#include "si_pipe.h"
#include "si_public.h"
+#include "sid.h"
+#include "radeon/radeon_llvm_emit.h"
#include "radeon/radeon_uvd.h"
-#include "util/u_blitter.h"
#include "util/u_memory.h"
-#include "util/u_simple_shaders.h"
#include "vl/vl_decoder.h"
+#include <llvm-c/Target.h>
+#include <llvm-c/TargetMachine.h>
+
/*
* pipe_context
*/
-static void si_flush(struct pipe_context *ctx, unsigned flags,
- struct pipe_fence_handle **fence)
-{
- struct si_context *sctx = (struct si_context *)ctx;
- struct pipe_query *render_cond = NULL;
- boolean render_cond_cond = FALSE;
- unsigned render_cond_mode = 0;
-
- /* Disable render condition. */
- if (sctx->b.current_render_cond) {
- render_cond = sctx->b.current_render_cond;
- render_cond_cond = sctx->b.current_render_cond_cond;
- render_cond_mode = sctx->b.current_render_cond_mode;
- ctx->render_condition(ctx, NULL, FALSE, 0);
- }
-
- si_context_flush(sctx, flags, fence);
-
- /* Re-enable render condition. */
- if (render_cond) {
- ctx->render_condition(ctx, render_cond, render_cond_cond, render_cond_mode);
- }
-}
-
-static void si_flush_from_st(struct pipe_context *ctx,
- struct pipe_fence_handle **fence,
- unsigned flags)
-{
- struct si_context *sctx = (struct si_context *)ctx;
- unsigned rflags = 0;
-
- if (flags & PIPE_FLUSH_END_OF_FRAME)
- rflags |= RADEON_FLUSH_END_OF_FRAME;
-
- if (sctx->b.rings.dma.cs) {
- sctx->b.rings.dma.flush(sctx, rflags, NULL);
- }
-
- si_flush(ctx, rflags, fence);
-}
-
-static void si_flush_gfx_ring(void *ctx, unsigned flags,
- struct pipe_fence_handle **fence)
-{
- si_flush(ctx, flags, fence);
-}
-
static void si_destroy_context(struct pipe_context *context)
{
struct si_context *sctx = (struct si_context *)context;
si_release_all_descriptors(sctx);
+ pipe_resource_reference(&sctx->esgs_ring, NULL);
+ pipe_resource_reference(&sctx->gsvs_ring, NULL);
pipe_resource_reference(&sctx->null_const_buf.buffer, NULL);
r600_resource_reference(&sctx->border_color_table, NULL);
if (sctx->dummy_pixel_shader) {
sctx->b.b.delete_fs_state(&sctx->b.b, sctx->dummy_pixel_shader);
}
- for (int i = 0; i < 8; i++) {
- sctx->b.b.delete_depth_stencil_alpha_state(&sctx->b.b, sctx->custom_dsa_flush_depth_stencil[i]);
- sctx->b.b.delete_depth_stencil_alpha_state(&sctx->b.b, sctx->custom_dsa_flush_depth[i]);
- sctx->b.b.delete_depth_stencil_alpha_state(&sctx->b.b, sctx->custom_dsa_flush_stencil[i]);
- }
- sctx->b.b.delete_depth_stencil_alpha_state(&sctx->b.b, sctx->custom_dsa_flush_inplace);
+ sctx->b.b.delete_depth_stencil_alpha_state(&sctx->b.b, sctx->custom_dsa_flush);
sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_resolve);
sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_decompress);
sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_fastclear);
sctx->b.b.screen = screen; /* this must be set first */
sctx->b.b.priv = priv;
sctx->b.b.destroy = si_destroy_context;
- sctx->b.b.flush = si_flush_from_st;
sctx->screen = sscreen; /* Easy accessing of screen/winsys. */
if (!r600_common_context_init(&sctx->b, &sscreen->b))
sctx->b.b.create_video_buffer = vl_video_buffer_create;
}
- sctx->b.rings.gfx.cs = ws->cs_create(ws, RING_GFX, si_flush_gfx_ring,
- sctx, NULL);
- sctx->b.rings.gfx.flush = si_flush_gfx_ring;
+ sctx->b.rings.gfx.cs = ws->cs_create(ws, RING_GFX, si_context_gfx_flush,
+ sctx, sscreen->b.trace_bo ?
+ sscreen->b.trace_bo->cs_buf : NULL);
+ sctx->b.rings.gfx.flush = si_context_gfx_flush;
si_init_all_descriptors(sctx);
/* Initialize cache_flush. */
sctx->cache_flush = si_atom_cache_flush;
- sctx->atoms.cache_flush = &sctx->cache_flush;
+ sctx->atoms.s.cache_flush = &sctx->cache_flush;
+
+ sctx->msaa_config = si_atom_msaa_config;
+ sctx->atoms.s.msaa_config = &sctx->msaa_config;
- sctx->atoms.streamout_begin = &sctx->b.streamout.begin_atom;
- sctx->atoms.streamout_enable = &sctx->b.streamout.enable_atom;
+ sctx->atoms.s.streamout_begin = &sctx->b.streamout.begin_atom;
+ sctx->atoms.s.streamout_enable = &sctx->b.streamout.enable_atom;
switch (sctx->b.chip_class) {
case SI:
case CIK:
si_init_state_functions(sctx);
+ si_init_shader_functions(sctx);
si_init_config(sctx);
break;
default:
goto fail;
}
+ if (sscreen->b.debug_flags & DBG_FORCE_DMA)
+ sctx->b.b.resource_copy_region = sctx->b.dma_copy;
+
sctx->blitter = util_blitter_create(&sctx->b.b);
if (sctx->blitter == NULL)
goto fail;
-
- sctx->dummy_pixel_shader =
- util_make_fragment_cloneinput_shader(&sctx->b.b, 0,
- TGSI_SEMANTIC_GENERIC,
- TGSI_INTERPOLATE_CONSTANT);
- sctx->b.b.bind_fs_state(&sctx->b.b, sctx->dummy_pixel_shader);
+ sctx->blitter->draw_rectangle = r600_draw_rectangle;
/* these must be last */
si_begin_new_cs(sctx);
sctx->null_const_buf.buffer_size = sctx->null_const_buf.buffer->width0;
for (shader = 0; shader < SI_NUM_SHADERS; shader++) {
- for (i = 0; i < NUM_CONST_BUFFERS; i++) {
+ for (i = 0; i < SI_NUM_CONST_BUFFERS; i++) {
sctx->b.b.set_constant_buffer(&sctx->b.b, shader, i,
&sctx->null_const_buf);
}
case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
+ case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
case PIPE_CAP_SM3:
case PIPE_CAP_SEAMLESS_CUBE_MAP:
case PIPE_CAP_PRIMITIVE_RESTART:
case PIPE_CAP_TGSI_INSTANCEID:
case PIPE_CAP_COMPUTE:
case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
- case PIPE_CAP_TGSI_VS_LAYER:
+ case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
+ case PIPE_CAP_CUBE_MAP_ARRAY:
+ case PIPE_CAP_SAMPLE_SHADING:
+ case PIPE_CAP_DRAW_INDIRECT:
+ case PIPE_CAP_CLIP_HALFZ:
+ case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
return 1;
case PIPE_CAP_TEXTURE_MULTISAMPLE:
/* 2D tiling on CIK is supported since DRM 2.35.0 */
- return HAVE_LLVM >= 0x0304 && (sscreen->b.chip_class < CIK ||
- sscreen->b.info.drm_minor >= 35);
+ return sscreen->b.chip_class < CIK ||
+ sscreen->b.info.drm_minor >= 35;
case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
return R600_MAP_BUFFER_ALIGNMENT;
return 4;
case PIPE_CAP_GLSL_FEATURE_LEVEL:
- return HAVE_LLVM >= 0x0305 ? 330 : 140;
+ return 330;
case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
return MIN2(sscreen->b.info.vram_size, 0xFFFFFFFF);
+ case PIPE_CAP_TEXTURE_QUERY_LOD:
+ case PIPE_CAP_TEXTURE_GATHER_SM5:
+ return HAVE_LLVM >= 0x0305;
+ case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
+ return HAVE_LLVM >= 0x0305 ? 4 : 0;
+
/* Unsupported features. */
case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
- case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
case PIPE_CAP_VERTEX_COLOR_CLAMPED:
case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
case PIPE_CAP_USER_VERTEX_BUFFERS:
- case PIPE_CAP_CUBE_MAP_ARRAY:
- case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
- case PIPE_CAP_TEXTURE_GATHER_SM5:
case PIPE_CAP_TGSI_TEXCOORD:
case PIPE_CAP_FAKE_SW_MSAA:
- case PIPE_CAP_TEXTURE_QUERY_LOD:
+ case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
+ case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
+ case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
+ case PIPE_CAP_SAMPLER_VIEW_TARGET:
+ case PIPE_CAP_VERTEXID_NOBASE:
return 0;
case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
return 1024;
case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
return 4095;
+ case PIPE_CAP_MAX_VERTEX_STREAMS:
+ return 1;
+
+ case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
+ return 2048;
/* Texturing. */
case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
case PIPE_CAP_MIN_TEXEL_OFFSET:
- return -8;
+ return -32;
case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
case PIPE_CAP_MAX_TEXEL_OFFSET:
- return 7;
+ return 31;
+
case PIPE_CAP_ENDIANNESS:
return PIPE_ENDIAN_LITTLE;
+
+ case PIPE_CAP_VENDOR_ID:
+ return 0x1002;
+ case PIPE_CAP_DEVICE_ID:
+ return sscreen->b.info.pci_id;
+ case PIPE_CAP_ACCELERATED:
+ return 1;
+ case PIPE_CAP_VIDEO_MEMORY:
+ return sscreen->b.info.vram_size >> 20;
+ case PIPE_CAP_UMA:
+ return 0;
}
return 0;
}
{
case PIPE_SHADER_FRAGMENT:
case PIPE_SHADER_VERTEX:
- break;
case PIPE_SHADER_GEOMETRY:
-#if HAVE_LLVM < 0x0305
- return 0;
-#endif
break;
case PIPE_SHADER_COMPUTE:
switch (param) {
case PIPE_SHADER_CAP_PREFERRED_IR:
+#if HAVE_LLVM < 0x0306
return PIPE_SHADER_IR_LLVM;
+#else
+ return PIPE_SHADER_IR_NATIVE;
+#endif
+ case PIPE_SHADER_CAP_DOUBLES:
+ return 0; /* XXX: Enable doubles once the compiler can
+ handle them. */
+ case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE: {
+ uint64_t max_const_buffer_size;
+ pscreen->get_compute_param(pscreen,
+ PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE,
+ &max_const_buffer_size);
+ return max_const_buffer_size;
+ }
default:
return 0;
}
case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
return 32;
case PIPE_SHADER_CAP_MAX_INPUTS:
- return 32;
+ return shader == PIPE_SHADER_VERTEX ? SI_NUM_VERTEX_BUFFERS : 32;
+ case PIPE_SHADER_CAP_MAX_OUTPUTS:
+ return shader == PIPE_SHADER_FRAGMENT ? 8 : 32;
case PIPE_SHADER_CAP_MAX_TEMPS:
return 256; /* Max native temporaries. */
- case PIPE_SHADER_CAP_MAX_ADDRS:
- /* FIXME Isn't this equal to TEMPS? */
- return 1; /* Max native address registers */
- case PIPE_SHADER_CAP_MAX_CONSTS:
- return 4096; /* actually only memory limits this */
+ case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
+ return 4096 * sizeof(float[4]); /* actually only memory limits this */
case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
- return NUM_PIPE_CONST_BUFFERS;
+ return SI_NUM_USER_CONST_BUFFERS;
case PIPE_SHADER_CAP_MAX_PREDS:
return 0; /* FIXME */
case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
return 16;
case PIPE_SHADER_CAP_PREFERRED_IR:
return PIPE_SHADER_IR_TGSI;
+ case PIPE_SHADER_CAP_DOUBLES:
+ return 0;
}
return 0;
}
return;
r600_destroy_common_screen(&sscreen->b);
+
+#if HAVE_LLVM >= 0x0306
+ LLVMDisposeTargetMachine(sscreen->tm);
+#endif
+}
+
+#define SI_TILE_MODE_COLOR_2D_8BPP 14
+
+/* Initialize pipe config. This is especially important for GPUs
+ * with 16 pipes and more where it's initialized incorrectly by
+ * the TILING_CONFIG ioctl. */
+static bool si_initialize_pipe_config(struct si_screen *sscreen)
+{
+ unsigned mode2d;
+
+ /* This is okay, because there can be no 2D tiling without
+ * the tile mode array, so we won't need the pipe config.
+ * Return "success".
+ */
+ if (!sscreen->b.info.si_tile_mode_array_valid)
+ return true;
+
+ /* The same index is used for the 2D mode on CIK too. */
+ mode2d = sscreen->b.info.si_tile_mode_array[SI_TILE_MODE_COLOR_2D_8BPP];
+
+ switch (G_009910_PIPE_CONFIG(mode2d)) {
+ case V_02803C_ADDR_SURF_P2:
+ sscreen->b.tiling_info.num_channels = 2;
+ break;
+ case V_02803C_X_ADDR_SURF_P4_8X16:
+ case V_02803C_X_ADDR_SURF_P4_16X16:
+ case V_02803C_X_ADDR_SURF_P4_16X32:
+ case V_02803C_X_ADDR_SURF_P4_32X32:
+ sscreen->b.tiling_info.num_channels = 4;
+ break;
+ case V_02803C_X_ADDR_SURF_P8_16X16_8X16:
+ case V_02803C_X_ADDR_SURF_P8_16X32_8X16:
+ case V_02803C_X_ADDR_SURF_P8_32X32_8X16:
+ case V_02803C_X_ADDR_SURF_P8_16X32_16X16:
+ case V_02803C_X_ADDR_SURF_P8_32X32_16X16:
+ case V_02803C_X_ADDR_SURF_P8_32X32_16X32:
+ case V_02803C_X_ADDR_SURF_P8_32X64_32X32:
+ sscreen->b.tiling_info.num_channels = 8;
+ break;
+ case V_02803C_X_ADDR_SURF_P16_32X32_8X16:
+ case V_02803C_X_ADDR_SURF_P16_32X32_16X16:
+ sscreen->b.tiling_info.num_channels = 16;
+ break;
+ default:
+ assert(0);
+ fprintf(stderr, "radeonsi: Unknown pipe config %i.\n",
+ G_009910_PIPE_CONFIG(mode2d));
+ return false;
+ }
+ return true;
}
struct pipe_screen *radeonsi_screen_create(struct radeon_winsys *ws)
{
struct si_screen *sscreen = CALLOC_STRUCT(si_screen);
+ LLVMTargetRef r600_target;
+#if HAVE_LLVM >= 0x0306
+ const char *triple = "amdgcn--";
+#else
+ const char *triple = "r600--";
+#endif
if (sscreen == NULL) {
return NULL;
}
sscreen->b.b.is_format_supported = si_is_format_supported;
sscreen->b.b.resource_create = r600_resource_create_common;
- if (!r600_common_screen_init(&sscreen->b, ws)) {
+ if (!r600_common_screen_init(&sscreen->b, ws) ||
+ !si_initialize_pipe_config(sscreen)) {
FREE(sscreen);
return NULL;
}
sscreen->b.has_cp_dma = true;
- sscreen->b.has_streamout = HAVE_LLVM >= 0x0304;
+ sscreen->b.has_streamout = true;
if (debug_get_bool_option("RADEON_DUMP_SHADERS", FALSE))
sscreen->b.debug_flags |= DBG_FS | DBG_VS | DBG_GS | DBG_PS | DBG_CS;
/* Create the auxiliary context. This must be done last. */
sscreen->b.aux_context = sscreen->b.b.context_create(&sscreen->b.b, NULL);
+#if HAVE_LLVM >= 0x0306
+ /* Initialize LLVM TargetMachine */
+ r600_target = radeon_llvm_get_r600_target(triple);
+ sscreen->tm = LLVMCreateTargetMachine(r600_target, triple,
+ r600_get_llvm_processor_name(sscreen->b.family),
+ "+DumpCode", LLVMCodeGenLevelDefault, LLVMRelocDefault,
+ LLVMCodeModelDefault);
+#endif
return &sscreen->b.b;
}
projects / mesa.git / blobdiff