-/* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */
-
/*
* Copyright (C) 2012 Rob Clark <robclark@freedesktop.org>
*
#include "util/u_inlines.h"
#include "util/u_format.h"
#include "util/u_format_s3tc.h"
+#include "util/u_screen.h"
#include "util/u_string.h"
#include "util/u_debug.h"
#include "util/os_time.h"
-#include <stdio.h>
+#include "drm-uapi/drm_fourcc.h"
#include <errno.h>
+#include <stdio.h>
#include <stdlib.h>
+#include <sys/sysinfo.h>
#include "freedreno_screen.h"
#include "freedreno_resource.h"
#include "a3xx/fd3_screen.h"
#include "a4xx/fd4_screen.h"
#include "a5xx/fd5_screen.h"
+#include "a6xx/fd6_screen.h"
+
#include "ir3/ir3_nir.h"
+#include "a2xx/ir2.h"
/* XXX this should go away */
#include "state_tracker/drm_driver.h"
static const struct debug_named_value debug_options[] = {
{"msgs", FD_DBG_MSGS, "Print debug messages"},
- {"disasm", FD_DBG_DISASM, "Dump TGSI and adreno shader disassembly"},
+ {"disasm", FD_DBG_DISASM, "Dump TGSI and adreno shader disassembly (a2xx only, see IR3_SHADER_DEBUG)"},
{"dclear", FD_DBG_DCLEAR, "Mark all state dirty after clear"},
{"ddraw", FD_DBG_DDRAW, "Mark all state dirty after draw"},
{"noscis", FD_DBG_NOSCIS, "Disable scissor optimization"},
{"nobypass", FD_DBG_NOBYPASS, "Disable GMEM bypass"},
{"fraghalf", FD_DBG_FRAGHALF, "Use half-precision in fragment shader"},
{"nobin", FD_DBG_NOBIN, "Disable hw binning"},
- {"optmsgs", FD_DBG_OPTMSGS,"Enable optimizer debug messages"},
{"glsl120", FD_DBG_GLSL120,"Temporary flag to force GLSL 1.20 (rather than 1.30) on a3xx+"},
{"shaderdb", FD_DBG_SHADERDB, "Enable shaderdb output"},
{"flush", FD_DBG_FLUSH, "Force flush after every draw"},
{"nogrow", FD_DBG_NOGROW, "Disable \"growable\" cmdstream buffers, even if kernel supports it"},
{"lrz", FD_DBG_LRZ, "Enable experimental LRZ support (a5xx+)"},
{"noindirect",FD_DBG_NOINDR, "Disable hw indirect draws (emulate on CPU)"},
+ {"noblit", FD_DBG_NOBLIT, "Disable blitter (fallback to generic blit path)"},
+ {"hiprio", FD_DBG_HIPRIO, "Force high-priority context"},
+ {"ttile", FD_DBG_TTILE, "Enable texture tiling (a5xx)"},
+ {"perfcntrs", FD_DBG_PERFC, "Expose performance counters"},
+ {"noubwc", FD_DBG_NOUBWC, "Disable UBWC for all internal buffers"},
DEBUG_NAMED_VALUE_END
};
fd_screen_get_name(struct pipe_screen *pscreen)
{
static char buffer[128];
- util_snprintf(buffer, sizeof(buffer), "FD%03d",
+ snprintf(buffer, sizeof(buffer), "FD%03d",
fd_screen(pscreen)->device_id);
return buffer;
}
if (screen->dev)
fd_device_del(screen->dev);
+ if (screen->ro)
+ FREE(screen->ro);
+
fd_bc_fini(&screen->batch_cache);
slab_destroy_parent(&screen->transfer_pool);
ralloc_free(screen->compiler);
+ free(screen->perfcntr_queries);
free(screen);
}
/* Supported features (boolean caps). */
case PIPE_CAP_NPOT_TEXTURES:
case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
- case PIPE_CAP_TWO_SIDED_STENCIL:
case PIPE_CAP_ANISOTROPIC_FILTER:
case PIPE_CAP_POINT_SPRITE:
- case PIPE_CAP_TEXTURE_SHADOW_MAP:
case PIPE_CAP_BLEND_EQUATION_SEPARATE:
case PIPE_CAP_TEXTURE_SWIZZLE:
case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
case PIPE_CAP_INVALIDATE_BUFFER:
return 1;
+ case PIPE_CAP_PACKED_UNIFORMS:
+ return !is_a2xx(screen);
+
+ case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR:
+ case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
+ return screen->has_robustness;
+
case PIPE_CAP_VERTEXID_NOBASE:
return is_a3xx(screen) || is_a4xx(screen);
- case PIPE_CAP_USER_CONSTANT_BUFFERS:
- return is_a4xx(screen) ? 0 : 1;
-
case PIPE_CAP_COMPUTE:
return has_compute(screen);
- case PIPE_CAP_SHADER_STENCIL_EXPORT:
- case PIPE_CAP_TGSI_TEXCOORD:
case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
- case PIPE_CAP_TEXTURE_MULTISAMPLE:
- case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
- case PIPE_CAP_QUERY_MEMORY_INFO:
case PIPE_CAP_PCI_GROUP:
case PIPE_CAP_PCI_BUS:
case PIPE_CAP_PCI_DEVICE:
case PIPE_CAP_PCI_FUNCTION:
+ case PIPE_CAP_DEPTH_CLIP_DISABLE_SEPARATE:
return 0;
- case PIPE_CAP_SM3:
+ case PIPE_CAP_FRAGMENT_SHADER_TEXTURE_LOD:
+ case PIPE_CAP_FRAGMENT_SHADER_DERIVATIVES:
+ case PIPE_CAP_VERTEX_SHADER_SATURATE:
case PIPE_CAP_PRIMITIVE_RESTART:
case PIPE_CAP_TGSI_INSTANCEID:
case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
case PIPE_CAP_CONDITIONAL_RENDER:
case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
- case PIPE_CAP_FAKE_SW_MSAA:
case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
case PIPE_CAP_CLIP_HALFZ:
- return is_a3xx(screen) || is_a4xx(screen) || is_a5xx(screen);
+ return is_a3xx(screen) || is_a4xx(screen) || is_a5xx(screen) || is_a6xx(screen);
+
+ case PIPE_CAP_FAKE_SW_MSAA:
+ return !fd_screen_get_param(pscreen, PIPE_CAP_TEXTURE_MULTISAMPLE);
+
+ case PIPE_CAP_TEXTURE_MULTISAMPLE:
+ return is_a5xx(screen) || is_a6xx(screen);
+
+ case PIPE_CAP_SURFACE_SAMPLE_COUNT:
+ return is_a6xx(screen);
case PIPE_CAP_DEPTH_CLIP_DISABLE:
return is_a3xx(screen) || is_a4xx(screen);
case PIPE_CAP_POLYGON_OFFSET_CLAMP:
- return is_a5xx(screen);
+ return is_a5xx(screen) || is_a6xx(screen);
- case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY:
- return 0;
case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
if (is_a3xx(screen)) return 16;
if (is_a4xx(screen)) return 32;
if (is_a5xx(screen)) return 32;
+ if (is_a6xx(screen)) return 64;
return 0;
case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
/* We could possibly emulate more by pretending 2d/rect textures and
if (is_a3xx(screen)) return 8192;
if (is_a4xx(screen)) return 16384;
if (is_a5xx(screen)) return 16384;
+ if (is_a6xx(screen)) return 1 << 27;
return 0;
case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
case PIPE_CAP_CUBE_MAP_ARRAY:
case PIPE_CAP_SAMPLER_VIEW_TARGET:
case PIPE_CAP_TEXTURE_QUERY_LOD:
- return is_a4xx(screen) || is_a5xx(screen);
+ return is_a4xx(screen) || is_a5xx(screen) || is_a6xx(screen);
case PIPE_CAP_START_INSTANCE:
/* Note that a5xx can do this, it just can't (at least with
return 64;
case PIPE_CAP_GLSL_FEATURE_LEVEL:
+ case PIPE_CAP_GLSL_FEATURE_LEVEL_COMPATIBILITY:
if (glsl120)
return 120;
return is_ir3(screen) ? 140 : 120;
+ case PIPE_CAP_ESSL_FEATURE_LEVEL:
+ /* we can probably enable 320 for a5xx too, but need to test: */
+ if (is_a6xx(screen)) return 320;
+ if (is_a5xx(screen)) return 310;
+ if (is_ir3(screen)) return 300;
+ return 120;
+
case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
- if (is_a5xx(screen))
- return 4;
+ if (is_a6xx(screen)) return 64;
+ if (is_a5xx(screen)) return 4;
return 0;
case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
- if (is_a4xx(screen) || is_a5xx(screen))
+ if (is_a4xx(screen) || is_a5xx(screen) || is_a6xx(screen))
return 4;
return 0;
- /* Unsupported features. */
- case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
- case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
- case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
- case PIPE_CAP_USER_VERTEX_BUFFERS:
- case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
- case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
- case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
- case PIPE_CAP_TEXTURE_GATHER_SM5:
- case PIPE_CAP_SAMPLE_SHADING:
- case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
- case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
- case PIPE_CAP_MULTI_DRAW_INDIRECT:
- case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
- case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
- case PIPE_CAP_MULTISAMPLE_Z_RESOLVE:
- case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
- case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
- case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
- case PIPE_CAP_DEPTH_BOUNDS_TEST:
- case PIPE_CAP_TGSI_TXQS:
/* TODO if we need this, do it in nir/ir3 backend to avoid breaking precompile: */
case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
- case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
- case PIPE_CAP_CLEAR_TEXTURE:
- case PIPE_CAP_DRAW_PARAMETERS:
- case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
- case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL:
- case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL:
- case PIPE_CAP_GENERATE_MIPMAP:
- case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS:
- case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR:
- case PIPE_CAP_CULL_DISTANCE:
- case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES:
- case PIPE_CAP_TGSI_VOTE:
- case PIPE_CAP_MAX_WINDOW_RECTANGLES:
- case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED:
- case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS:
- case PIPE_CAP_TGSI_ARRAY_COMPONENTS:
- case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
- case PIPE_CAP_TGSI_FS_FBFETCH:
- case PIPE_CAP_TGSI_MUL_ZERO_WINS:
- case PIPE_CAP_DOUBLES:
- case PIPE_CAP_INT64:
- case PIPE_CAP_INT64_DIVMOD:
- case PIPE_CAP_TGSI_TEX_TXF_LZ:
- case PIPE_CAP_TGSI_CLOCK:
- case PIPE_CAP_POLYGON_MODE_FILL_RECTANGLE:
- case PIPE_CAP_SPARSE_BUFFER_PAGE_SIZE:
- case PIPE_CAP_TGSI_BALLOT:
- case PIPE_CAP_TGSI_TES_LAYER_VIEWPORT:
- case PIPE_CAP_CAN_BIND_CONST_BUFFER_AS_VERTEX:
+ return 0;
+
+ case PIPE_CAP_FBFETCH:
+ if (fd_device_version(screen->dev) >= FD_VERSION_GMEM_BASE &&
+ is_a6xx(screen))
+ return 1;
+ return 0;
+ case PIPE_CAP_SAMPLE_SHADING:
+ if (is_a6xx(screen)) return 1;
+ return 0;
+
case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION:
- case PIPE_CAP_POST_DEPTH_COVERAGE:
- case PIPE_CAP_BINDLESS_TEXTURE:
- case PIPE_CAP_NIR_SAMPLERS_AS_DEREF:
- case PIPE_CAP_QUERY_SO_OVERFLOW:
- case PIPE_CAP_MEMOBJ:
- case PIPE_CAP_TGSI_ANY_REG_AS_ADDRESS:
- case PIPE_CAP_TILE_RASTER_ORDER:
- case PIPE_CAP_MAX_COMBINED_SHADER_OUTPUT_RESOURCES:
- case PIPE_CAP_SIGNED_VERTEX_BUFFER_OFFSET:
return 0;
+ case PIPE_CAP_CONTEXT_PRIORITY_MASK:
+ return screen->priority_mask;
+
case PIPE_CAP_DRAW_INDIRECT:
- if (is_a4xx(screen) || is_a5xx(screen))
+ if (is_a4xx(screen) || is_a5xx(screen) || is_a6xx(screen))
return 1;
return 0;
case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT:
- if (is_a4xx(screen) || is_a5xx(screen))
+ if (is_a4xx(screen) || is_a5xx(screen) || is_a6xx(screen))
return 1;
return 0;
case PIPE_CAP_MAX_VIEWPORTS:
return 1;
+ case PIPE_CAP_MAX_VARYINGS:
+ return 16;
+
case PIPE_CAP_SHAREABLE_SHADERS:
case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
/* manage the variants for these ourself, to avoid breaking precompile: */
return 0;
case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS:
+ case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL:
if (is_ir3(screen))
return 1;
return 0;
return 16 * 4; /* should only be shader out limit? */
return 0;
- /* Geometry shader output, unsupported. */
- case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
- case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
- case PIPE_CAP_MAX_VERTEX_STREAMS:
- return 0;
-
- case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
- return 2048;
-
/* Texturing. */
- case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
+ case PIPE_CAP_MAX_TEXTURE_2D_SIZE:
+ return 1 << (MAX_MIP_LEVELS - 1);
case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
return MAX_MIP_LEVELS;
case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
return 11;
case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
- return (is_a3xx(screen) || is_a4xx(screen) || is_a5xx(screen)) ? 256 : 0;
+ return (is_a3xx(screen) || is_a4xx(screen) || is_a5xx(screen) || is_a6xx(screen)) ? 256 : 0;
/* Render targets. */
case PIPE_CAP_MAX_RENDER_TARGETS:
return is_a3xx(screen) ? 1 : 0;
/* Queries. */
- case PIPE_CAP_QUERY_BUFFER_OBJECT:
- return 0;
case PIPE_CAP_OCCLUSION_QUERY:
- return is_a3xx(screen) || is_a4xx(screen) || is_a5xx(screen);
+ return is_a3xx(screen) || is_a4xx(screen) || is_a5xx(screen) || is_a6xx(screen);
case PIPE_CAP_QUERY_TIMESTAMP:
case PIPE_CAP_QUERY_TIME_ELAPSED:
/* only a4xx, requires new enough kernel so we know max_freq: */
- return (screen->max_freq > 0) && (is_a4xx(screen) || is_a5xx(screen));
-
- case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
- case PIPE_CAP_MIN_TEXEL_OFFSET:
- return -8;
-
- case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
- case PIPE_CAP_MAX_TEXEL_OFFSET:
- return 7;
-
- case PIPE_CAP_ENDIANNESS:
- return PIPE_ENDIAN_LITTLE;
-
- case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
- return 64;
+ return (screen->max_freq > 0) && (is_a4xx(screen) || is_a5xx(screen) || is_a6xx(screen));
case PIPE_CAP_VENDOR_ID:
return 0x5143;
return 1;
case PIPE_CAP_NATIVE_FENCE_FD:
return fd_device_version(screen->dev) >= FD_VERSION_FENCE_FD;
+ default:
+ return u_pipe_screen_get_param_defaults(pscreen, param);
}
- debug_printf("unknown param %d\n", param);
- return 0;
}
static float
return 16.0f;
case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
return 15.0f;
- case PIPE_CAPF_GUARD_BAND_LEFT:
- case PIPE_CAPF_GUARD_BAND_TOP:
- case PIPE_CAPF_GUARD_BAND_RIGHT:
- case PIPE_CAPF_GUARD_BAND_BOTTOM:
+ case PIPE_CAPF_MIN_CONSERVATIVE_RASTER_DILATE:
+ case PIPE_CAPF_MAX_CONSERVATIVE_RASTER_DILATE:
+ case PIPE_CAPF_CONSERVATIVE_RASTER_DILATE_GRANULARITY:
return 0.0f;
}
debug_printf("unknown paramf %d\n", param);
* split between VS and FS. Use lower limit of 256 to
* avoid getting into impossible situations:
*/
- return ((is_a3xx(screen) || is_a4xx(screen) || is_a5xx(screen)) ? 4096 : 64) * sizeof(float[4]);
+ return ((is_a3xx(screen) || is_a4xx(screen) || is_a5xx(screen) || is_a6xx(screen)) ? 4096 : 64) * sizeof(float[4]);
case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
return is_ir3(screen) ? 16 : 1;
case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
case PIPE_SHADER_CAP_TGSI_LDEXP_SUPPORTED:
case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
+ case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS:
+ case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD:
+ case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS:
+ case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS:
return 0;
case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
return 1;
case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
return 16;
case PIPE_SHADER_CAP_PREFERRED_IR:
- if (is_ir3(screen))
- return PIPE_SHADER_IR_NIR;
- return PIPE_SHADER_IR_TGSI;
+ return PIPE_SHADER_IR_NIR;
case PIPE_SHADER_CAP_SUPPORTED_IRS:
- if (is_ir3(screen)) {
- return (1 << PIPE_SHADER_IR_NIR) | (1 << PIPE_SHADER_IR_TGSI);
- } else {
- return (1 << PIPE_SHADER_IR_TGSI);
- }
- return 0;
+ return (1 << PIPE_SHADER_IR_NIR) | (1 << PIPE_SHADER_IR_TGSI);
case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
return 32;
- case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD:
- case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS:
- case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS:
- case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS:
- return 0;
+ case PIPE_SHADER_CAP_SCALAR_ISA:
+ return is_ir3(screen) ? 1 : 0;
case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
- if (is_a5xx(screen)) {
+ if (is_a5xx(screen) || is_a6xx(screen)) {
/* a5xx (and a4xx for that matter) has one state-block
* for compute-shader SSBO's and another that is shared
* by VS/HS/DS/GS/FS.. so to simplify things for now
if (!has_compute(screen))
return 0;
+#define RET(x) do { \
+ if (ret) \
+ memcpy(ret, x, sizeof(x)); \
+ return sizeof(x); \
+} while (0)
+
switch (param) {
case PIPE_COMPUTE_CAP_ADDRESS_BITS:
- if (ret) {
- uint32_t *address_bits = ret;
- address_bits[0] = 32;
-
- if (is_a5xx(screen))
- address_bits[0] = 64;
- }
- return 1 * sizeof(uint32_t);
+// don't expose 64b pointer support yet, until ir3 supports 64b
+// math, otherwise spir64 target is used and we get 64b pointer
+// calculations that we can't do yet
+// if (is_a5xx(screen))
+// RET((uint32_t []){ 64 });
+ RET((uint32_t []){ 32 });
case PIPE_COMPUTE_CAP_IR_TARGET:
if (ret)
- sprintf(ret, ir);
+ sprintf(ret, "%s", ir);
return strlen(ir) * sizeof(char);
case PIPE_COMPUTE_CAP_GRID_DIMENSION:
- if (ret) {
- uint64_t *grid_dimension = ret;
- grid_dimension[0] = 3;
- }
- return 1 * sizeof(uint64_t);
+ RET((uint64_t []) { 3 });
case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
- if (ret) {
- uint64_t *grid_size = ret;
- grid_size[0] = 65535;
- grid_size[1] = 65535;
- grid_size[2] = 65535;
- }
- return 3 * sizeof(uint64_t) ;
+ RET(((uint64_t []) { 65535, 65535, 65535 }));
case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
- if (ret) {
- uint64_t *block_size = ret;
- block_size[0] = 1024;
- block_size[1] = 1024;
- block_size[2] = 64;
- }
- return 3 * sizeof(uint64_t) ;
+ RET(((uint64_t []) { 1024, 1024, 64 }));
case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
- if (ret) {
- uint64_t *max_threads_per_block = ret;
- *max_threads_per_block = 1024;
- }
- return sizeof(uint64_t);
+ RET((uint64_t []) { 1024 });
case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE:
+ RET((uint64_t []) { screen->ram_size });
+
case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE:
- if (ret) {
- uint64_t *local_size = ret;
- *local_size = 32768;
- }
- return sizeof(uint64_t);
+ RET((uint64_t []) { 32768 });
+
case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE:
case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE:
- break;
+ RET((uint64_t []) { 4096 });
+
case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE:
- if (ret) {
- uint64_t *max = ret;
- *max = 32768;
- }
- return sizeof(uint64_t);
+ RET((uint64_t []) { screen->ram_size });
+
case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY:
+ RET((uint32_t []) { screen->max_freq / 1000000 });
+
case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS:
+ RET((uint32_t []) { 9999 }); // TODO
+
case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED:
+ RET((uint32_t []) { 1 });
+
case PIPE_COMPUTE_CAP_SUBGROUP_SIZE:
+ RET((uint32_t []) { 32 }); // TODO
+
case PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK:
- break;
+ RET((uint64_t []) { 1024 }); // TODO
}
return 0;
if (is_ir3(screen))
return ir3_get_compiler_options(screen->compiler);
- return NULL;
+ return ir2_get_compiler_options();
}
-boolean
+bool
fd_screen_bo_get_handle(struct pipe_screen *pscreen,
struct fd_bo *bo,
+ struct renderonly_scanout *scanout,
unsigned stride,
struct winsys_handle *whandle)
{
whandle->stride = stride;
- if (whandle->type == DRM_API_HANDLE_TYPE_SHARED) {
+ if (whandle->type == WINSYS_HANDLE_TYPE_SHARED) {
return fd_bo_get_name(bo, &whandle->handle) == 0;
- } else if (whandle->type == DRM_API_HANDLE_TYPE_KMS) {
+ } else if (whandle->type == WINSYS_HANDLE_TYPE_KMS) {
+ if (renderonly_get_handle(scanout, whandle))
+ return true;
whandle->handle = fd_bo_handle(bo);
- return TRUE;
- } else if (whandle->type == DRM_API_HANDLE_TYPE_FD) {
+ return true;
+ } else if (whandle->type == WINSYS_HANDLE_TYPE_FD) {
whandle->handle = fd_bo_dmabuf(bo);
- return TRUE;
+ return true;
} else {
- return FALSE;
+ return false;
}
}
+static void
+fd_screen_query_dmabuf_modifiers(struct pipe_screen *pscreen,
+ enum pipe_format format,
+ int max, uint64_t *modifiers,
+ unsigned int *external_only,
+ int *count)
+{
+ struct fd_screen *screen = fd_screen(pscreen);
+ int i, num = 0;
+
+ max = MIN2(max, screen->num_supported_modifiers);
+
+ if (!max) {
+ max = screen->num_supported_modifiers;
+ external_only = NULL;
+ modifiers = NULL;
+ }
+
+ for (i = 0; i < max; i++) {
+ if (modifiers)
+ modifiers[num] = screen->supported_modifiers[i];
+
+ if (external_only)
+ external_only[num] = 0;
+
+ num++;
+ }
+
+ *count = num;
+}
+
struct fd_bo *
fd_screen_bo_from_handle(struct pipe_screen *pscreen,
struct winsys_handle *whandle)
struct fd_screen *screen = fd_screen(pscreen);
struct fd_bo *bo;
- if (whandle->type == DRM_API_HANDLE_TYPE_SHARED) {
+ if (whandle->type == WINSYS_HANDLE_TYPE_SHARED) {
bo = fd_bo_from_name(screen->dev, whandle->handle);
- } else if (whandle->type == DRM_API_HANDLE_TYPE_KMS) {
+ } else if (whandle->type == WINSYS_HANDLE_TYPE_KMS) {
bo = fd_bo_from_handle(screen->dev, whandle->handle, 0);
- } else if (whandle->type == DRM_API_HANDLE_TYPE_FD) {
+ } else if (whandle->type == WINSYS_HANDLE_TYPE_FD) {
bo = fd_bo_from_dmabuf(screen->dev, whandle->handle);
} else {
DBG("Attempt to import unsupported handle type %d", whandle->type);
}
struct pipe_screen *
-fd_screen_create(struct fd_device *dev)
+fd_screen_create(struct fd_device *dev, struct renderonly *ro)
{
struct fd_screen *screen = CALLOC_STRUCT(fd_screen);
struct pipe_screen *pscreen;
screen->dev = dev;
screen->refcnt = 1;
+ if (ro) {
+ screen->ro = renderonly_dup(ro);
+ if (!screen->ro) {
+ DBG("could not create renderonly object");
+ goto fail;
+ }
+ }
+
// maybe this should be in context?
screen->pipe = fd_pipe_new(screen->dev, FD_PIPE_3D);
if (!screen->pipe) {
}
screen->gmemsize_bytes = val;
+ if (fd_device_version(dev) >= FD_VERSION_GMEM_BASE) {
+ fd_pipe_get_param(screen->pipe, FD_GMEM_BASE, &screen->gmem_base);
+ }
+
if (fd_pipe_get_param(screen->pipe, FD_DEVICE_ID, &val)) {
DBG("could not get device-id");
goto fail;
}
screen->chip_id = val;
+ if (fd_pipe_get_param(screen->pipe, FD_NR_RINGS, &val)) {
+ DBG("could not get # of rings");
+ screen->priority_mask = 0;
+ } else {
+ /* # of rings equates to number of unique priority values: */
+ screen->priority_mask = (1 << val) - 1;
+ }
+
+ if ((fd_device_version(dev) >= FD_VERSION_ROBUSTNESS) &&
+ (fd_pipe_get_param(screen->pipe, FD_PP_PGTABLE, &val) == 0)) {
+ screen->has_robustness = val;
+ }
+
+ struct sysinfo si;
+ sysinfo(&si);
+ screen->ram_size = si.totalram;
+
DBG("Pipe Info:");
DBG(" GPU-id: %d", screen->gpu_id);
DBG(" Chip-id: 0x%08x", screen->chip_id);
* send a patch ;-)
*/
switch (screen->gpu_id) {
+ case 200:
+ case 201:
+ case 205:
case 220:
fd2_screen_init(pscreen);
break;
fd4_screen_init(pscreen);
break;
case 530:
+ case 540:
fd5_screen_init(pscreen);
break;
+ case 630:
+ fd6_screen_init(pscreen);
+ break;
default:
debug_printf("unsupported GPU: a%03d\n", screen->gpu_id);
goto fail;
}
- if (screen->gpu_id >= 500) {
+ if (screen->gpu_id >= 600) {
+ screen->gmem_alignw = 32;
+ screen->gmem_alignh = 32;
+ screen->num_vsc_pipes = 32;
+ } else if (screen->gpu_id >= 500) {
screen->gmem_alignw = 64;
screen->gmem_alignh = 32;
screen->num_vsc_pipes = 16;
screen->num_vsc_pipes = 8;
}
- /* NOTE: don't enable reordering on a2xx, since completely untested.
- * Also, don't enable if we have too old of a kernel to support
+ /* NOTE: don't enable if we have too old of a kernel to support
* growable cmdstream buffers, since memory requirement for cmdstream
* buffers would be too much otherwise.
*/
- if ((screen->gpu_id >= 300) && (fd_device_version(dev) >= FD_VERSION_UNLIMITED_CMDS))
+ if (fd_device_version(dev) >= FD_VERSION_UNLIMITED_CMDS)
screen->reorder = !(fd_mesa_debug & FD_DBG_INORDER);
fd_bc_init(&screen->batch_cache);
pscreen->fence_finish = fd_fence_finish;
pscreen->fence_get_fd = fd_fence_get_fd;
+ pscreen->query_dmabuf_modifiers = fd_screen_query_dmabuf_modifiers;
+
+ if (!screen->supported_modifiers) {
+ static const uint64_t supported_modifiers[] = {
+ DRM_FORMAT_MOD_LINEAR,
+ };
+
+ screen->supported_modifiers = supported_modifiers;
+ screen->num_supported_modifiers = ARRAY_SIZE(supported_modifiers);
+ }
+
slab_create_parent(&screen->transfer_pool, sizeof(struct fd_transfer), 16);
return pscreen;