONFIG; - info.value = (uintptr_t)&tiling_config; - r = drmCommandWriteRead(radeon->fd, DRM_RADEON_INFO, &info, - sizeof(struct drm_radeon_info)); - - if (r) - return 0; - switch ((tiling_config & 0xe) >> 1) { case 0: radeon->tiling_info.num_channels = 1; @@ -111,10 +136,117 @@ static int radeon_drm_get_tiling(struct radeon *radeon) return 0; } +static int eg_interpret_tiling(struct radeon *radeon, uint32_t tiling_config) +{ + switch (tiling_config & 0xf) { + case 0: + radeon->tiling_info.num_channels = 1; + break; + case 1: + radeon->tiling_info.num_channels = 2; + break; + case 2: + radeon->tiling_info.num_channels = 4; + break; + case 3: + radeon->tiling_info.num_channels = 8; + break; + default: + return -EINVAL; + } + + radeon->tiling_info.num_banks = (tiling_config & 0xf0) >> 4; + + switch ((tiling_config & 0xf00) >> 8) { + case 0: + radeon->tiling_info.group_bytes = 256; + break; + case 1: + radeon->tiling_info.group_bytes = 512; + break; + default: + return -EINVAL; + } + return 0; +} + +static int radeon_drm_get_tiling(struct radeon *radeon) +{ + struct drm_radeon_info info; + int r; + uint32_t tiling_config = 0; + + info.request = RADEON_INFO_TILING_CONFIG; + info.value = (uintptr_t)&tiling_config; + r = drmCommandWriteRead(radeon->fd, DRM_RADEON_INFO, &info, + sizeof(struct drm_radeon_info)); + + if (r) + return 0; + + if (radeon->chip_class == R600 || radeon->chip_class == R700) { + r = r600_interpret_tiling(radeon, tiling_config); + } else { + r = eg_interpret_tiling(radeon, tiling_config); + } + return r; +} + +static int radeon_get_clock_crystal_freq(struct radeon *radeon) +{ + struct drm_radeon_info info; + uint32_t clock_crystal_freq; + int r; + + radeon->device = 0; + info.request = RADEON_INFO_CLOCK_CRYSTAL_FREQ; + info.value = (uintptr_t)&clock_crystal_freq; + r = drmCommandWriteRead(radeon->fd, DRM_RADEON_INFO, &info, + sizeof(struct drm_radeon_info)); + if (r) + return r; + + radeon->clock_crystal_freq = clock_crystal_freq; + return 0; +} + + +static int radeon_get_num_backends(struct radeon *radeon) +{ + struct drm_radeon_info info; + uint32_t num_backends; + int r; + + radeon->device = 0; + info.request = RADEON_INFO_NUM_BACKENDS; + info.value = (uintptr_t)&num_backends; + r = drmCommandWriteRead(radeon->fd, DRM_RADEON_INFO, &info, + sizeof(struct drm_radeon_info)); + if (r) + return r; + + radeon->num_backends = num_backends; + return 0; +} + + +static int radeon_init_fence(struct radeon *radeon) +{ + radeon->fence = 1; + radeon->fence_bo = r600_bo(radeon, 4096, 0, 0, 0); + if (radeon->fence_bo == NULL) { + return -ENOMEM; + } + radeon->cfence = r600_bo_map(radeon, radeon->fence_bo, PB_USAGE_UNSYNCHRONIZED, NULL); + *radeon->cfence = 0; + return 0; +} + static struct radeon *radeon_new(int fd, unsigned device) { struct radeon *radeon; int r; + drmVersionPtr version; radeon = calloc(1, sizeof(*radeon)); if (radeon == NULL) { @@ -123,71 +255,32 @@ static struct radeon *radeon_new(int fd, unsigned device) radeon->fd = fd; radeon->device = device; radeon->refcount = 1; - if (fd >= 0) { - r = radeon_get_device(radeon); - if (r) { - fprintf(stderr, "Failed to get device id\n"); - return radeon_decref(radeon); - } + + version = drmGetVersion(radeon->fd); + if (version->version_major != 2) { + fprintf(stderr, "%s: DRM version is %d.%d.%d but this driver is " + "only compatible with 2.x.x\n", __FUNCTION__, + version->version_major, version->version_minor, + version->version_patchlevel); + drmFreeVersion(version); + exit(1); + } + + radeon->minor_version = version->version_minor; + + drmFreeVersion(version); + + r = radeon_get_device(radeon); + if (r) { + fprintf(stderr, "Failed to get device id\n"); + return radeon_decref(radeon); } + radeon->family = radeon_family_from_device(radeon->device); if (radeon->family == CHIP_UNKNOWN) { fprintf(stderr, "Unknown chipset 0x%04X\n", radeon->device); return radeon_decref(radeon); } - switch (radeon->family) { - case CHIP_R600: - case CHIP_RV610: - case CHIP_RV630: - case CHIP_RV670: - case CHIP_RV620: - case CHIP_RV635: - case CHIP_RS780: - case CHIP_RS880: - case CHIP_RV770: - case CHIP_RV730: - case CHIP_RV710: - case CHIP_RV740: - case CHIP_CEDAR: - case CHIP_REDWOOD: - case CHIP_JUNIPER: - case CHIP_CYPRESS: - case CHIP_HEMLOCK: - case CHIP_PALM: - break; - case CHIP_R100: - case CHIP_RV100: - case CHIP_RS100: - case CHIP_RV200: - case CHIP_RS200: - case CHIP_R200: - case CHIP_RV250: - case CHIP_RS300: - case CHIP_RV280: - case CHIP_R300: - case CHIP_R350: - case CHIP_RV350: - case CHIP_RV380: - case CHIP_R420: - case CHIP_R423: - case CHIP_RV410: - case CHIP_RS400: - case CHIP_RS480: - case CHIP_RS600: - case CHIP_RS690: - case CHIP_RS740: - case CHIP_RV515: - case CHIP_R520: - case CHIP_RV530: - case CHIP_RV560: - case CHIP_RV570: - case CHIP_R580: - default: - fprintf(stderr, "%s unknown or unsupported chipset 0x%04X\n", - __func__, radeon->device); - break; - } - /* setup class */ switch (radeon->family) { case CHIP_R600: @@ -216,6 +309,9 @@ static struct radeon *radeon_new(int fd, unsigned device) case CHIP_CYPRESS: case CHIP_HEMLOCK: case CHIP_PALM: + case CHIP_BARTS: + case CHIP_TURKS: + case CHIP_CAICOS: radeon->chip_class = EVERGREEN; /* set default group bytes, overridden by tiling info ioctl */ radeon->tiling_info.group_bytes = 512; @@ -226,16 +322,24 @@ static struct radeon *radeon_new(int fd, unsigned device) break; } - if (radeon->chip_class == R600 || radeon->chip_class == R700) { - if (radeon_drm_get_tiling(radeon)) - return NULL; - } - radeon->kman = radeon_bo_pbmgr_create(radeon); - if (!radeon->kman) + if (radeon_drm_get_tiling(radeon)) return NULL; - radeon->cman = pb_cache_manager_create(radeon->kman, 1000000); - if (!radeon->cman) + + /* get the GPU counter frequency, failure is non fatal */ + radeon_get_clock_crystal_freq(radeon); + + if (radeon->minor_version >= 9) + radeon_get_num_backends(radeon); + + radeon->bomgr = r600_bomgr_create(radeon, 1000000); + if (radeon->bomgr == NULL) { return NULL; + } + r = radeon_init_fence(radeon); + if (r) { + radeon_decref(radeon); + return NULL; + } return radeon; } @@ -252,14 +356,12 @@ struct radeon *radeon_decref(struct radeon *radeon) return NULL; } - if (radeon->cman) - radeon->cman->destroy(radeon->cman); - - if (radeon->kman) - radeon->kman->destroy(radeon->kman); + if (radeon->fence_bo) { + r600_bo_reference(radeon, &radeon->fence_bo, NULL); + } - if (radeon->fd >= 0) - drmClose(radeon->fd); + if (radeon->bomgr) + r600_bomgr_destroy(radeon->bomgr); free(radeon); return NULL; diff --git a/src/gallium/winsys/r600/drm/r600_hw_context.c b/src/gallium/winsys/r600/drm/r600_hw_context.c index 0f2724f61c..a7c21784e5 100644 --- a/src/gallium/winsys/r600/drm/r600_hw_context.c +++ b/src/gallium/winsys/r600/drm/r600_hw_context.c @@ -28,39 +28,25 @@ #include <string.h> #include <stdlib.h> #include <assert.h> +#include <pipe/p_compiler.h> +#include <util/u_inlines.h> +#include <util/u_memory.h> +#include <pipebuffer/pb_bufmgr.h> #include "xf86drm.h" -#include "r600.h" -#include "r600d.h" #include "radeon_drm.h" -#include "bof.h" -#include "pipe/p_compiler.h" -#include "util/u_inlines.h" -#include "util/u_memory.h" -#include <pipebuffer/pb_bufmgr.h> #include "r600_priv.h" +#include "bof.h" +#include "r600d.h" #define GROUP_FORCE_NEW_BLOCK 0 -int r600_context_init_fence(struct r600_context *ctx) -{ - ctx->fence = 1; - ctx->fence_bo = r600_bo(ctx->radeon, 4096, 0, 0, 0); - if (ctx->fence_bo == NULL) { - return -ENOMEM; - } - ctx->cfence = r600_bo_map(ctx->radeon, ctx->fence_bo, PB_USAGE_UNSYNCHRONIZED, NULL); - *ctx->cfence = 0; - LIST_INITHEAD(&ctx->fenced_bo); - return 0; -} - static void INLINE r600_context_update_fenced_list(struct r600_context *ctx) { for (int i = 0; i < ctx->creloc; i++) { if (!LIST_IS_EMPTY(&ctx->bo[i]->fencedlist)) LIST_DELINIT(&ctx->bo[i]->fencedlist); LIST_ADDTAIL(&ctx->bo[i]->fencedlist, &ctx->fenced_bo); - ctx->bo[i]->fence = ctx->fence; + ctx->bo[i]->fence = ctx->radeon->fence; ctx->bo[i]->ctx = ctx; } } @@ -71,7 +57,7 @@ static void INLINE r600_context_fence_wraparound(struct r600_context *ctx, unsig struct radeon_bo *tmp; LIST_FOR_EACH_ENTRY_SAFE(bo, tmp, &ctx->fenced_bo, fencedlist) { - if (bo->fence <= *ctx->cfence) { + if (bo->fence <= *ctx->radeon->cfence) { LIST_DELINIT(&bo->fencedlist); bo->fence = 0; } else { @@ -121,7 +107,7 @@ int r600_context_add_block(struct r600_context *ctx, const struct r600_reg *reg, /* initialize block */ block->start_offset = reg[i].offset; - block->pm4[block->pm4_ndwords++] = PKT3(reg[i].opcode, n); + block->pm4[block->pm4_ndwords++] = PKT3(reg[i].opcode, n, 0); block->pm4[block->pm4_ndwords++] = (block->start_offset - reg[i].offset_base) >> 2; block->reg = &block->pm4[block->pm4_ndwords]; block->pm4_ndwords += n; @@ -133,7 +119,7 @@ int r600_context_add_block(struct r600_context *ctx, const struct r600_reg *reg, block->nbo++; assert(block->nbo < R600_BLOCK_MAX_BO); block->pm4_bo_index[j] = block->nbo; - block->pm4[block->pm4_ndwords++] = PKT3(PKT3_NOP, 0); + block->pm4[block->pm4_ndwords++] = PKT3(PKT3_NOP, 0, 0); block->pm4[block->pm4_ndwords++] = 0x00000000; block->reloc[block->nbo].flush_flags = reg[i+j].flush_flags; block->reloc[block->nbo].flush_mask = reg[i+j].flush_mask; @@ -632,9 +618,6 @@ void r600_context_fini(struct r600_context *ctx) free(ctx->pm4); r600_context_clear_fenced_bo(ctx); - if (ctx->fence_bo) { - r600_bo_reference(ctx->radeon, &ctx->fence_bo, NULL); - } memset(ctx, 0, sizeof(struct r600_context)); } @@ -654,7 +637,8 @@ int r600_context_init(struct r600_context *ctx, struct radeon *radeon) ctx->range[i].end_offset = ((i + 1) << ctx->hash_shift) - 1; ctx->range[i].blocks = calloc(1 << ctx->hash_shift, sizeof(void*)); if (ctx->range[i].blocks == NULL) { - return -ENOMEM; + r = -ENOMEM; + goto out_err; } } @@ -763,19 +747,30 @@ int r600_context_init(struct r600_context *ctx, struct radeon *radeon) /* save 16dwords space for fence mecanism */ ctx->pm4_ndwords -= 16; - r = r600_context_init_fence(ctx); - if (r) { - goto out_err; - } + LIST_INITHEAD(&ctx->fenced_bo); /* init dirty list */ LIST_INITHEAD(&ctx->dirty); + + ctx->max_db = 4; + return 0; out_err: r600_context_fini(ctx); return r; } +static void rv6xx_context_surface_base_update(struct r600_context *ctx, + unsigned base_update_flags) +{ + /* need to emit surface base update on rv6xx */ + if ((ctx->radeon->family > CHIP_R600) && + (ctx->radeon->family < CHIP_RV770)) { + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SURFACE_BASE_UPDATE, 0, 0); + ctx->pm4[ctx->pm4_cdwords++] = base_update_flags; + } +} + void r600_context_bo_flush(struct r600_context *ctx, unsigned flush_flags, unsigned flush_mask, struct r600_bo *rbo) { @@ -787,12 +782,12 @@ void r600_context_bo_flush(struct r600_context *ctx, unsigned flush_flags, bo->last_flush &= flush_mask; return; } - ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SURFACE_SYNC, 3); + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SURFACE_SYNC, 3, ctx->predicate_drawing); ctx->pm4[ctx->pm4_cdwords++] = flush_flags; ctx->pm4[ctx->pm4_cdwords++] = (bo->size + 255) >> 8; ctx->pm4[ctx->pm4_cdwords++] = 0x00000000; ctx->pm4[ctx->pm4_cdwords++] = 0x0000000A; - ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0); + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, ctx->predicate_drawing); ctx->pm4[ctx->pm4_cdwords++] = bo->reloc_id; bo->last_flush = (bo->last_flush | flush_flags) & flush_mask; } @@ -814,6 +809,7 @@ void r600_context_bo_reloc(struct r600_context *ctx, u32 *pm4, struct r600_bo *r ctx->reloc[ctx->creloc].write_domain = rbo->domains & (RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM); ctx->reloc[ctx->creloc].flags = 0; radeon_bo_reference(ctx->radeon, &ctx->bo[ctx->creloc], bo); + rbo->fence = ctx->radeon->fence; ctx->creloc++; /* set PKT3 to point to proper reloc */ *pm4 = bo->reloc_id; @@ -836,6 +832,7 @@ void r600_context_pipe_state_set(struct r600_context *ctx, struct r600_pipe_stat /* find relocation */ id = block->pm4_bo_index[id]; r600_bo_reference(ctx->radeon, &block->reloc[id].bo, state->regs[i].bo); + state->regs[i].bo->fence = ctx->radeon->fence; } if (!(block->status & R600_BLOCK_STATUS_DIRTY)) { block->status |= R600_BLOCK_STATUS_ENABLED; @@ -875,10 +872,13 @@ static inline void r600_context_pipe_state_set_resource(struct r600_context *ctx */ r600_bo_reference(ctx->radeon, &block->reloc[1].bo, state->regs[0].bo); r600_bo_reference(ctx->radeon, &block->reloc[2].bo, state->regs[0].bo); + state->regs[0].bo->fence = ctx->radeon->fence; } else { /* TEXTURE RESOURCE */ r600_bo_reference(ctx->radeon, &block->reloc[1].bo, state->regs[2].bo); r600_bo_reference(ctx->radeon, &block->reloc[2].bo, state->regs[3].bo); + state->regs[2].bo->fence = ctx->radeon->fence; + state->regs[3].bo->fence = ctx->radeon->fence; } if (!(block->status & R600_BLOCK_STATUS_DIRTY)) { block->status |= R600_BLOCK_STATUS_ENABLED; @@ -1002,6 +1002,7 @@ void r600_context_draw(struct r600_context *ctx, const struct r600_draw *draw) unsigned ndwords = 9; struct r600_block *dirty_block = NULL; struct r600_block *next_block; + unsigned rv6xx_surface_base_update = 0; if (draw->indices) { ndwords = 13; @@ -1024,10 +1025,14 @@ void r600_context_draw(struct r600_context *ctx, const struct r600_draw *draw) for (int i = 0; i < 8; i++) { if (cb[i]) { ndwords += 7; + rv6xx_surface_base_update |= SURFACE_BASE_UPDATE_COLOR(i); } } - if (db) + if (db) { ndwords += 7; + rv6xx_surface_base_update |= SURFACE_BASE_UPDATE_DEPTH; + } + /* XXX also need to update SURFACE_BASE_UPDATE_STRMOUT when we support it */ /* queries need some special values */ if (ctx->num_query_running) { @@ -1054,30 +1059,34 @@ void r600_context_draw(struct r600_context *ctx, const struct r600_draw *draw) } /* enough room to copy packet */ - LIST_FOR_EACH_ENTRY_SAFE(dirty_block, next_block, &ctx->dirty,list) { + LIST_FOR_EACH_ENTRY_SAFE(dirty_block, next_block, &ctx->dirty, list) { r600_context_block_emit_dirty(ctx, dirty_block); } + /* rv6xx surface base udpate */ + if (rv6xx_surface_base_update) + rv6xx_context_surface_base_update(ctx, rv6xx_surface_base_update); + /* draw packet */ - ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_INDEX_TYPE, 0); + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_INDEX_TYPE, 0, ctx->predicate_drawing); ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_index_type; - ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NUM_INSTANCES, 0); + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NUM_INSTANCES, 0, ctx->predicate_drawing); ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_num_instances; if (draw->indices) { - ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_DRAW_INDEX, 3); + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_DRAW_INDEX, 3, ctx->predicate_drawing); ctx->pm4[ctx->pm4_cdwords++] = draw->indices_bo_offset + r600_bo_offset(draw->indices); ctx->pm4[ctx->pm4_cdwords++] = 0; ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_num_indices; ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_draw_initiator; - ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0); + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, ctx->predicate_drawing); ctx->pm4[ctx->pm4_cdwords++] = 0; r600_context_bo_reloc(ctx, &ctx->pm4[ctx->pm4_cdwords - 1], draw->indices); } else { - ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_DRAW_INDEX_AUTO, 1); + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_DRAW_INDEX_AUTO, 1, ctx->predicate_drawing); ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_num_indices; ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_draw_initiator; } - ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0); + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0, ctx->predicate_drawing); ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_EVENT) | EVENT_INDEX(0); /* flush color buffer */ @@ -1099,7 +1108,7 @@ void r600_context_draw(struct r600_context *ctx, const struct r600_draw *draw) void r600_context_flush(struct r600_context *ctx) { - struct drm_radeon_cs drmib; + struct drm_radeon_cs drmib = {}; struct drm_radeon_cs_chunk chunks[2]; uint64_t chunk_array[2]; unsigned fence; @@ -1112,17 +1121,17 @@ void r600_context_flush(struct r600_context *ctx) r600_context_queries_suspend(ctx); /* emit fence */ - ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0); + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0, 0); ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4); - ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE_EOP, 4); + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0); ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5); ctx->pm4[ctx->pm4_cdwords++] = 0; ctx->pm4[ctx->pm4_cdwords++] = (1 << 29) | (0 << 24); - ctx->pm4[ctx->pm4_cdwords++] = ctx->fence; + ctx->pm4[ctx->pm4_cdwords++] = ctx->radeon->fence; ctx->pm4[ctx->pm4_cdwords++] = 0; - ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0); + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); ctx->pm4[ctx->pm4_cdwords++] = 0; - r600_context_bo_reloc(ctx, &ctx->pm4[ctx->pm4_cdwords - 1], ctx->fence_bo); + r600_context_bo_reloc(ctx, &ctx->pm4[ctx->pm4_cdwords - 1], ctx->radeon->fence_bo); #if 1 /* emit cs */ @@ -1139,18 +1148,18 @@ void r600_context_flush(struct r600_context *ctx) r = drmCommandWriteRead(ctx->radeon->fd, DRM_RADEON_CS, &drmib, sizeof(struct drm_radeon_cs)); #else - *ctx->cfence = ctx->fence; + *ctx->radeon->cfence = ctx->radeon->fence; #endif r600_context_update_fenced_list(ctx); - fence = ctx->fence + 1; - if (fence < ctx->fence) { + fence = ctx->radeon->fence + 1; + if (fence < ctx->radeon->fence) { /* wrap around */ fence = 1; r600_context_fence_wraparound(ctx, fence); } - ctx->fence = fence; + ctx->radeon->fence = fence; /* restart */ for (int i = 0; i < ctx->creloc; i++) { @@ -1261,44 +1270,90 @@ out_err: bof_decref(root); } -static void r600_query_result(struct r600_context *ctx, struct r600_query *query) +static boolean r600_query_result(struct r600_context *ctx, struct r600_query *query, boolean wait) { u64 start, end; u32 *results; int i; + int size; + + if (wait) + results = r600_bo_map(ctx->radeon, query->buffer, PB_USAGE_CPU_READ, NULL); + else + results = r600_bo_map(ctx->radeon, query->buffer, PB_USAGE_DONTBLOCK | PB_USAGE_CPU_READ, NULL); + if (!results) + return FALSE; - results = r600_bo_map(ctx->radeon, query->buffer, 0, NULL); - for (i = 0; i < query->num_results; i += 4) { + size = query->num_results * (query->type == PIPE_QUERY_OCCLUSION_COUNTER ? ctx->max_db : 1); + for (i = 0; i < size; i += 4) { start = (u64)results[i] | (u64)results[i + 1] << 32; end = (u64)results[i + 2] | (u64)results[i + 3] << 32; - if ((start & 0x8000000000000000UL) && (end & 0x8000000000000000UL)) { + if (((start & 0x8000000000000000UL) && (end & 0x8000000000000000UL)) + || query->type == PIPE_QUERY_TIME_ELAPSED) { query->result += end - start; } } r600_bo_unmap(ctx->radeon, query->buffer); query->num_results = 0; + + return TRUE; } void r600_query_begin(struct r600_context *ctx, struct r600_query *query) { - /* query request needs 6 dwords for begin + 6 dwords for end */ - if ((12 + ctx->pm4_cdwords) > ctx->pm4_ndwords) { + unsigned required_space; + int num_backends = r600_get_num_backends(ctx->radeon); + + /* query request needs 6/8 dwords for begin + 6/8 dwords for end */ + if (query->type == PIPE_QUERY_TIME_ELAPSED) + required_space = 16; + else + required_space = 12; + + if ((required_space + ctx->pm4_cdwords) > ctx->pm4_ndwords) { /* need to flush */ r600_context_flush(ctx); } /* if query buffer is full force a flush */ - if (query->num_results >= ((query->buffer_size >> 2) - 2)) { + if (query->num_results*4 >= query->buffer_size - 16) { r600_context_flush(ctx); - r600_query_result(ctx, query); + r600_query_result(ctx, query, TRUE); } + if (query->type == PIPE_QUERY_OCCLUSION_COUNTER && + num_backends > 0 && num_backends < ctx->max_db) { + /* as per info on ZPASS the driver must set the unusued DB top bits */ + u32 *results; + int i; + + results = r600_bo_map(ctx->radeon, query->buffer, PB_USAGE_DONTBLOCK | PB_USAGE_CPU_WRITE, NULL); + if (results) { + memset(results + (query->num_results * 4), 0, ctx->max_db * 4 * 4); + + for (i = num_backends; i < ctx->max_db; i++) { + results[(i * 4)+1] = 0x80000000; + results[(i * 4)+3] = 0x80000000; + } + r600_bo_unmap(ctx->radeon, query->buffer); + } + } + /* emit begin query */ - ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 2); - ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_ZPASS_DONE) | EVENT_INDEX(1); - ctx->pm4[ctx->pm4_cdwords++] = query->num_results + r600_bo_offset(query->buffer); - ctx->pm4[ctx->pm4_cdwords++] = 0; - ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0); + if (query->type == PIPE_QUERY_TIME_ELAPSED) { + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0); + ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5); + ctx->pm4[ctx->pm4_cdwords++] = query->num_results*4 + r600_bo_offset(query->buffer); + ctx->pm4[ctx->pm4_cdwords++] = (3 << 29); + ctx->pm4[ctx->pm4_cdwords++] = 0; + ctx->pm4[ctx->pm4_cdwords++] = 0; + } else { + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 2, 0); + ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_ZPASS_DONE) | EVENT_INDEX(1); + ctx->pm4[ctx->pm4_cdwords++] = query->num_results*4 + r600_bo_offset(query->buffer); + ctx->pm4[ctx->pm4_cdwords++] = 0; + } + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); ctx->pm4[ctx->pm4_cdwords++] = 0; r600_context_bo_reloc(ctx, &ctx->pm4[ctx->pm4_cdwords - 1], query->buffer); @@ -1310,25 +1365,56 @@ void r600_query_begin(struct r600_context *ctx, struct r600_query *query) void r600_query_end(struct r600_context *ctx, struct r600_query *query) { /* emit begin query */ - ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 2); - ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_ZPASS_DONE) | EVENT_INDEX(1); - ctx->pm4[ctx->pm4_cdwords++] = query->num_results + 8 + r600_bo_offset(query->buffer); - ctx->pm4[ctx->pm4_cdwords++] = 0; - ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0); + if (query->type == PIPE_QUERY_TIME_ELAPSED) { + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0); + ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5); + ctx->pm4[ctx->pm4_cdwords++] = query->num_results*4 + 8 + r600_bo_offset(query->buffer); + ctx->pm4[ctx->pm4_cdwords++] = (3 << 29); + ctx->pm4[ctx->pm4_cdwords++] = 0; + ctx->pm4[ctx->pm4_cdwords++] = 0; + } else { + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 2, 0); + ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_ZPASS_DONE) | EVENT_INDEX(1); + ctx->pm4[ctx->pm4_cdwords++] = query->num_results*4 + 8 + r600_bo_offset(query->buffer); + ctx->pm4[ctx->pm4_cdwords++] = 0; + } + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); ctx->pm4[ctx->pm4_cdwords++] = 0; r600_context_bo_reloc(ctx, &ctx->pm4[ctx->pm4_cdwords - 1], query->buffer); - query->num_results += 16; + query->num_results += 4 * (query->type == PIPE_QUERY_OCCLUSION_COUNTER ? ctx->max_db : 1); query->state ^= R600_QUERY_STATE_STARTED; query->state |= R600_QUERY_STATE_ENDED; ctx->num_query_running--; } +void r600_query_predication(struct r600_context *ctx, struct r600_query *query, int operation, + int flag_wait) +{ + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_PREDICATION, 1, 0); + + if (operation == PREDICATION_OP_CLEAR) { + ctx->pm4[ctx->pm4_cdwords++] = 0; + ctx->pm4[ctx->pm4_cdwords++] = PRED_OP(PREDICATION_OP_CLEAR); + } else { + int results_base = query->num_results - (4 * ctx->max_db); + + if (results_base < 0) + results_base = 0; + + ctx->pm4[ctx->pm4_cdwords++] = results_base*4 + r600_bo_offset(query->buffer); + ctx->pm4[ctx->pm4_cdwords++] = PRED_OP(operation) | (flag_wait ? PREDICATION_HINT_WAIT : PREDICATION_HINT_NOWAIT_DRAW) | PREDICATION_DRAW_VISIBLE; + ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0); + ctx->pm4[ctx->pm4_cdwords++] = 0; + r600_context_bo_reloc(ctx, &ctx->pm4[ctx->pm4_cdwords - 1], query->buffer); + } +} + struct r600_query *r600_context_query_create(struct r600_context *ctx, unsigned query_type) { struct r600_query *query; - if (query_type != PIPE_QUERY_OCCLUSION_COUNTER) + if (query_type != PIPE_QUERY_OCCLUSION_COUNTER && query_type != PIPE_QUERY_TIME_ELAPSED) return NULL; query = calloc(1, sizeof(struct r600_query)); @@ -1370,8 +1456,12 @@ boolean r600_context_query_result(struct r600_context *ctx, if (query->num_results) { r600_context_flush(ctx); } - r600_query_result(ctx, query); - *result = query->result; + if (!r600_query_result(ctx, query, wait)) + return FALSE; + if (query->type == PIPE_QUERY_TIME_ELAPSED) + *result = (1000000*query->result)/r600_get_clock_crystal_freq(ctx->radeon); + else + *result = query->result; query->result = 0; return TRUE; } diff --git a/src/gallium/winsys/r600/drm/r600_priv.h b/src/gallium/winsys/r600/drm/r600_priv.h index 193af984f8..41c5ee02c3 100644 --- a/src/gallium/winsys/r600/drm/r600_priv.h +++ b/src/gallium/winsys/r600/drm/r600_priv.h @@ -30,24 +30,30 @@ #include <stdint.h> #include <stdlib.h> #include <assert.h> -#include <pipebuffer/pb_bufmgr.h> -#include "util/u_double_list.h" +#include <util/u_double_list.h> +#include <util/u_inlines.h> +#include <os/os_thread.h> #include "r600.h" +struct r600_bomgr; +struct r600_bo; + struct radeon { int fd; int refcount; unsigned device; unsigned family; enum chip_class chip_class; - struct pb_manager *kman; /* kernel bo manager */ - struct pb_manager *cman; /* cached bo manager */ - struct r600_tiling_info tiling_info; + struct r600_tiling_info tiling_info; + struct r600_bomgr *bomgr; + unsigned fence; + unsigned *cfence; + struct r600_bo *fence_bo; + unsigned clock_crystal_freq; + unsigned num_backends; + unsigned minor_version; }; -struct radeon *r600_new(int fd, unsigned device); -void r600_delete(struct radeon *r600); - struct r600_reg { unsigned opcode; unsigned offset_base; @@ -75,25 +81,49 @@ struct radeon_bo { struct r600_bo { struct pipe_reference reference; - struct pb_buffer *pb; unsigned size; unsigned tiling_flags; unsigned kernel_pitch; unsigned domains; + struct radeon_bo *bo; + unsigned fence; + /* manager data */ + struct list_head list; + unsigned manager_id; + unsigned alignment; + unsigned offset; + int64_t start; + int64_t end; }; +struct r600_bomgr { + struct radeon *radeon; + unsigned usecs; + pipe_mutex mutex; + struct list_head delayed; + unsigned num_delayed; +}; -/* radeon_pciid.c */ +/* + * r600_drm.c + */ +struct radeon *r600_new(int fd, unsigned device); +void r600_delete(struct radeon *r600); + +/* + * radeon_pciid.c + */ unsigned radeon_family_from_device(unsigned device); -/* radeon_bo.c */ +/* + * radeon_bo.c + */ struct radeon_bo *radeon_bo(struct radeon *radeon, unsigned handle, unsigned size, unsigned alignment); void radeon_bo_reference(struct radeon *radeon, struct radeon_bo **dst, struct radeon_bo *src); int radeon_bo_wait(struct radeon *radeon, struct radeon_bo *bo); int radeon_bo_busy(struct radeon *radeon, struct radeon_bo *bo, uint32_t *domain); -void radeon_bo_pbmgr_flush_maps(struct pb_manager *_mgr); int radeon_bo_fencelist(struct radeon *radeon, struct radeon_bo **bolist, uint32_t num_bo); int radeon_bo_get_tiling_flags(struct radeon *radeon, struct radeon_bo *bo, @@ -103,13 +133,9 @@ int radeon_bo_get_name(struct radeon *radeon, struct radeon_bo *bo, uint32_t *name); -/* radeon_bo_pb.c */ -struct radeon_bo *radeon_bo_pb_get_bo(struct pb_buffer *_buf); -struct pb_manager *radeon_bo_pbmgr_create(struct radeon *radeon); -struct pb_buffer *radeon_bo_pb_create_buffer_from_handle(struct pb_manager *_mgr, - uint32_t handle); - -/* r600_hw_context.c */ +/* + * r600_hw_context.c + */ int r600_context_init_fence(struct r600_context *ctx); void r600_context_bo_reloc(struct r600_context *ctx, u32 *pm4, struct r600_bo *rbo); void r600_context_bo_flush(struct r600_context *ctx, unsigned flush_flags, @@ -117,14 +143,27 @@ void r600_context_bo_flush(struct r600_context *ctx, unsigned flush_flags, struct r600_bo *r600_context_reg_bo(struct r600_context *ctx, unsigned offset); int r600_context_add_block(struct r600_context *ctx, const struct r600_reg *reg, unsigned nreg); -/* r600_bo.c */ -unsigned r600_bo_get_handle(struct r600_bo *bo); -unsigned r600_bo_get_size(struct r600_bo *bo); -static INLINE struct radeon_bo *r600_bo_get_bo(struct r600_bo *bo) -{ - return radeon_bo_pb_get_bo(bo->pb); -} +/* + * r600_bo.c + */ +void r600_bo_destroy(struct radeon *radeon, struct r600_bo *bo); +/* + * r600_bomgr.c + */ +struct r600_bomgr *r600_bomgr_create(struct radeon *radeon, unsigned usecs); +void r600_bomgr_destroy(struct r600_bomgr *mgr); +bool r600_bomgr_bo_destroy(struct r600_bomgr *mgr, struct r600_bo *bo); +void r600_bomgr_bo_init(struct r600_bomgr *mgr, struct r600_bo *bo); +struct r600_bo *r600_bomgr_bo_create(struct r600_bomgr *mgr, + unsigned size, + unsigned alignment, + unsigned cfence); + + +/* + * helpers + */ #define CTX_RANGE_ID(ctx, offset) (((offset) >> (ctx)->hash_shift) & 255) #define CTX_BLOCK_ID(ctx, offset) ((offset) & ((1 << (ctx)->hash_shift) - 1)) @@ -172,6 +211,9 @@ static inline void r600_context_block_emit_dirty(struct r600_context *ctx, struc LIST_DELINIT(&block->list); } +/* + * radeon_bo.c + */ static inline int radeon_bo_map(struct radeon *radeon, struct radeon_bo *bo) { bo->map_count++; @@ -184,4 +226,35 @@ static inline void radeon_bo_unmap(struct radeon *radeon, struct radeon_bo *bo) assert(bo->map_count >= 0); } +/* + * r600_bo + */ +static inline struct radeon_bo *r600_bo_get_bo(struct r600_bo *bo) +{ + return bo->bo; +} + +static unsigned inline r600_bo_get_handle(struct r600_bo *bo) +{ + return bo->bo->handle; +} + +static unsigned inline r600_bo_get_size(struct r600_bo *bo) +{ + return bo->size; +} + +/* + * fence + */ +static inline bool fence_is_after(unsigned fence, unsigned ofence) +{ + /* handle wrap around */ + if (fence < 0x80000000 && ofence > 0x80000000) + return TRUE; + if (fence > ofence) + return TRUE; + return FALSE; +} + #endif diff --git a/src/gallium/winsys/r600/drm/r600d.h b/src/gallium/winsys/r600/drm/r600d.h index 1c1ac76fe6..8042481804 100644 --- a/src/gallium/winsys/r600/drm/r600d.h +++ b/src/gallium/winsys/r600/drm/r600d.h @@ -90,6 +90,9 @@ #define PKT3_SET_SAMPLER 0x6E #define PKT3_SET_CTL_CONST 0x6F #define PKT3_SURFACE_BASE_UPDATE 0x73 +#define SURFACE_BASE_UPDATE_DEPTH (1 << 0) +#define SURFACE_BASE_UPDATE_COLOR(x) (2 << (x)) +#define SURFACE_BASE_UPDATE_STRMOUT(x) (0x200 << (x)) #define EVENT_TYPE_PS_PARTIAL_FLUSH 0x10 #define EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT 0x14 @@ -105,6 +108,18 @@ * 5 - TS events */ +#define PREDICATION_OP_CLEAR 0x0 +#define PREDICATION_OP_ZPASS 0x1 +#define PREDICATION_OP_PRIMCOUNT 0x2 + +#define PRED_OP(x) ((x) << 16) + +#define PREDICATION_HINT_WAIT (0 << 12) +#define PREDICATION_HINT_NOWAIT_DRAW (1 << 12) + +#define PREDICATION_DRAW_NOT_VISIBLE (0 << 8) +#define PREDICATION_DRAW_VISIBLE (1 << 8) + #define PKT_TYPE_S(x) (((x) & 0x3) << 30) #define PKT_TYPE_G(x) (((x) >> 30) & 0x3) #define PKT_TYPE_C 0x3FFFFFFF @@ -117,8 +132,9 @@ #define PKT3_IT_OPCODE_S(x) (((x) & 0xFF) << 8) #define PKT3_IT_OPCODE_G(x) (((x) >> 8) & 0xFF) #define PKT3_IT_OPCODE_C 0xFFFF00FF +#define PKT3_PRED_S(x) (((x) >> 0) & 0x1) #define PKT0(index, count) (PKT_TYPE_S(0) | PKT0_BASE_INDEX_S(index) | PKT_COUNT_S(count)) -#define PKT3(op, count) (PKT_TYPE_S(3) | PKT3_IT_OPCODE_S(op) | PKT_COUNT_S(count)) +#define PKT3(op, count, predicate) (PKT_TYPE_S(3) | PKT3_IT_OPCODE_S(op) | PKT_COUNT_S(count) | PKT3_PRED_S(predicate)) /* Registers */ #define R_0280A0_CB_COLOR0_INFO 0x0280A0 diff --git a/src/gallium/winsys/r600/drm/radeon_bo.c b/src/gallium/winsys/r600/drm/radeon_bo.c index 557cfb9597..ba02a3c2bd 100644 --- a/src/gallium/winsys/r600/drm/radeon_bo.c +++ b/src/gallium/winsys/r600/drm/radeon_bo.c @@ -33,6 +33,25 @@ #include "xf86drm.h" #include "radeon_drm.h" +#ifdef ANDROID + +extern void* __mmap2(void*, size_t, int, int, int, size_t); + +#define MMAP2_SHIFT 12 +static void* android_mmap2(void *addr, size_t size, int prot, int flags, int fd, unsigned long long offset) +{ + if ( offset & ((1UL << MMAP2_SHIFT)-1) ) { + errno = EINVAL; + return MAP_FAILED; + } + + return __mmap2(addr, size, prot, flags, fd, (size_t)(offset >> MMAP2_SHIFT)); +} + +#define mmap(addr, size, prot, flags, fd, offset) android_mmap2(addr, size, prot, flags, fd, offset) + +#endif /* ANDROID */ + static int radeon_bo_fixed_map(struct radeon *radeon, struct radeon_bo *bo) { struct drm_radeon_gem_mmap args; @@ -98,7 +117,7 @@ struct radeon_bo *radeon_bo(struct radeon *radeon, unsigned handle, bo->size = open_arg.size; bo->shared = TRUE; } else { - struct drm_radeon_gem_create args; + struct drm_radeon_gem_create args = {}; args.size = size; args.alignment = alignment; @@ -156,7 +175,7 @@ int radeon_bo_wait(struct radeon *radeon, struct radeon_bo *bo) if (!bo->shared) { if (!bo->fence) return 0; - if (bo->fence <= *bo->ctx->cfence) { + if (bo->fence <= *radeon->cfence) { LIST_DELINIT(&bo->fencedlist); bo->fence = 0; return 0; @@ -181,7 +200,7 @@ int radeon_bo_busy(struct radeon *radeon, struct radeon_bo *bo, uint32_t *domain if (!bo->shared) { if (!bo->fence) return 0; - if (bo->fence <= *bo->ctx->cfence) { + if (bo->fence <= *radeon->cfence) { LIST_DELINIT(&bo->fencedlist); bo->fence = 0; return 0; @@ -204,15 +223,15 @@ int radeon_bo_get_tiling_flags(struct radeon *radeon, uint32_t *tiling_flags, uint32_t *pitch) { - struct drm_radeon_gem_get_tiling args; + struct drm_radeon_gem_get_tiling args = {}; int ret; - + args.handle = bo->handle; ret = drmCommandWriteRead(radeon->fd, DRM_RADEON_GEM_GET_TILING, &args, sizeof(args)); if (ret) return ret; - + *tiling_flags = args.tiling_flags; *pitch = args.pitch; return ret; diff --git a/src/gallium/winsys/r600/drm/radeon_bo_pb.c b/src/gallium/winsys/r600/drm/radeon_bo_pb.c deleted file mode 100644 index 4bd3ae3ca1..0000000000 --- a/src/gallium/winsys/r600/drm/radeon_bo_pb.c +++ /dev/null @@ -1,260 +0,0 @@ -/* - * Copyright 2010 Dave Airlie - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * on the rights to use, copy, modify, merge, publish, distribute, sub - * license, and/or sell copies of the Software, and to permit persons to whom - * the Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * 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 - * THE AUTHOR(S) AND/OR THEIR 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. - * - * Authors: - * Dave Airlie - */ -#include <util/u_inlines.h> -#include <util/u_memory.h> -#include <util/u_double_list.h> -#include <pipebuffer/pb_buffer.h> -#include <pipebuffer/pb_bufmgr.h> -#include "r600_priv.h" - -struct radeon_bo_pb { - struct pb_buffer b; - struct radeon_bo *bo; - - struct radeon_bo_pbmgr *mgr; -}; - -extern const struct pb_vtbl radeon_bo_pb_vtbl; - -static INLINE struct radeon_bo_pb *radeon_bo_pb(struct pb_buffer *buf) -{ - assert(buf); - assert(buf->vtbl == &radeon_bo_pb_vtbl); - return (struct radeon_bo_pb *)buf; -} - -struct radeon_bo_pbmgr { - struct pb_manager b; - struct radeon *radeon; -}; - -static INLINE struct radeon_bo_pbmgr *radeon_bo_pbmgr(struct pb_manager *mgr) -{ - assert(mgr); - return (struct radeon_bo_pbmgr *)mgr; -} - -static void radeon_bo_pb_destroy(struct pb_buffer *_buf) -{ - struct radeon_bo_pb *buf = radeon_bo_pb(_buf); - - /* If this buffer is on the list of buffers to unmap, - * do the unmapping now. - */ - radeon_bo_unmap(buf->mgr->radeon, buf->bo); - radeon_bo_reference(buf->mgr->radeon, &buf->bo, NULL); - FREE(buf); -} - -static void * -radeon_bo_pb_map_internal(struct pb_buffer *_buf, - unsigned flags, void *ctx) -{ - struct radeon_bo_pb *buf = radeon_bo_pb(_buf); - struct pipe_context *pctx = ctx; - - if (flags & PB_USAGE_UNSYNCHRONIZED) { - if (radeon_bo_map(buf->mgr->radeon, buf->bo)) { - return NULL; - } - return buf->bo->data; - } - - if (p_atomic_read(&buf->bo->reference.count) > 1) { - if (flags & PB_USAGE_DONTBLOCK) { - return NULL; - } - if (ctx) { - pctx->flush(pctx, 0, NULL); - } - } - - if (flags & PB_USAGE_DONTBLOCK) { - uint32_t domain; - if (radeon_bo_busy(buf->mgr->radeon, buf->bo, &domain)) - return NULL; - if (radeon_bo_map(buf->mgr->radeon, buf->bo)) { - return NULL; - } - goto out; - } - - if (radeon_bo_map(buf->mgr->radeon, buf->bo)) { - return NULL; - } - if (radeon_bo_wait(buf->mgr->radeon, buf->bo)) { - radeon_bo_unmap(buf->mgr->radeon, buf->bo); - return NULL; - } -out: - return buf->bo->data; -} - -static void radeon_bo_pb_unmap_internal(struct pb_buffer *_buf) -{ -} - -static void -radeon_bo_pb_get_base_buffer(struct pb_buffer *buf, - struct pb_buffer **base_buf, - unsigned *offset) -{ - *base_buf = buf; - *offset = 0; -} - -static enum pipe_error -radeon_bo_pb_validate(struct pb_buffer *_buf, - struct pb_validate *vl, - unsigned flags) -{ - /* Always pinned */ - return PIPE_OK; -} - -static void -radeon_bo_pb_fence(struct pb_buffer *buf, - struct pipe_fence_handle *fence) -{ -} - -const struct pb_vtbl radeon_bo_pb_vtbl = { - radeon_bo_pb_destroy, - radeon_bo_pb_map_internal, - radeon_bo_pb_unmap_internal, - radeon_bo_pb_validate, - radeon_bo_pb_fence, - radeon_bo_pb_get_base_buffer, -}; - -struct pb_buffer * -radeon_bo_pb_create_buffer_from_handle(struct pb_manager *_mgr, - uint32_t handle) -{ - struct radeon_bo_pbmgr *mgr = radeon_bo_pbmgr(_mgr); - struct radeon *radeon = mgr->radeon; - struct radeon_bo_pb *bo; - struct radeon_bo *hw_bo; - - hw_bo = radeon_bo(radeon, handle, 0, 0); - if (hw_bo == NULL) - return NULL; - - bo = CALLOC_STRUCT(radeon_bo_pb); - if (!bo) { - radeon_bo_reference(radeon, &hw_bo, NULL); - return NULL; - } - - pipe_reference_init(&bo->b.base.reference, 1); - bo->b.base.alignment = 0; - bo->b.base.usage = PB_USAGE_GPU_WRITE | PB_USAGE_GPU_READ; - bo->b.base.size = hw_bo->size; - bo->b.vtbl = &radeon_bo_pb_vtbl; - bo->mgr = mgr; - - bo->bo = hw_bo; - - return &bo->b; -} - -static struct pb_buffer * -radeon_bo_pb_create_buffer(struct pb_manager *_mgr, - pb_size size, - const struct pb_desc *desc) -{ - struct radeon_bo_pbmgr *mgr = radeon_bo_pbmgr(_mgr); - struct radeon *radeon = mgr->radeon; - struct radeon_bo_pb *bo; - - bo = CALLOC_STRUCT(radeon_bo_pb); - if (!bo) - goto error1; - - pipe_reference_init(&bo->b.base.reference, 1); - bo->b.base.alignment = desc->alignment; - bo->b.base.usage = desc->usage; - bo->b.base.size = size; - bo->b.vtbl = &radeon_bo_pb_vtbl; - bo->mgr = mgr; - - bo->bo = radeon_bo(radeon, 0, size, desc->alignment); - if (bo->bo == NULL) - goto error2; - return &bo->b; - -error2: - FREE(bo); -error1: - return NULL; -} - -static void -radeon_bo_pbmgr_flush(struct pb_manager *mgr) -{ - /* NOP */ -} - -static void -radeon_bo_pbmgr_destroy(struct pb_manager *_mgr) -{ - struct radeon_bo_pbmgr *mgr = radeon_bo_pbmgr(_mgr); - FREE(mgr); -} - -struct pb_manager *radeon_bo_pbmgr_create(struct radeon *radeon) -{ - struct radeon_bo_pbmgr *mgr; - - mgr = CALLOC_STRUCT(radeon_bo_pbmgr); - if (!mgr) - return NULL; - - mgr->b.destroy = radeon_bo_pbmgr_destroy; - mgr->b.create_buffer = radeon_bo_pb_create_buffer; - mgr->b.flush = radeon_bo_pbmgr_flush; - - mgr->radeon = radeon; - return &mgr->b; -} - -struct radeon_bo *radeon_bo_pb_get_bo(struct pb_buffer *_buf) -{ - struct radeon_bo_pb *buf; - if (_buf->vtbl == &radeon_bo_pb_vtbl) { - buf = radeon_bo_pb(_buf); - return buf->bo; - } else { - struct pb_buffer *base_buf; - pb_size offset; - pb_get_base_buffer(_buf, &base_buf, &offset); - if (base_buf->vtbl == &radeon_bo_pb_vtbl) { - buf = radeon_bo_pb(base_buf); - return buf->bo; - } - } - return NULL; -} diff --git a/src/gallium/winsys/r600/drm/radeon_pciid.c b/src/gallium/winsys/r600/drm/radeon_pciid.c index 92560a488a..f19956931d 100644 --- a/src/gallium/winsys/r600/drm/radeon_pciid.c +++ b/src/gallium/winsys/r600/drm/radeon_pciid.c @@ -32,7 +32,7 @@ struct pci_id { unsigned family; }; -struct pci_id radeon_pci_id[] = { +static const struct pci_id radeon_pci_id[] = { {0x1002, 0x3150, CHIP_RV380}, {0x1002, 0x3152, CHIP_RV380}, {0x1002, 0x3154, CHIP_RV380}, @@ -40,29 +40,29 @@ struct pci_id radeon_pci_id[] = { {0x1002, 0x3E54, CHIP_RV380}, {0x1002, 0x4136, CHIP_RS100}, {0x1002, 0x4137, CHIP_RS200}, - {0x1002, 0x4144, CHIP_R300}, - {0x1002, 0x4145, CHIP_R300}, - {0x1002, 0x4146, CHIP_R300}, - {0x1002, 0x4147, CHIP_R300}, - {0x1002, 0x4148, CHIP_R350}, - {0x1002, 0x4149, CHIP_R350}, - {0x1002, 0x414A, CHIP_R350}, - {0x1002, 0x414B, CHIP_R350}, - {0x1002, 0x4150, CHIP_RV350}, - {0x1002, 0x4151, CHIP_RV350}, - {0x1002, 0x4152, CHIP_RV350}, - {0x1002, 0x4153, CHIP_RV350}, - {0x1002, 0x4154, CHIP_RV350}, - {0x1002, 0x4155, CHIP_RV350}, - {0x1002, 0x4156, CHIP_RV350}, + {0x1002, 0x4144, CHIP_R300}, + {0x1002, 0x4145, CHIP_R300}, + {0x1002, 0x4146, CHIP_R300}, + {0x1002, 0x4147, CHIP_R300}, + {0x1002, 0x4148, CHIP_R350}, + {0x1002, 0x4149, CHIP_R350}, + {0x1002, 0x414A, CHIP_R350}, + {0x1002, 0x414B, CHIP_R350}, + {0x1002, 0x4150, CHIP_RV350}, + {0x1002, 0x4151, CHIP_RV350}, + {0x1002, 0x4152, CHIP_RV350}, + {0x1002, 0x4153, CHIP_RV350}, + {0x1002, 0x4154, CHIP_RV350}, + {0x1002, 0x4155, CHIP_RV350}, + {0x1002, 0x4156, CHIP_RV350}, {0x1002, 0x4237, CHIP_RS200}, - {0x1002, 0x4242, CHIP_R200}, - {0x1002, 0x4243, CHIP_R200}, + {0x1002, 0x4242, CHIP_R200}, + {0x1002, 0x4243, CHIP_R200}, {0x1002, 0x4336, CHIP_RS100}, {0x1002, 0x4337, CHIP_RS200}, {0x1002, 0x4437, CHIP_RS200}, - {0x1002, 0x4966, CHIP_RV250}, - {0x1002, 0x4967, CHIP_RV250}, + {0x1002, 0x4966, CHIP_RV250}, + {0x1002, 0x4967, CHIP_RV250}, {0x1002, 0x4A48, CHIP_R420}, {0x1002, 0x4A49, CHIP_R420}, {0x1002, 0x4A4A, CHIP_R420}, @@ -85,14 +85,14 @@ struct pci_id radeon_pci_id[] = { {0x1002, 0x4C64, CHIP_RV250}, {0x1002, 0x4C66, CHIP_RV250}, {0x1002, 0x4C67, CHIP_RV250}, - {0x1002, 0x4E44, CHIP_R300}, - {0x1002, 0x4E45, CHIP_R300}, - {0x1002, 0x4E46, CHIP_R300}, - {0x1002, 0x4E47, CHIP_R300}, - {0x1002, 0x4E48, CHIP_R350}, - {0x1002, 0x4E49, CHIP_R350}, - {0x1002, 0x4E4A, CHIP_R350}, - {0x1002, 0x4E4B, CHIP_R350}, + {0x1002, 0x4E44, CHIP_R300}, + {0x1002, 0x4E45, CHIP_R300}, + {0x1002, 0x4E46, CHIP_R300}, + {0x1002, 0x4E47, CHIP_R300}, + {0x1002, 0x4E48, CHIP_R350}, + {0x1002, 0x4E49, CHIP_R350}, + {0x1002, 0x4E4A, CHIP_R350}, + {0x1002, 0x4E4B, CHIP_R350}, {0x1002, 0x4E50, CHIP_RV350}, {0x1002, 0x4E51, CHIP_RV350}, {0x1002, 0x4E52, CHIP_RV350}, @@ -103,13 +103,13 @@ struct pci_id radeon_pci_id[] = { {0x1002, 0x5145, CHIP_R100}, {0x1002, 0x5146, CHIP_R100}, {0x1002, 0x5147, CHIP_R100}, - {0x1002, 0x5148, CHIP_R200}, - {0x1002, 0x514C, CHIP_R200}, - {0x1002, 0x514D, CHIP_R200}, - {0x1002, 0x5157, CHIP_RV200}, - {0x1002, 0x5158, CHIP_RV200}, - {0x1002, 0x5159, CHIP_RV100}, - {0x1002, 0x515A, CHIP_RV100}, + {0x1002, 0x5148, CHIP_R200}, + {0x1002, 0x514C, CHIP_R200}, + {0x1002, 0x514D, CHIP_R200}, + {0x1002, 0x5157, CHIP_RV200}, + {0x1002, 0x5158, CHIP_RV200}, + {0x1002, 0x5159, CHIP_RV100}, + {0x1002, 0x515A, CHIP_RV100}, {0x1002, 0x515E, CHIP_RV100}, {0x1002, 0x5460, CHIP_RV380}, {0x1002, 0x5462, CHIP_RV380}, @@ -138,10 +138,10 @@ struct pci_id radeon_pci_id[] = { {0x1002, 0x5955, CHIP_RS480}, {0x1002, 0x5974, CHIP_RS480}, {0x1002, 0x5975, CHIP_RS480}, - {0x1002, 0x5960, CHIP_RV280}, - {0x1002, 0x5961, CHIP_RV280}, - {0x1002, 0x5962, CHIP_RV280}, - {0x1002, 0x5964, CHIP_RV280}, + {0x1002, 0x5960, CHIP_RV280}, + {0x1002, 0x5961, CHIP_RV280}, + {0x1002, 0x5962, CHIP_RV280}, + {0x1002, 0x5964, CHIP_RV280}, {0x1002, 0x5965, CHIP_RV280}, {0x1002, 0x5969, CHIP_RV100}, {0x1002, 0x5a41, CHIP_RS400}, @@ -445,6 +445,42 @@ struct pci_id radeon_pci_id[] = { {0x1002, 0x9803, CHIP_PALM}, {0x1002, 0x9804, CHIP_PALM}, {0x1002, 0x9805, CHIP_PALM}, + {0x1002, 0x6720, CHIP_BARTS}, + {0x1002, 0x6721, CHIP_BARTS}, + {0x1002, 0x6722, CHIP_BARTS}, + {0x1002, 0x6723, CHIP_BARTS}, + {0x1002, 0x6724, CHIP_BARTS}, + {0x1002, 0x6725, CHIP_BARTS}, + {0x1002, 0x6726, CHIP_BARTS}, + {0x1002, 0x6727, CHIP_BARTS}, + {0x1002, 0x6728, CHIP_BARTS}, + {0x1002, 0x6729, CHIP_BARTS}, + {0x1002, 0x6738, CHIP_BARTS}, + {0x1002, 0x6739, CHIP_BARTS}, + {0x1002, 0x6740, CHIP_TURKS}, + {0x1002, 0x6741, CHIP_TURKS}, + {0x1002, 0x6742, CHIP_TURKS}, + {0x1002, 0x6743, CHIP_TURKS}, + {0x1002, 0x6744, CHIP_TURKS}, + {0x1002, 0x6745, CHIP_TURKS}, + {0x1002, 0x6746, CHIP_TURKS}, + {0x1002, 0x6747, CHIP_TURKS}, + {0x1002, 0x6748, CHIP_TURKS}, + {0x1002, 0x6749, CHIP_TURKS}, + {0x1002, 0x6750, CHIP_TURKS}, + {0x1002, 0x6758, CHIP_TURKS}, + {0x1002, 0x6759, CHIP_TURKS}, + {0x1002, 0x6760, CHIP_CAICOS}, + {0x1002, 0x6761, CHIP_CAICOS}, + {0x1002, 0x6762, CHIP_CAICOS}, + {0x1002, 0x6763, CHIP_CAICOS}, + {0x1002, 0x6764, CHIP_CAICOS}, + {0x1002, 0x6765, CHIP_CAICOS}, + {0x1002, 0x6766, CHIP_CAICOS}, + {0x1002, 0x6767, CHIP_CAICOS}, + {0x1002, 0x6768, CHIP_CAICOS}, + {0x1002, 0x6770, CHIP_CAICOS}, + {0x1002, 0x6779, CHIP_CAICOS}, {0, 0}, }; |