#include "util/u_format_s3tc.h"
#include "util/u_screen.h"
#include "pipe/p_screen.h"
+#include "compiler/nir/nir.h"
#include "nouveau_vp3_video.h"
#include "nv50/g80_texture.xml.h"
-static boolean
+static bool
nvc0_screen_is_format_supported(struct pipe_screen *pscreen,
enum pipe_format format,
enum pipe_texture_target target,
nvc0_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
{
const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
- struct nouveau_device *dev = nouveau_screen(pscreen)->device;
+ const struct nouveau_screen *screen = nouveau_screen(pscreen);
+ struct nouveau_device *dev = screen->device;
switch (param) {
/* non-boolean caps */
- case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
+ case PIPE_CAP_MAX_TEXTURE_2D_SIZE:
+ return 16384;
case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
return 15;
case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
return class_3d >= GM200_3D_CLASS ? 8 : 0;
case PIPE_CAP_MAX_TEXTURE_UPLOAD_MEMORY_BUDGET:
return 64 * 1024 * 1024;
+ case PIPE_CAP_MAX_VARYINGS:
+ /* NOTE: These only count our slots for GENERIC varyings.
+ * The address space may be larger, but the actual hard limit seems to be
+ * less than what the address space layout permits, so don't add TEXCOORD,
+ * COLOR, etc. here.
+ */
+ return 0x1f0 / 16;
/* supported caps */
case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
case PIPE_CAP_DEPTH_CLIP_DISABLE:
case PIPE_CAP_POINT_SPRITE:
case PIPE_CAP_TGSI_TEXCOORD:
- 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_CAN_BIND_CONST_BUFFER_AS_VERTEX:
case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION:
case PIPE_CAP_QUERY_SO_OVERFLOW:
+ case PIPE_CAP_DEST_SURFACE_SRGB_CONTROL:
+ case PIPE_CAP_TGSI_DIV:
+ case PIPE_CAP_TGSI_ATOMINC_WRAP:
return 1;
case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
return nouveau_screen(pscreen)->vram_domain & NOUVEAU_BO_VRAM ? 1 : 0;
- case PIPE_CAP_TGSI_FS_FBFETCH:
- return class_3d >= NVE4_3D_CLASS; /* needs testing on fermi */
+ case PIPE_CAP_FBFETCH:
+ return class_3d >= NVE4_3D_CLASS ? 1 : 0; /* needs testing on fermi */
case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
case PIPE_CAP_TGSI_BALLOT:
+ return class_3d >= NVE4_3D_CLASS;
case PIPE_CAP_BINDLESS_TEXTURE:
return class_3d >= NVE4_3D_CLASS;
case PIPE_CAP_TGSI_ATOMFADD:
case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_TRIANGLES:
return class_3d >= GP100_3D_CLASS;
+ /* caps has to be turned on with nir */
+ case PIPE_CAP_INT64_DIVMOD:
+ return screen->prefer_nir ? 1 : 0;
+
/* unsupported caps */
case PIPE_CAP_DEPTH_CLIP_DISABLE_SEPARATE:
case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
case PIPE_CAP_NATIVE_FENCE_FD:
case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
- case PIPE_CAP_INT64_DIVMOD:
case PIPE_CAP_SPARSE_BUFFER_PAGE_SIZE:
case PIPE_CAP_NIR_SAMPLERS_AS_DEREF:
case PIPE_CAP_MEMOBJ:
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_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_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:
return 0;
case PIPE_CAP_VENDOR_ID:
return dev->vram_size >> 20;
case PIPE_CAP_UMA:
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_MAX_FRAMES_IN_FLIGHT:
return u_pipe_screen_get_param_defaults(pscreen, param);
}
}
enum pipe_shader_type shader,
enum pipe_shader_cap param)
{
- const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
+ const struct nouveau_screen *screen = nouveau_screen(pscreen);
+ const uint16_t class_3d = screen->class_3d;
switch (shader) {
case PIPE_SHADER_VERTEX:
switch (param) {
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_SUPPORTED_IRS:
- return 1 << PIPE_SHADER_IR_TGSI;
+ return 1 << PIPE_SHADER_IR_TGSI |
+ 1 << PIPE_SHADER_IR_NIR;
case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
return 16;
case PIPE_SHADER_CAP_MAX_INPUTS:
- if (shader == PIPE_SHADER_VERTEX)
- return 32;
- /* NOTE: These only count our slots for GENERIC varyings.
- * The address space may be larger, but the actual hard limit seems to be
- * less than what the address space layout permits, so don't add TEXCOORD,
- * COLOR, etc. here.
- */
- if (shader == PIPE_SHADER_FRAGMENT)
- return 0x1f0 / 16;
- /* Actually this counts CLIPVERTEX, which occupies the last generic slot,
- * and excludes 0x60 per-patch inputs.
- */
return 0x200 / 16;
case PIPE_SHADER_CAP_MAX_OUTPUTS:
return 32;
IMMED_NVC0(push, NVC0_3D(CB_BIND(stage)), (index << 4) | (size >= 0));
}
+static const nir_shader_compiler_options nir_options = {
+ .lower_fdiv = false,
+ .lower_ffma = false,
+ .fuse_ffma = false, /* nir doesn't track mad vs fma */
+ .lower_flrp32 = true,
+ .lower_flrp64 = true,
+ .lower_fpow = false,
+ .lower_fsat = false,
+ .lower_fsqrt = false, // TODO: only before gm200
+ .lower_fmod = true,
+ .lower_bitfield_extract = false,
+ .lower_bitfield_extract_to_shifts = false,
+ .lower_bitfield_insert = false,
+ .lower_bitfield_insert_to_shifts = false,
+ .lower_bitfield_reverse = false,
+ .lower_bit_count = false,
+ .lower_ifind_msb = false,
+ .lower_find_lsb = false,
+ .lower_uadd_carry = true, // TODO
+ .lower_usub_borrow = true, // TODO
+ .lower_mul_high = false,
+ .lower_negate = false,
+ .lower_sub = false, // TODO
+ .lower_scmp = true, // TODO: not implemented yet
+ .lower_idiv = true,
+ .lower_isign = false, // TODO
+ .fdot_replicates = false, // TODO
+ .lower_ffloor = false, // TODO
+ .lower_ffract = true,
+ .lower_fceil = false, // TODO
+ .lower_ldexp = 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,
+ .vertex_id_zero_based = false,
+ .lower_base_vertex = false,
+ .lower_helper_invocation = false,
+ .lower_cs_local_index_from_id = true,
+ .lower_cs_local_id_from_index = false,
+ .lower_device_index_to_zero = false, // TODO
+ .lower_wpos_pntc = false, // TODO
+ .lower_hadd = true, // TODO
+ .lower_add_sat = true, // TODO
+ .use_interpolated_input_intrinsics = true,
+ .lower_mul_2x32_64 = true, // TODO
+ .max_unroll_iterations = 32,
+ .lower_int64_options = nir_lower_divmod64, // TODO
+ .lower_doubles_options = nir_lower_dmod, // TODO
+};
+
+static const void *
+nvc0_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;
+}
+
#define FAIL_SCREEN_INIT(str, err) \
do { \
NOUVEAU_ERR(str, err); \
pscreen->get_sample_pixel_grid = nvc0_screen_get_sample_pixel_grid;
pscreen->get_driver_query_info = nvc0_screen_get_driver_query_info;
pscreen->get_driver_query_group_info = nvc0_screen_get_driver_query_group_info;
+ /* nir stuff */
+ pscreen->get_compiler_options = nvc0_screen_get_compiler_options;
nvc0_screen_init_resource_functions(pscreen);
for (i = 0; i < NVC0_MAX_VIEWPORTS; i++) {
BEGIN_NVC0(push, NVC0_3D(SCISSOR_ENABLE(i)), 3);
PUSH_DATA (push, 1);
- PUSH_DATA (push, 8192 << 16);
- PUSH_DATA (push, 8192 << 16);
+ PUSH_DATA (push, 16384 << 16);
+ PUSH_DATA (push, 16384 << 16);
}
#define MK_MACRO(m, n) i = nvc0_graph_set_macro(screen, m, i, sizeof(n), n);
MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT_COUNT, mme9097_draw_elts_indirect_count);
MK_MACRO(NVC0_3D_MACRO_QUERY_BUFFER_WRITE, mme9097_query_buffer_write);
MK_MACRO(NVC0_3D_MACRO_CONSERVATIVE_RASTER_STATE, mme9097_conservative_raster_state);
+ MK_MACRO(NVC0_3D_MACRO_COMPUTE_COUNTER, mme9097_compute_counter);
+ MK_MACRO(NVC0_3D_MACRO_COMPUTE_COUNTER_TO_QUERY, mme9097_compute_counter_to_query);
MK_MACRO(NVC0_CP_MACRO_LAUNCH_GRID_INDIRECT, mme90c0_launch_grid_indirect);
BEGIN_NVC0(push, NVC0_3D(RASTERIZE_ENABLE), 1);
projects / mesa.git / blobdiff