#include <errno.h>
#include <xf86drm.h>
#include <nouveau_drm.h>
-#include "util/u_format.h"
-#include "util/u_format_s3tc.h"
+#include "util/format/u_format.h"
+#include "util/format/u_format_s3tc.h"
+#include "util/u_screen.h"
#include "pipe/p_screen.h"
+#include "compiler/nir/nir.h"
#include "nv50/nv50_context.h"
#include "nv50/nv50_screen.h"
#define THREADS_IN_WARP 32
-static boolean
+static bool
nv50_screen_is_format_supported(struct pipe_screen *pscreen,
enum pipe_format format,
enum pipe_texture_target target,
unsigned sample_count,
+ unsigned storage_sample_count,
unsigned bindings)
{
if (sample_count > 8)
if (sample_count == 8 && util_format_get_blocksizebits(format) >= 128)
return false;
- if (!util_format_is_supported(format, bindings))
+ if (MAX2(1, sample_count) != MAX2(1, storage_sample_count))
return false;
switch (format) {
switch (param) {
/* non-boolean caps */
- case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
- return 14;
+ case PIPE_CAP_MAX_TEXTURE_2D_SIZE:
+ return 8192;
case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
return 12;
case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
return 128 * 1024 * 1024;
case PIPE_CAP_GLSL_FEATURE_LEVEL:
return 330;
+ case PIPE_CAP_GLSL_FEATURE_LEVEL_COMPATIBILITY:
+ return 330;
case PIPE_CAP_MAX_RENDER_TARGETS:
return 8;
case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
return 1;
+ case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS:
+ case PIPE_CAP_RASTERIZER_SUBPIXEL_BITS:
+ return 8;
case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
return 4;
case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
return 1024;
case PIPE_CAP_MAX_VERTEX_STREAMS:
return 1;
+ case PIPE_CAP_MAX_GS_INVOCATIONS:
+ return 0;
+ case PIPE_CAP_MAX_SHADER_BUFFER_SIZE:
+ return 0;
case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
return 2048;
+ case PIPE_CAP_MAX_VERTEX_ELEMENT_SRC_OFFSET:
+ return 2047;
case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
return 256;
case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
return (class_3d >= NVA3_3D_CLASS) ? 4 : 0;
case PIPE_CAP_MAX_WINDOW_RECTANGLES:
return NV50_MAX_WINDOW_RECTANGLES;
+ case PIPE_CAP_MAX_TEXTURE_UPLOAD_MEMORY_BUDGET:
+ return 16 * 1024 * 1024;
+ case PIPE_CAP_MAX_VARYINGS:
+ return 15;
/* supported caps */
case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
+ case PIPE_CAP_TEXTURE_MIRROR_CLAMP_TO_EDGE:
case PIPE_CAP_TEXTURE_SWIZZLE:
- case PIPE_CAP_TEXTURE_SHADOW_MAP:
case PIPE_CAP_NPOT_TEXTURES:
case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
case PIPE_CAP_MIXED_COLOR_DEPTH_BITS:
case PIPE_CAP_ANISOTROPIC_FILTER:
case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
- case PIPE_CAP_TWO_SIDED_STENCIL:
case PIPE_CAP_DEPTH_CLIP_DISABLE:
case PIPE_CAP_POINT_SPRITE:
- 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_FRAGMENT_COLOR_CLAMPED:
case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
case PIPE_CAP_VERTEX_COLOR_CLAMPED:
case PIPE_CAP_TEXTURE_BARRIER:
case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
case PIPE_CAP_START_INSTANCE:
- case PIPE_CAP_USER_CONSTANT_BUFFERS:
case PIPE_CAP_USER_VERTEX_BUFFERS:
case PIPE_CAP_TEXTURE_MULTISAMPLE:
case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
case PIPE_CAP_SHAREABLE_SHADERS:
case PIPE_CAP_CLEAR_TEXTURE:
- case PIPE_CAP_COMPUTE:
case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL:
case PIPE_CAP_INVALIDATE_BUFFER:
case PIPE_CAP_STRING_MARKER:
case PIPE_CAP_TGSI_ARRAY_COMPONENTS:
case PIPE_CAP_TGSI_MUL_ZERO_WINS:
case PIPE_CAP_TGSI_TEX_TXF_LZ:
+ case PIPE_CAP_TGSI_CLOCK:
+ case PIPE_CAP_CAN_BIND_CONST_BUFFER_AS_VERTEX:
+ case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION:
+ case PIPE_CAP_DEST_SURFACE_SRGB_CONTROL:
+ case PIPE_CAP_TGSI_DIV:
return 1;
case PIPE_CAP_SEAMLESS_CUBE_MAP:
return 1; /* class_3d >= NVA0_3D_CLASS; */
return class_3d >= NVA3_3D_CLASS;
/* unsupported caps */
+ case PIPE_CAP_DEPTH_CLIP_DISABLE_SEPARATE:
case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES:
case PIPE_CAP_TGSI_VOTE:
case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED:
- case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS:
case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS:
case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
case PIPE_CAP_NATIVE_FENCE_FD:
case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
- case PIPE_CAP_TGSI_FS_FBFETCH:
+ case PIPE_CAP_FBFETCH:
case PIPE_CAP_DOUBLES:
case PIPE_CAP_INT64:
case PIPE_CAP_INT64_DIVMOD:
- 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_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_LOAD_CONSTBUF:
+ 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_FRAMEBUFFER_MSAA_CONSTRAINTS:
+ case PIPE_CAP_SIGNED_VERTEX_BUFFER_OFFSET:
+ case PIPE_CAP_CONTEXT_PRIORITY_MASK:
+ case PIPE_CAP_FENCE_SIGNAL:
+ case PIPE_CAP_CONSTBUF0_FLAGS:
+ case PIPE_CAP_PACKED_UNIFORMS:
+ case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_TRIANGLES:
+ case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_POINTS_LINES:
+ case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_TRIANGLES:
+ case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_POINTS_LINES:
+ case PIPE_CAP_CONSERVATIVE_RASTER_POST_DEPTH_COVERAGE:
+ case PIPE_CAP_MAX_CONSERVATIVE_RASTER_SUBPIXEL_PRECISION_BIAS:
+ case PIPE_CAP_PROGRAMMABLE_SAMPLE_LOCATIONS:
+ case PIPE_CAP_MAX_COMBINED_SHADER_BUFFERS:
+ case PIPE_CAP_MAX_COMBINED_HW_ATOMIC_COUNTERS:
+ case PIPE_CAP_MAX_COMBINED_HW_ATOMIC_COUNTER_BUFFERS:
+ case PIPE_CAP_SURFACE_SAMPLE_COUNT:
+ case PIPE_CAP_TGSI_ATOMFADD:
+ case PIPE_CAP_QUERY_PIPELINE_STATISTICS_SINGLE:
+ case PIPE_CAP_RGB_OVERRIDE_DST_ALPHA_BLEND:
+ case PIPE_CAP_GLSL_TESS_LEVELS_AS_INPUTS:
+ case PIPE_CAP_NIR_COMPACT_ARRAYS:
+ case PIPE_CAP_COMPUTE:
+ case PIPE_CAP_IMAGE_LOAD_FORMATTED:
+ case PIPE_CAP_COMPUTE_SHADER_DERIVATIVES:
+ case PIPE_CAP_ATOMIC_FLOAT_MINMAX:
+ case PIPE_CAP_CONSERVATIVE_RASTER_INNER_COVERAGE:
+ case PIPE_CAP_FRAGMENT_SHADER_INTERLOCK:
+ case PIPE_CAP_CS_DERIVED_SYSTEM_VALUES_SUPPORTED:
+ case PIPE_CAP_FBFETCH_COHERENT:
+ case PIPE_CAP_TGSI_SKIP_SHRINK_IO_ARRAYS:
+ case PIPE_CAP_TGSI_ATOMINC_WRAP:
+ case PIPE_CAP_DEMOTE_TO_HELPER_INVOCATION:
return 0;
case PIPE_CAP_VENDOR_ID:
return dev->vram_size >> 20;
case PIPE_CAP_UMA:
return 0;
- }
- NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
- return 0;
+ default:
+ debug_printf("%s: unhandled cap %d\n", __func__, param);
+ /* fallthrough */
+ /* caps where we want the default value */
+ case PIPE_CAP_DMABUF:
+ case PIPE_CAP_ESSL_FEATURE_LEVEL:
+ case PIPE_CAP_THROTTLE:
+ return u_pipe_screen_get_param_defaults(pscreen, param);
+ }
}
static int
enum pipe_shader_type shader,
enum pipe_shader_cap param)
{
+ const struct nouveau_screen *screen = nouveau_screen(pscreen);
+
switch (shader) {
case PIPE_SHADER_VERTEX:
case PIPE_SHADER_GEOMETRY:
return 1;
case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
return 1;
+ case PIPE_SHADER_CAP_INT64_ATOMICS:
+ case PIPE_SHADER_CAP_FP16:
case PIPE_SHADER_CAP_SUBROUTINES:
return 0; /* please inline, or provide function declarations */
case PIPE_SHADER_CAP_INTEGERS:
return 1;
+ case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS:
+ return 1;
case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
/* The chip could handle more sampler views than samplers */
case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
return MIN2(16, PIPE_MAX_SAMPLERS);
case PIPE_SHADER_CAP_PREFERRED_IR:
- return PIPE_SHADER_IR_TGSI;
+ return screen->prefer_nir ? PIPE_SHADER_IR_NIR : PIPE_SHADER_IR_TGSI;
case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
return 32;
case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_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_SHADER_BUFFERS:
case PIPE_SHADER_CAP_SUPPORTED_IRS:
case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD:
+ case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS:
+ case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS:
return 0;
default:
NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
return 16.0f;
case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
return 4.0f;
- case PIPE_CAPF_GUARD_BAND_LEFT:
- case PIPE_CAPF_GUARD_BAND_TOP:
+ case PIPE_CAPF_MIN_CONSERVATIVE_RASTER_DILATE:
+ case PIPE_CAPF_MAX_CONSERVATIVE_RASTER_DILATE:
+ case PIPE_CAPF_CONSERVATIVE_RASTER_DILATE_GRANULARITY:
return 0.0f;
- case PIPE_CAPF_GUARD_BAND_RIGHT:
- case PIPE_CAPF_GUARD_BAND_BOTTOM:
- return 0.0f; /* that or infinity */
}
NOUVEAU_ERR("unknown PIPE_CAPF %d\n", param);
PUSH_DATA (push, 0);
}
+ BEGIN_NV04(push, NV50_3D(UNK0FDC), 1);
+ PUSH_DATA (push, 1);
+ BEGIN_NV04(push, NV50_3D(UNK19C0), 1);
+ PUSH_DATA (push, 1);
+
PUSH_KICK (push);
}
return 1;
}
+static const nir_shader_compiler_options nir_options = {
+ .fuse_ffma = false, /* nir doesn't track mad vs fma */
+ .lower_flrp32 = true,
+ .lower_flrp64 = true,
+ .lower_fpow = false,
+ .lower_uadd_carry = true,
+ .lower_usub_borrow = true,
+ .lower_sub = true,
+ .lower_ffract = true,
+ .lower_pack_half_2x16 = true,
+ .lower_pack_unorm_2x16 = true,
+ .lower_pack_snorm_2x16 = true,
+ .lower_pack_unorm_4x8 = true,
+ .lower_pack_snorm_4x8 = true,
+ .lower_unpack_half_2x16 = true,
+ .lower_unpack_unorm_2x16 = true,
+ .lower_unpack_snorm_2x16 = true,
+ .lower_unpack_unorm_4x8 = true,
+ .lower_unpack_snorm_4x8 = true,
+ .lower_extract_byte = true,
+ .lower_extract_word = true,
+ .lower_all_io_to_temps = false,
+ .lower_cs_local_index_from_id = true,
+ .lower_rotate = true,
+ .lower_to_scalar = true,
+ .use_interpolated_input_intrinsics = true,
+ .max_unroll_iterations = 32,
+};
+
+static const void *
+nv50_screen_get_compiler_options(struct pipe_screen *pscreen,
+ enum pipe_shader_ir ir,
+ enum pipe_shader_type shader)
+{
+ if (ir == PIPE_SHADER_IR_NIR)
+ return &nir_options;
+ return NULL;
+}
+
struct nouveau_screen *
nv50_screen_create(struct nouveau_device *dev)
{
pscreen->get_driver_query_info = nv50_screen_get_driver_query_info;
pscreen->get_driver_query_group_info = nv50_screen_get_driver_query_group_info;
+ /* nir stuff */
+ pscreen->get_compiler_options = nv50_screen_get_compiler_options;
+
nv50_screen_init_resource_functions(pscreen);
if (screen->base.device->chipset < 0x84 ||
nouveau_getparam(dev, NOUVEAU_GETPARAM_GRAPH_UNITS, &value);
screen->TPs = util_bitcount(value & 0xffff);
- screen->MPsInTP = util_bitcount((value >> 24) & 0xf);
+ screen->MPsInTP = util_bitcount(value & 0x0f000000);
screen->mp_count = screen->TPs * screen->MPsInTP;
projects / mesa.git / blobdiff