* Christian König <christian.koenig@amd.com>
*/
+/* The compiler middle-end architecture: Explaining (non-)monolithic shaders
+ * -------------------------------------------------------------------------
+ *
+ * Typically, there is one-to-one correspondence between API and HW shaders,
+ * that is, for every API shader, there is exactly one shader binary in
+ * the driver.
+ *
+ * The problem with that is that we also have to emulate some API states
+ * (e.g. alpha-test, and many others) in shaders too. The two obvious ways
+ * to deal with it are:
+ * - each shader has multiple variants for each combination of emulated states,
+ * and the variants are compiled on demand, possibly relying on a shader
+ * cache for good performance
+ * - patch shaders at the binary level
+ *
+ * This driver uses something completely different. The emulated states are
+ * usually implemented at the beginning or end of shaders. Therefore, we can
+ * split the shader into 3 parts:
+ * - prolog part (shader code dependent on states)
+ * - main part (the API shader)
+ * - epilog part (shader code dependent on states)
+ *
+ * Each part is compiled as a separate shader and the final binaries are
+ * concatenated. This type of shader is called non-monolithic, because it
+ * consists of multiple independent binaries. Creating a new shader variant
+ * is therefore only a concatenation of shader parts (binaries) and doesn't
+ * involve any compilation. The main shader parts are the only parts that are
+ * compiled when applications create shader objects. The prolog and epilog
+ * parts are compiled on the first use and saved, so that their binaries can
+ * be reused by many other shaders.
+ *
+ * One of the roles of the prolog part is to compute vertex buffer addresses
+ * for vertex shaders. A few of the roles of the epilog part are color buffer
+ * format conversions in pixel shaders that we have to do manually, and write
+ * tessellation factors in tessellation control shaders. The prolog and epilog
+ * have many other important responsibilities in various shader stages.
+ * They don't just "emulate legacy stuff".
+ *
+ * Monolithic shaders are shaders where the parts are combined before LLVM
+ * compilation, and the whole thing is compiled and optimized as one unit with
+ * one binary on the output. The result is the same as the non-monolithic
+ * shader, but the final code can be better, because LLVM can optimize across
+ * all shader parts. Monolithic shaders aren't usually used except for these
+ * special cases:
+ *
+ * 1) Some rarely-used states require modification of the main shader part
+ * itself, and in such cases, only the monolithic shader variant is
+ * compiled, and that's always done on the first use.
+ *
+ * 2) When we do cross-stage optimizations for separate shader objects and
+ * e.g. eliminate unused shader varyings, the resulting optimized shader
+ * variants are always compiled as monolithic shaders, and always
+ * asynchronously (i.e. not stalling ongoing rendering). We call them
+ * "optimized monolithic" shaders. The important property here is that
+ * the non-monolithic unoptimized shader variant is always available for use
+ * when the asynchronous compilation of the optimized shader is not done
+ * yet.
+ *
+ * Starting with GFX9 chips, some shader stages are merged, and the number of
+ * shader parts per shader increased. The complete new list of shader parts is:
+ * - 1st shader: prolog part
+ * - 1st shader: main part
+ * - 2nd shader: prolog part
+ * - 2nd shader: main part
+ * - 2nd shader: epilog part
+ */
+
/* How linking shader inputs and outputs between vertex, tessellation, and
* geometry shaders works.
*
/* TES */
SI_SGPR_TES_OFFCHIP_LAYOUT = SI_NUM_RESOURCE_SGPRS,
+ SI_SGPR_TES_OFFCHIP_ADDR_BASE64K,
SI_TES_NUM_USER_SGPR,
/* GFX6-8: TCS only */
GFX6_SGPR_TCS_OUT_OFFSETS,
GFX6_SGPR_TCS_OUT_LAYOUT,
GFX6_SGPR_TCS_IN_LAYOUT,
+ GFX6_SGPR_TCS_OFFCHIP_ADDR_BASE64K,
+ GFX6_SGPR_TCS_FACTOR_ADDR_BASE64K,
GFX6_TCS_NUM_USER_SGPR,
/* GFX9: Merged LS-HS (VS-TCS) only. */
GFX9_SGPR_TCS_OFFCHIP_LAYOUT = SI_VS_NUM_USER_SGPR,
GFX9_SGPR_TCS_OUT_OFFSETS,
GFX9_SGPR_TCS_OUT_LAYOUT,
+ GFX9_SGPR_TCS_OFFCHIP_ADDR_BASE64K,
+ GFX9_SGPR_TCS_FACTOR_ADDR_BASE64K,
GFX9_SGPR_unused_to_align_the_next_pointer,
GFX9_SGPR_TCS_CONST_BUFFERS,
GFX9_SGPR_TCS_CONST_BUFFERS_HI,
/* PS only */
SI_SGPR_ALPHA_REF = SI_NUM_RESOURCE_SGPRS,
SI_PS_NUM_USER_SGPR,
-
- /* CS only */
- SI_SGPR_GRID_SIZE = SI_NUM_RESOURCE_SGPRS,
- SI_SGPR_BLOCK_SIZE = SI_SGPR_GRID_SIZE + 3,
- SI_CS_NUM_USER_SGPR = SI_SGPR_BLOCK_SIZE + 3
};
/* LLVM function parameter indices */
SI_PARAM_SAMPLE_COVERAGE,
SI_PARAM_POS_FIXED_PT,
- /* CS only parameters */
- SI_PARAM_GRID_SIZE = SI_NUM_RESOURCE_PARAMS,
- SI_PARAM_BLOCK_SIZE,
- SI_PARAM_BLOCK_ID,
- SI_PARAM_THREAD_ID,
-
SI_NUM_PARAMS = SI_PARAM_POS_FIXED_PT + 9, /* +8 for COLOR[0..1] */
};
* binaries for one TGSI program. This can be shared by multiple contexts.
*/
struct si_shader_selector {
+ struct pipe_reference reference;
struct si_screen *screen;
struct util_queue_fence ready;
struct si_compiler_ctx_state compiler_ctx_state;
unsigned instance_divisors[SI_MAX_ATTRIBS];
};
-/* Common VS bits between the shader key and the epilog key. */
-struct si_vs_epilog_bits {
- unsigned export_prim_id:1; /* when PS needs it and GS is disabled */
-};
-
/* Common TCS bits between the shader key and the epilog key. */
struct si_tcs_epilog_bits {
unsigned prim_mode:3;
/* For merged stages such as LS-HS, HS input VGPRs are first. */
unsigned num_merged_next_stage_vgprs:3;
unsigned last_input:4;
+ unsigned as_ls:1;
/* Prologs for monolithic shaders shouldn't set EXEC. */
unsigned is_monolithic:1;
} vs_prolog;
- struct {
- struct si_vs_epilog_bits states;
- unsigned prim_id_param_offset:5;
- } vs_epilog;
struct {
struct si_tcs_epilog_bits states;
} tcs_epilog;
struct {
struct si_gs_prolog_bits states;
+ /* Prologs of monolithic shaders shouldn't set EXEC. */
+ unsigned is_monolithic:1;
} gs_prolog;
struct {
struct si_ps_prolog_bits states;
union {
struct {
struct si_vs_prolog_bits prolog;
- struct si_vs_epilog_bits epilog;
} vs;
struct {
struct si_vs_prolog_bits ls_prolog; /* for merged LS-HS */
struct si_tcs_epilog_bits epilog;
} tcs; /* tessellation control shader */
struct {
- struct si_vs_epilog_bits epilog; /* same as VS */
- } tes; /* tessellation evaluation shader */
- struct {
+ struct si_vs_prolog_bits vs_prolog; /* for merged ES-GS */
+ struct si_shader_selector *es; /* for merged ES-GS */
struct si_gs_prolog_bits prolog;
} gs;
struct {
/* One byte for every input: SI_FIX_FETCH_* enums. */
uint8_t vs_fix_fetch[SI_MAX_ATTRIBS];
uint64_t ff_tcs_inputs_to_copy; /* for fixed-func TCS */
+ /* When PS needs PrimID and GS is disabled. */
+ unsigned vs_export_prim_id:1;
} mono;
/* Optimization flags for asynchronous compilation only. */
- union {
+ struct {
struct {
uint64_t kill_outputs; /* "get_unique_index" bits */
uint32_t kill_outputs2; /* "get_unique_index2" bits */
unsigned clip_disable:1;
} hw_vs; /* HW VS (it can be VS, TES, GS) */
+
+ /* For shaders where monolithic variants have better code.
+ *
+ * This is a flag that has no effect on code generation,
+ * but forces monolithic shaders to be used as soon as
+ * possible, because it's in the "opt" group.
+ */
+ unsigned prefer_mono:1;
} opt;
};
struct si_compiler_ctx_state compiler_ctx_state;
struct si_shader_selector *selector;
+ struct si_shader_selector *previous_stage_sel; /* for refcounting */
struct si_shader *next_variant;
struct si_shader_part *prolog;
int si_shader_create(struct si_screen *sscreen, LLVMTargetMachineRef tm,
struct si_shader *shader,
struct pipe_debug_callback *debug);
-int si_compile_llvm(struct si_screen *sscreen,
- struct ac_shader_binary *binary,
- struct si_shader_config *conf,
- LLVMTargetMachineRef tm,
- LLVMModuleRef mod,
- struct pipe_debug_callback *debug,
- unsigned processor,
- const char *name);
void si_shader_destroy(struct si_shader *shader);
+unsigned si_shader_io_get_unique_index_patch(unsigned semantic_name, unsigned index);
unsigned si_shader_io_get_unique_index(unsigned semantic_name, unsigned index);
unsigned si_shader_io_get_unique_index2(unsigned name, unsigned index);
int si_shader_binary_upload(struct si_screen *sscreen, struct si_shader *shader);
FILE *f, bool check_debug_option);
void si_multiwave_lds_size_workaround(struct si_screen *sscreen,
unsigned *lds_size);
-void si_shader_apply_scratch_relocs(struct si_context *sctx,
- struct si_shader *shader,
- struct si_shader_config *config,
- uint64_t scratch_va);
+void si_shader_apply_scratch_relocs(struct si_shader *shader,
+ uint64_t scratch_va);
void si_shader_binary_read_config(struct ac_shader_binary *binary,
struct si_shader_config *conf,
unsigned symbol_offset);
projects / mesa.git / blobdiff