radeonsi: kill point size VS output if it's not used by the rasterizer
[mesa.git] / src / gallium / drivers / radeonsi / si_state.h
index c4ba2767d0d74b32961142ff78a0c42013f548b9..4d42a40d5171277f7cfe167cc93c610ea1ba9cbb 100644 (file)
 #ifndef SI_STATE_H
 #define SI_STATE_H
 
-#include "si_pm4.h"
-#include "radeon/r600_pipe_common.h"
-
 #include "pipebuffer/pb_slab.h"
+#include "si_pm4.h"
+#include "util/u_blitter.h"
 
-#define SI_NUM_GRAPHICS_SHADERS (PIPE_SHADER_TESS_EVAL+1)
-#define SI_NUM_SHADERS (PIPE_SHADER_COMPUTE+1)
+#define SI_NUM_GRAPHICS_SHADERS (PIPE_SHADER_TESS_EVAL + 1)
+#define SI_NUM_SHADERS          (PIPE_SHADER_COMPUTE + 1)
 
-#define SI_MAX_ATTRIBS                 16
-#define SI_NUM_VERTEX_BUFFERS          SI_MAX_ATTRIBS
-#define SI_NUM_SAMPLERS                        32 /* OpenGL textures units per shader */
-#define SI_NUM_CONST_BUFFERS           16
-#define SI_NUM_IMAGES                  16
-#define SI_NUM_SHADER_BUFFERS          16
+#define SI_NUM_VERTEX_BUFFERS SI_MAX_ATTRIBS
+#define SI_NUM_SAMPLERS       32 /* OpenGL textures units per shader */
+#define SI_NUM_CONST_BUFFERS  16
+#define SI_NUM_IMAGES         16
+#define SI_NUM_IMAGE_SLOTS    (SI_NUM_IMAGES * 2) /* the second half are FMASK slots */
+#define SI_NUM_SHADER_BUFFERS 32
 
 struct si_screen;
 struct si_shader;
+struct si_shader_ctx_state;
 struct si_shader_selector;
+struct si_texture;
+struct si_qbo_state;
 
 struct si_state_blend {
-       struct si_pm4_state     pm4;
-       uint32_t                cb_target_mask;
-       /* Set 0xf or 0x0 (4 bits) per render target if the following is
-        * true. ANDed with spi_shader_col_format.
-        */
-       unsigned                cb_target_enabled_4bit;
-       unsigned                blend_enable_4bit;
-       unsigned                need_src_alpha_4bit;
-       unsigned                commutative_4bit;
-       bool                    alpha_to_coverage:1;
-       bool                    alpha_to_one:1;
-       bool                    dual_src_blend:1;
-       bool                    logicop_enable:1;
+   struct si_pm4_state pm4;
+   uint32_t cb_target_mask;
+   /* Set 0xf or 0x0 (4 bits) per render target if the following is
+    * true. ANDed with spi_shader_col_format.
+    */
+   unsigned cb_target_enabled_4bit;
+   unsigned blend_enable_4bit;
+   unsigned need_src_alpha_4bit;
+   unsigned commutative_4bit;
+   unsigned dcc_msaa_corruption_4bit;
+   bool alpha_to_coverage : 1;
+   bool alpha_to_one : 1;
+   bool dual_src_blend : 1;
+   bool logicop_enable : 1;
 };
 
 struct si_state_rasterizer {
-       struct si_pm4_state     pm4;
-       /* poly offset states for 16-bit, 24-bit, and 32-bit zbuffers */
-       struct si_pm4_state     *pm4_poly_offset;
-       unsigned                pa_sc_line_stipple;
-       unsigned                pa_cl_clip_cntl;
-       float                   line_width;
-       float                   max_point_size;
-       unsigned                sprite_coord_enable:8;
-       unsigned                clip_plane_enable:8;
-       unsigned                flatshade:1;
-       unsigned                two_side:1;
-       unsigned                multisample_enable:1;
-       unsigned                force_persample_interp:1;
-       unsigned                line_stipple_enable:1;
-       unsigned                poly_stipple_enable:1;
-       unsigned                line_smooth:1;
-       unsigned                poly_smooth:1;
-       unsigned                uses_poly_offset:1;
-       unsigned                clamp_fragment_color:1;
-       unsigned                clamp_vertex_color:1;
-       unsigned                rasterizer_discard:1;
-       unsigned                scissor_enable:1;
-       unsigned                clip_halfz:1;
+   struct si_pm4_state pm4;
+   /* poly offset states for 16-bit, 24-bit, and 32-bit zbuffers */
+   struct si_pm4_state *pm4_poly_offset;
+   unsigned pa_sc_line_stipple;
+   unsigned pa_cl_clip_cntl;
+   float line_width;
+   float max_point_size;
+   unsigned sprite_coord_enable : 8;
+   unsigned clip_plane_enable : 8;
+   unsigned half_pixel_center : 1;
+   unsigned flatshade : 1;
+   unsigned flatshade_first : 1;
+   unsigned two_side : 1;
+   unsigned multisample_enable : 1;
+   unsigned force_persample_interp : 1;
+   unsigned line_stipple_enable : 1;
+   unsigned poly_stipple_enable : 1;
+   unsigned line_smooth : 1;
+   unsigned poly_smooth : 1;
+   unsigned uses_poly_offset : 1;
+   unsigned clamp_fragment_color : 1;
+   unsigned clamp_vertex_color : 1;
+   unsigned rasterizer_discard : 1;
+   unsigned scissor_enable : 1;
+   unsigned clip_halfz : 1;
+   unsigned cull_front : 1;
+   unsigned cull_back : 1;
+   unsigned depth_clamp_any : 1;
+   unsigned provoking_vertex_first : 1;
+   unsigned polygon_mode_enabled : 1;
+   unsigned polygon_mode_is_lines : 1;
+   unsigned polygon_mode_is_points : 1;
 };
 
 struct si_dsa_stencil_ref_part {
-       uint8_t                 valuemask[2];
-       uint8_t                 writemask[2];
+   uint8_t valuemask[2];
+   uint8_t writemask[2];
 };
 
 struct si_dsa_order_invariance {
-       /** Whether the final result in Z/S buffers is guaranteed to be
-        * invariant under changes to the order in which fragments arrive. */
-       bool zs:1;
-
-       /** Whether the set of fragments that pass the combined Z/S test is
-        * guaranteed to be invariant under changes to the order in which
-        * fragments arrive. */
-       bool pass_set:1;
-
-       /** Whether the last fragment that passes the combined Z/S test at each
-        * sample is guaranteed to be invariant under changes to the order in
-        * which fragments arrive. */
-       bool pass_last:1;
+   /** Whether the final result in Z/S buffers is guaranteed to be
+    * invariant under changes to the order in which fragments arrive. */
+   bool zs : 1;
+
+   /** Whether the set of fragments that pass the combined Z/S test is
+    * guaranteed to be invariant under changes to the order in which
+    * fragments arrive. */
+   bool pass_set : 1;
+
+   /** Whether the last fragment that passes the combined Z/S test at each
+    * sample is guaranteed to be invariant under changes to the order in
+    * which fragments arrive. */
+   bool pass_last : 1;
 };
 
 struct si_state_dsa {
-       struct si_pm4_state             pm4;
-       struct si_dsa_stencil_ref_part  stencil_ref;
-
-       /* 0 = without stencil buffer, 1 = when both Z and S buffers are present */
-       struct si_dsa_order_invariance  order_invariance[2];
-
-       ubyte                           alpha_func:3;
-       bool                            depth_enabled:1;
-       bool                            depth_write_enabled:1;
-       bool                            stencil_enabled:1;
-       bool                            stencil_write_enabled:1;
-       bool                            db_can_write:1;
-
+   struct si_pm4_state pm4;
+   struct si_dsa_stencil_ref_part stencil_ref;
+
+   /* 0 = without stencil buffer, 1 = when both Z and S buffers are present */
+   struct si_dsa_order_invariance order_invariance[2];
+
+   ubyte alpha_func : 3;
+   bool depth_enabled : 1;
+   bool depth_write_enabled : 1;
+   bool stencil_enabled : 1;
+   bool stencil_write_enabled : 1;
+   bool db_can_write : 1;
 };
 
 struct si_stencil_ref {
-       struct r600_atom                atom;
-       struct pipe_stencil_ref         state;
-       struct si_dsa_stencil_ref_part  dsa_part;
+   struct pipe_stencil_ref state;
+   struct si_dsa_stencil_ref_part dsa_part;
 };
 
-struct si_vertex_elements
-{
-       uint32_t                        instance_divisors[SI_MAX_ATTRIBS];
-       uint32_t                        rsrc_word3[SI_MAX_ATTRIBS];
-       uint16_t                        src_offset[SI_MAX_ATTRIBS];
-       uint8_t                         fix_fetch[SI_MAX_ATTRIBS];
-       uint8_t                         format_size[SI_MAX_ATTRIBS];
-       uint8_t                         vertex_buffer_index[SI_MAX_ATTRIBS];
-
-       uint8_t                         count;
-       bool                            uses_instance_divisors;
-
-       uint16_t                        first_vb_use_mask;
-       /* Vertex buffer descriptor list size aligned for optimal prefetch. */
-       uint16_t                        desc_list_byte_size;
-       uint16_t                        instance_divisor_is_one; /* bitmask of inputs */
-       uint16_t                        instance_divisor_is_fetched;  /* bitmask of inputs */
+struct si_vertex_elements {
+   struct si_resource *instance_divisor_factor_buffer;
+   uint32_t rsrc_word3[SI_MAX_ATTRIBS];
+   uint16_t src_offset[SI_MAX_ATTRIBS];
+   uint8_t fix_fetch[SI_MAX_ATTRIBS];
+   uint8_t format_size[SI_MAX_ATTRIBS];
+   uint8_t vertex_buffer_index[SI_MAX_ATTRIBS];
+
+   /* Bitmask of elements that always need a fixup to be applied. */
+   uint16_t fix_fetch_always;
+
+   /* Bitmask of elements whose fetch should always be opencoded. */
+   uint16_t fix_fetch_opencode;
+
+   /* Bitmask of elements which need to be opencoded if the vertex buffer
+    * is unaligned. */
+   uint16_t fix_fetch_unaligned;
+
+   /* For elements in fix_fetch_unaligned: whether the effective
+    * element load size as seen by the hardware is a dword (as opposed
+    * to a short).
+    */
+   uint16_t hw_load_is_dword;
+
+   /* Bitmask of vertex buffers requiring alignment check */
+   uint16_t vb_alignment_check_mask;
+
+   uint8_t count;
+   bool uses_instance_divisors;
+
+   uint16_t first_vb_use_mask;
+   /* Vertex buffer descriptor list size aligned for optimal prefetch. */
+   uint16_t vb_desc_list_alloc_size;
+   uint16_t instance_divisor_is_one;     /* bitmask of inputs */
+   uint16_t instance_divisor_is_fetched; /* bitmask of inputs */
 };
 
 union si_state {
-       struct {
-               struct si_state_blend           *blend;
-               struct si_state_rasterizer      *rasterizer;
-               struct si_state_dsa             *dsa;
-               struct si_pm4_state             *poly_offset;
-               struct si_pm4_state             *ls;
-               struct si_pm4_state             *hs;
-               struct si_pm4_state             *es;
-               struct si_pm4_state             *gs;
-               struct si_pm4_state             *vgt_shader_config;
-               struct si_pm4_state             *vs;
-               struct si_pm4_state             *ps;
-       } named;
-       struct si_pm4_state     *array[0];
+   struct si_state_named {
+      struct si_state_blend *blend;
+      struct si_state_rasterizer *rasterizer;
+      struct si_state_dsa *dsa;
+      struct si_pm4_state *poly_offset;
+      struct si_pm4_state *ls;
+      struct si_pm4_state *hs;
+      struct si_pm4_state *es;
+      struct si_pm4_state *gs;
+      struct si_pm4_state *vgt_shader_config;
+      struct si_pm4_state *vs;
+      struct si_pm4_state *ps;
+   } named;
+   struct si_pm4_state *array[sizeof(struct si_state_named) / sizeof(struct si_pm4_state *)];
 };
 
-#define SI_NUM_STATES (sizeof(union si_state) / sizeof(struct si_pm4_state *))
+#define SI_STATE_IDX(name) (offsetof(union si_state, named.name) / sizeof(struct si_pm4_state *))
+#define SI_STATE_BIT(name) (1 << SI_STATE_IDX(name))
+#define SI_NUM_STATES      (sizeof(union si_state) / sizeof(struct si_pm4_state *))
+
+static inline unsigned si_states_that_always_roll_context(void)
+{
+   return (SI_STATE_BIT(blend) | SI_STATE_BIT(rasterizer) | SI_STATE_BIT(dsa) |
+           SI_STATE_BIT(poly_offset) | SI_STATE_BIT(vgt_shader_config));
+}
 
 union si_state_atoms {
-       struct {
-               /* The order matters. */
-               struct r600_atom *render_cond;
-               struct r600_atom *streamout_begin;
-               struct r600_atom *streamout_enable; /* must be after streamout_begin */
-               struct r600_atom *framebuffer;
-               struct r600_atom *msaa_sample_locs;
-               struct r600_atom *db_render_state;
-               struct r600_atom *dpbb_state;
-               struct r600_atom *msaa_config;
-               struct r600_atom *sample_mask;
-               struct r600_atom *cb_render_state;
-               struct r600_atom *blend_color;
-               struct r600_atom *clip_regs;
-               struct r600_atom *clip_state;
-               struct r600_atom *shader_pointers;
-               struct r600_atom *scissors;
-               struct r600_atom *viewports;
-               struct r600_atom *stencil_ref;
-               struct r600_atom *spi_map;
-               struct r600_atom *scratch_state;
-       } s;
-       struct r600_atom *array[0];
+   struct si_atoms_s {
+      /* The order matters. */
+      struct si_atom render_cond;
+      struct si_atom streamout_begin;
+      struct si_atom streamout_enable; /* must be after streamout_begin */
+      struct si_atom framebuffer;
+      struct si_atom msaa_sample_locs;
+      struct si_atom db_render_state;
+      struct si_atom dpbb_state;
+      struct si_atom msaa_config;
+      struct si_atom sample_mask;
+      struct si_atom cb_render_state;
+      struct si_atom blend_color;
+      struct si_atom clip_regs;
+      struct si_atom clip_state;
+      struct si_atom shader_pointers;
+      struct si_atom guardband;
+      struct si_atom scissors;
+      struct si_atom viewports;
+      struct si_atom stencil_ref;
+      struct si_atom spi_map;
+      struct si_atom scratch_state;
+      struct si_atom window_rectangles;
+      struct si_atom shader_query;
+   } s;
+   struct si_atom array[sizeof(struct si_atoms_s) / sizeof(struct si_atom)];
 };
 
-#define SI_NUM_ATOMS (sizeof(union si_state_atoms)/sizeof(struct r600_atom*))
+#define SI_ATOM_BIT(name) (1 << (offsetof(union si_state_atoms, s.name) / sizeof(struct si_atom)))
+#define SI_NUM_ATOMS      (sizeof(union si_state_atoms) / sizeof(struct si_atom))
+
+static inline unsigned si_atoms_that_always_roll_context(void)
+{
+   return (SI_ATOM_BIT(streamout_begin) | SI_ATOM_BIT(streamout_enable) | SI_ATOM_BIT(framebuffer) |
+           SI_ATOM_BIT(msaa_sample_locs) | SI_ATOM_BIT(sample_mask) | SI_ATOM_BIT(blend_color) |
+           SI_ATOM_BIT(clip_state) | SI_ATOM_BIT(scissors) | SI_ATOM_BIT(viewports) |
+           SI_ATOM_BIT(stencil_ref) | SI_ATOM_BIT(scratch_state) | SI_ATOM_BIT(window_rectangles));
+}
 
 struct si_shader_data {
-       struct r600_atom        atom;
-       uint32_t                sh_base[SI_NUM_SHADERS];
+   uint32_t sh_base[SI_NUM_SHADERS];
+};
+
+#define SI_TRACKED_PA_CL_VS_OUT_CNTL__VS_MASK                                                      \
+   (S_02881C_USE_VTX_POINT_SIZE(1) | S_02881C_USE_VTX_EDGE_FLAG(1) |                               \
+    S_02881C_USE_VTX_RENDER_TARGET_INDX(1) | S_02881C_USE_VTX_VIEWPORT_INDX(1) |                   \
+    S_02881C_VS_OUT_MISC_VEC_ENA(1) | S_02881C_VS_OUT_MISC_SIDE_BUS_ENA(1))
+
+/* The list of registers whose emitted values are remembered by si_context. */
+enum si_tracked_reg
+{
+   SI_TRACKED_DB_RENDER_CONTROL, /* 2 consecutive registers */
+   SI_TRACKED_DB_COUNT_CONTROL,
+
+   SI_TRACKED_DB_RENDER_OVERRIDE2,
+   SI_TRACKED_DB_SHADER_CONTROL,
+
+   SI_TRACKED_CB_TARGET_MASK,
+   SI_TRACKED_CB_DCC_CONTROL,
+
+   SI_TRACKED_SX_PS_DOWNCONVERT, /* 3 consecutive registers */
+   SI_TRACKED_SX_BLEND_OPT_EPSILON,
+   SI_TRACKED_SX_BLEND_OPT_CONTROL,
+
+   SI_TRACKED_PA_SC_LINE_CNTL, /* 2 consecutive registers */
+   SI_TRACKED_PA_SC_AA_CONFIG,
+
+   SI_TRACKED_DB_EQAA,
+   SI_TRACKED_PA_SC_MODE_CNTL_1,
+
+   SI_TRACKED_PA_SU_PRIM_FILTER_CNTL,
+   SI_TRACKED_PA_SU_SMALL_PRIM_FILTER_CNTL,
+
+   SI_TRACKED_PA_CL_VS_OUT_CNTL__VS, /* set with SI_TRACKED_PA_CL_VS_OUT_CNTL__VS_MASK*/
+   SI_TRACKED_PA_CL_VS_OUT_CNTL__CL, /* set with ~SI_TRACKED_PA_CL_VS_OUT_CNTL__VS_MASK */
+   SI_TRACKED_PA_CL_CLIP_CNTL,
+
+   SI_TRACKED_PA_SC_BINNER_CNTL_0,
+   SI_TRACKED_DB_DFSM_CONTROL,
+
+   SI_TRACKED_PA_CL_GB_VERT_CLIP_ADJ, /* 4 consecutive registers */
+   SI_TRACKED_PA_CL_GB_VERT_DISC_ADJ,
+   SI_TRACKED_PA_CL_GB_HORZ_CLIP_ADJ,
+   SI_TRACKED_PA_CL_GB_HORZ_DISC_ADJ,
+
+   SI_TRACKED_PA_SU_HARDWARE_SCREEN_OFFSET,
+   SI_TRACKED_PA_SU_VTX_CNTL,
+
+   SI_TRACKED_PA_SC_CLIPRECT_RULE,
+
+   SI_TRACKED_PA_SC_LINE_STIPPLE,
+
+   SI_TRACKED_VGT_ESGS_RING_ITEMSIZE,
+
+   SI_TRACKED_VGT_GSVS_RING_OFFSET_1, /* 3 consecutive registers */
+   SI_TRACKED_VGT_GSVS_RING_OFFSET_2,
+   SI_TRACKED_VGT_GSVS_RING_OFFSET_3,
+
+   SI_TRACKED_VGT_GSVS_RING_ITEMSIZE,
+   SI_TRACKED_VGT_GS_MAX_VERT_OUT,
+
+   SI_TRACKED_VGT_GS_VERT_ITEMSIZE, /* 4 consecutive registers */
+   SI_TRACKED_VGT_GS_VERT_ITEMSIZE_1,
+   SI_TRACKED_VGT_GS_VERT_ITEMSIZE_2,
+   SI_TRACKED_VGT_GS_VERT_ITEMSIZE_3,
+
+   SI_TRACKED_VGT_GS_INSTANCE_CNT,
+   SI_TRACKED_VGT_GS_ONCHIP_CNTL,
+   SI_TRACKED_VGT_GS_MAX_PRIMS_PER_SUBGROUP,
+   SI_TRACKED_VGT_GS_MODE,
+   SI_TRACKED_VGT_PRIMITIVEID_EN,
+   SI_TRACKED_VGT_REUSE_OFF,
+   SI_TRACKED_SPI_VS_OUT_CONFIG,
+   SI_TRACKED_PA_CL_VTE_CNTL,
+   SI_TRACKED_PA_CL_NGG_CNTL,
+   SI_TRACKED_GE_MAX_OUTPUT_PER_SUBGROUP,
+   SI_TRACKED_GE_NGG_SUBGRP_CNTL,
+
+   SI_TRACKED_SPI_SHADER_IDX_FORMAT, /* 2 consecutive registers */
+   SI_TRACKED_SPI_SHADER_POS_FORMAT,
+
+   SI_TRACKED_SPI_PS_INPUT_ENA, /* 2 consecutive registers */
+   SI_TRACKED_SPI_PS_INPUT_ADDR,
+
+   SI_TRACKED_SPI_BARYC_CNTL,
+   SI_TRACKED_SPI_PS_IN_CONTROL,
+
+   SI_TRACKED_SPI_SHADER_Z_FORMAT, /* 2 consecutive registers */
+   SI_TRACKED_SPI_SHADER_COL_FORMAT,
+
+   SI_TRACKED_CB_SHADER_MASK,
+   SI_TRACKED_VGT_TF_PARAM,
+   SI_TRACKED_VGT_VERTEX_REUSE_BLOCK_CNTL,
+
+   SI_TRACKED_GE_PC_ALLOC,
+
+   SI_NUM_TRACKED_REGS,
+};
+
+struct si_tracked_regs {
+   uint64_t reg_saved;
+   uint32_t reg_value[SI_NUM_TRACKED_REGS];
+   uint32_t spi_ps_input_cntl[32];
 };
 
 /* Private read-write buffer slots. */
-enum {
-       SI_ES_RING_ESGS,
-       SI_GS_RING_ESGS,
-
-       SI_RING_GSVS,
-
-       SI_VS_STREAMOUT_BUF0,
-       SI_VS_STREAMOUT_BUF1,
-       SI_VS_STREAMOUT_BUF2,
-       SI_VS_STREAMOUT_BUF3,
-
-       SI_HS_CONST_DEFAULT_TESS_LEVELS,
-       SI_VS_CONST_INSTANCE_DIVISORS,
-       SI_VS_CONST_CLIP_PLANES,
-       SI_PS_CONST_POLY_STIPPLE,
-       SI_PS_CONST_SAMPLE_POSITIONS,
-
-       /* Image descriptor of color buffer 0 for KHR_blend_equation_advanced. */
-       SI_PS_IMAGE_COLORBUF0,
-       SI_PS_IMAGE_COLORBUF0_HI,
-       SI_PS_IMAGE_COLORBUF0_FMASK,
-       SI_PS_IMAGE_COLORBUF0_FMASK_HI,
-
-       SI_NUM_RW_BUFFERS,
+enum
+{
+   SI_ES_RING_ESGS,
+   SI_GS_RING_ESGS,
+
+   SI_RING_GSVS,
+
+   SI_VS_STREAMOUT_BUF0,
+   SI_VS_STREAMOUT_BUF1,
+   SI_VS_STREAMOUT_BUF2,
+   SI_VS_STREAMOUT_BUF3,
+
+   SI_HS_CONST_DEFAULT_TESS_LEVELS,
+   SI_VS_CONST_INSTANCE_DIVISORS,
+   SI_VS_CONST_CLIP_PLANES,
+   SI_PS_CONST_POLY_STIPPLE,
+   SI_PS_CONST_SAMPLE_POSITIONS,
+
+   /* Image descriptor of color buffer 0 for KHR_blend_equation_advanced. */
+   SI_PS_IMAGE_COLORBUF0,
+   SI_PS_IMAGE_COLORBUF0_HI,
+   SI_PS_IMAGE_COLORBUF0_FMASK,
+   SI_PS_IMAGE_COLORBUF0_FMASK_HI,
+
+   GFX10_GS_QUERY_BUF,
+
+   SI_NUM_RW_BUFFERS,
 };
 
 /* Indices into sctx->descriptors, laid out so that gfx and compute pipelines
@@ -238,247 +390,241 @@ enum {
  *  11 - compute const and shader buffers
  *  12 - compute samplers and images
  */
-enum {
-       SI_SHADER_DESCS_CONST_AND_SHADER_BUFFERS,
-       SI_SHADER_DESCS_SAMPLERS_AND_IMAGES,
-       SI_NUM_SHADER_DESCS,
+enum
+{
+   SI_SHADER_DESCS_CONST_AND_SHADER_BUFFERS,
+   SI_SHADER_DESCS_SAMPLERS_AND_IMAGES,
+   SI_NUM_SHADER_DESCS,
 };
 
-#define SI_DESCS_RW_BUFFERS            0
-#define SI_DESCS_FIRST_SHADER          1
-#define SI_DESCS_FIRST_COMPUTE         (SI_DESCS_FIRST_SHADER + \
-                                        PIPE_SHADER_COMPUTE * SI_NUM_SHADER_DESCS)
-#define SI_NUM_DESCS                   (SI_DESCS_FIRST_SHADER + \
-                                        SI_NUM_SHADERS * SI_NUM_SHADER_DESCS)
+#define SI_DESCS_RW_BUFFERS    0
+#define SI_DESCS_FIRST_SHADER  1
+#define SI_DESCS_FIRST_COMPUTE (SI_DESCS_FIRST_SHADER + PIPE_SHADER_COMPUTE * SI_NUM_SHADER_DESCS)
+#define SI_NUM_DESCS           (SI_DESCS_FIRST_SHADER + SI_NUM_SHADERS * SI_NUM_SHADER_DESCS)
+
+#define SI_DESCS_SHADER_MASK(name)                                                                 \
+   u_bit_consecutive(SI_DESCS_FIRST_SHADER + PIPE_SHADER_##name * SI_NUM_SHADER_DESCS,             \
+                     SI_NUM_SHADER_DESCS)
+
+static inline unsigned si_const_and_shader_buffer_descriptors_idx(unsigned shader)
+{
+   return SI_DESCS_FIRST_SHADER + shader * SI_NUM_SHADER_DESCS +
+          SI_SHADER_DESCS_CONST_AND_SHADER_BUFFERS;
+}
 
-#define SI_DESCS_SHADER_MASK(name) \
-       u_bit_consecutive(SI_DESCS_FIRST_SHADER + \
-                         PIPE_SHADER_##name * SI_NUM_SHADER_DESCS, \
-                         SI_NUM_SHADER_DESCS)
+static inline unsigned si_sampler_and_image_descriptors_idx(unsigned shader)
+{
+   return SI_DESCS_FIRST_SHADER + shader * SI_NUM_SHADER_DESCS +
+          SI_SHADER_DESCS_SAMPLERS_AND_IMAGES;
+}
 
 /* This represents descriptors in memory, such as buffer resources,
  * image resources, and sampler states.
  */
 struct si_descriptors {
-       /* The list of descriptors in malloc'd memory. */
-       uint32_t *list;
-       /* The list in mapped GPU memory. */
-       uint32_t *gpu_list;
-
-       /* The buffer where the descriptors have been uploaded. */
-       struct r600_resource *buffer;
-       uint64_t gpu_address;
-
-       /* The maximum number of descriptors. */
-       uint32_t num_elements;
-
-       /* Slots that are used by currently-bound shaders.
-        * It determines which slots are uploaded.
-        */
-       uint32_t first_active_slot;
-       uint32_t num_active_slots;
-
-       /* The SH register offset relative to USER_DATA*_0 where the pointer
-        * to the descriptor array will be stored. */
-       short shader_userdata_offset;
-       /* The size of one descriptor. */
-       ubyte element_dw_size;
-       /* If there is only one slot enabled, bind it directly instead of
-        * uploading descriptors. -1 if disabled. */
-       signed char slot_index_to_bind_directly;
+   /* The list of descriptors in malloc'd memory. */
+   uint32_t *list;
+   /* The list in mapped GPU memory. */
+   uint32_t *gpu_list;
+
+   /* The buffer where the descriptors have been uploaded. */
+   struct si_resource *buffer;
+   uint64_t gpu_address;
+
+   /* The maximum number of descriptors. */
+   uint32_t num_elements;
+
+   /* Slots that are used by currently-bound shaders.
+    * It determines which slots are uploaded.
+    */
+   uint32_t first_active_slot;
+   uint32_t num_active_slots;
+
+   /* The SH register offset relative to USER_DATA*_0 where the pointer
+    * to the descriptor array will be stored. */
+   short shader_userdata_offset;
+   /* The size of one descriptor. */
+   ubyte element_dw_size;
+   /* If there is only one slot enabled, bind it directly instead of
+    * uploading descriptors. -1 if disabled. */
+   signed char slot_index_to_bind_directly;
 };
 
 struct si_buffer_resources {
-       struct pipe_resource            **buffers; /* this has num_buffers elements */
+   struct pipe_resource **buffers; /* this has num_buffers elements */
+   unsigned *offsets;              /* this has num_buffers elements */
 
-       enum radeon_bo_usage            shader_usage:4; /* READ, WRITE, or READWRITE */
-       enum radeon_bo_usage            shader_usage_constbuf:4;
-       enum radeon_bo_priority         priority:6;
-       enum radeon_bo_priority         priority_constbuf:6;
+   enum radeon_bo_priority priority : 6;
+   enum radeon_bo_priority priority_constbuf : 6;
 
-       /* The i-th bit is set if that element is enabled (non-NULL resource). */
-       unsigned                        enabled_mask;
+   /* The i-th bit is set if that element is enabled (non-NULL resource). */
+   uint64_t enabled_mask;
+   uint64_t writable_mask;
 };
 
-#define si_pm4_block_idx(member) \
-       (offsetof(union si_state, named.member) / sizeof(struct si_pm4_state *))
-
-#define si_pm4_state_changed(sctx, member) \
-       ((sctx)->queued.named.member != (sctx)->emitted.named.member)
+#define si_pm4_state_changed(sctx, member)                                                         \
+   ((sctx)->queued.named.member != (sctx)->emitted.named.member)
 
-#define si_pm4_state_enabled_and_changed(sctx, member) \
-       ((sctx)->queued.named.member && si_pm4_state_changed(sctx, member))
+#define si_pm4_state_enabled_and_changed(sctx, member)                                             \
+   ((sctx)->queued.named.member && si_pm4_state_changed(sctx, member))
 
-#define si_pm4_bind_state(sctx, member, value) \
-       do { \
-               (sctx)->queued.named.member = (value); \
-               (sctx)->dirty_states |= 1 << si_pm4_block_idx(member); \
-       } while(0)
+#define si_pm4_bind_state(sctx, member, value)                                                     \
+   do {                                                                                            \
+      (sctx)->queued.named.member = (value);                                                       \
+      (sctx)->dirty_states |= SI_STATE_BIT(member);                                                \
+   } while (0)
 
-#define si_pm4_delete_state(sctx, member, value) \
-       do { \
-               if ((sctx)->queued.named.member == (value)) { \
-                       (sctx)->queued.named.member = NULL; \
-               } \
-               si_pm4_free_state(sctx, (struct si_pm4_state *)(value), \
-                                 si_pm4_block_idx(member)); \
-       } while(0)
+#define si_pm4_delete_state(sctx, member, value)                                                   \
+   do {                                                                                            \
+      if ((sctx)->queued.named.member == (value)) {                                                \
+         (sctx)->queued.named.member = NULL;                                                       \
+      }                                                                                            \
+      si_pm4_free_state(sctx, (struct si_pm4_state *)(value), SI_STATE_IDX(member));               \
+   } while (0)
 
 /* si_descriptors.c */
-void si_set_mutable_tex_desc_fields(struct si_screen *sscreen,
-                                   struct r600_texture *tex,
-                                   const struct legacy_surf_level *base_level_info,
-                                   unsigned base_level, unsigned first_level,
-                                   unsigned block_width, bool is_stencil,
-                                   uint32_t *state);
+void si_set_mutable_tex_desc_fields(struct si_screen *sscreen, struct si_texture *tex,
+                                    const struct legacy_surf_level *base_level_info,
+                                    unsigned base_level, unsigned first_level, unsigned block_width,
+                                    bool is_stencil, bool force_dcc_off, uint32_t *state);
 void si_update_ps_colorbuf0_slot(struct si_context *sctx);
-void si_get_pipe_constant_buffer(struct si_context *sctx, uint shader,
-                                uint slot, struct pipe_constant_buffer *cbuf);
-void si_get_shader_buffers(struct si_context *sctx,
-                          enum pipe_shader_type shader,
-                          uint start_slot, uint count,
-                          struct pipe_shader_buffer *sbuf);
-void si_set_ring_buffer(struct si_context *sctx, uint slot,
-                       struct pipe_resource *buffer,
-                       unsigned stride, unsigned num_records,
-                       bool add_tid, bool swizzle,
-                       unsigned element_size, unsigned index_stride, uint64_t offset);
+void si_get_pipe_constant_buffer(struct si_context *sctx, uint shader, uint slot,
+                                 struct pipe_constant_buffer *cbuf);
+void si_get_shader_buffers(struct si_context *sctx, enum pipe_shader_type shader, uint start_slot,
+                           uint count, struct pipe_shader_buffer *sbuf);
+void si_set_ring_buffer(struct si_context *sctx, uint slot, struct pipe_resource *buffer,
+                        unsigned stride, unsigned num_records, bool add_tid, bool swizzle,
+                        unsigned element_size, unsigned index_stride, uint64_t offset);
 void si_init_all_descriptors(struct si_context *sctx);
 bool si_upload_vertex_buffer_descriptors(struct si_context *sctx);
 bool si_upload_graphics_shader_descriptors(struct si_context *sctx);
 bool si_upload_compute_shader_descriptors(struct si_context *sctx);
 void si_release_all_descriptors(struct si_context *sctx);
-void si_all_descriptors_begin_new_cs(struct si_context *sctx);
-void si_all_resident_buffers_begin_new_cs(struct si_context *sctx);
-void si_upload_const_buffer(struct si_context *sctx, struct r600_resource **rbuffer,
-                           const uint8_t *ptr, unsigned size, uint32_t *const_offset);
+void si_gfx_resources_add_all_to_bo_list(struct si_context *sctx);
+void si_compute_resources_add_all_to_bo_list(struct si_context *sctx);
+bool si_gfx_resources_check_encrypted(struct si_context *sctx);
+bool si_compute_resources_check_encrypted(struct si_context *sctx);
+void si_shader_pointers_mark_dirty(struct si_context *sctx);
+void si_add_all_descriptors_to_bo_list(struct si_context *sctx);
+void si_upload_const_buffer(struct si_context *sctx, struct si_resource **buf, const uint8_t *ptr,
+                            unsigned size, uint32_t *const_offset);
 void si_update_all_texture_descriptors(struct si_context *sctx);
 void si_shader_change_notify(struct si_context *sctx);
 void si_update_needs_color_decompress_masks(struct si_context *sctx);
-void si_emit_graphics_shader_pointers(struct si_context *sctx,
-                                      struct r600_atom *atom);
+void si_emit_graphics_shader_pointers(struct si_context *sctx);
 void si_emit_compute_shader_pointers(struct si_context *sctx);
-void si_set_rw_buffer(struct si_context *sctx,
-                     uint slot, const struct pipe_constant_buffer *input);
+void si_set_rw_buffer(struct si_context *sctx, uint slot, const struct pipe_constant_buffer *input);
+void si_set_rw_shader_buffer(struct si_context *sctx, uint slot,
+                             const struct pipe_shader_buffer *sbuffer);
 void si_set_active_descriptors(struct si_context *sctx, unsigned desc_idx,
-                              uint64_t new_active_mask);
-void si_set_active_descriptors_for_shader(struct si_context *sctx,
-                                         struct si_shader_selector *sel);
-bool si_bindless_descriptor_can_reclaim_slab(void *priv,
-                                            struct pb_slab_entry *entry);
-struct pb_slab *si_bindless_descriptor_slab_alloc(void *priv, unsigned heap,
-                                                 unsigned entry_size,
-                                                 unsigned group_index);
+                               uint64_t new_active_mask);
+void si_set_active_descriptors_for_shader(struct si_context *sctx, struct si_shader_selector *sel);
+bool si_bindless_descriptor_can_reclaim_slab(void *priv, struct pb_slab_entry *entry);
+struct pb_slab *si_bindless_descriptor_slab_alloc(void *priv, unsigned heap, unsigned entry_size,
+                                                  unsigned group_index);
 void si_bindless_descriptor_slab_free(void *priv, struct pb_slab *pslab);
-void si_rebind_buffer(struct si_context *sctx, struct pipe_resource *buf,
-                     uint64_t old_va);
+void si_rebind_buffer(struct si_context *sctx, struct pipe_resource *buf);
 /* si_state.c */
-struct si_shader_selector;
-
-void si_init_atom(struct si_context *sctx, struct r600_atom *atom,
-                 struct r600_atom **list_elem,
-                 void (*emit_func)(struct si_context *ctx, struct r600_atom *state));
+void si_init_state_compute_functions(struct si_context *sctx);
 void si_init_state_functions(struct si_context *sctx);
 void si_init_screen_state_functions(struct si_screen *sscreen);
-void
-si_make_buffer_descriptor(struct si_screen *screen, struct r600_resource *buf,
-                         enum pipe_format format,
-                         unsigned offset, unsigned size,
-                         uint32_t *state);
-void
-si_make_texture_descriptor(struct si_screen *screen,
-                          struct r600_texture *tex,
-                          bool sampler,
-                          enum pipe_texture_target target,
-                          enum pipe_format pipe_format,
-                          const unsigned char state_swizzle[4],
-                          unsigned first_level, unsigned last_level,
-                          unsigned first_layer, unsigned last_layer,
-                          unsigned width, unsigned height, unsigned depth,
-                          uint32_t *state,
-                          uint32_t *fmask_state);
-struct pipe_sampler_view *
-si_create_sampler_view_custom(struct pipe_context *ctx,
-                             struct pipe_resource *texture,
-                             const struct pipe_sampler_view *state,
-                             unsigned width0, unsigned height0,
-                             unsigned force_level);
+void si_init_cs_preamble_state(struct si_context *sctx, bool uses_reg_shadowing);
+void si_make_buffer_descriptor(struct si_screen *screen, struct si_resource *buf,
+                               enum pipe_format format, unsigned offset, unsigned size,
+                               uint32_t *state);
+struct pipe_sampler_view *si_create_sampler_view_custom(struct pipe_context *ctx,
+                                                        struct pipe_resource *texture,
+                                                        const struct pipe_sampler_view *state,
+                                                        unsigned width0, unsigned height0,
+                                                        unsigned force_level);
 void si_update_fb_dirtiness_after_rendering(struct si_context *sctx);
 void si_update_ps_iter_samples(struct si_context *sctx);
-void si_save_qbo_state(struct si_context *sctx, struct r600_qbo_state *st);
-void si_set_occlusion_query_state(struct si_context *sctx,
-                                 bool old_perfect_enable);
+void si_save_qbo_state(struct si_context *sctx, struct si_qbo_state *st);
+void si_restore_qbo_state(struct si_context *sctx, struct si_qbo_state *st);
+void si_set_occlusion_query_state(struct si_context *sctx, bool old_perfect_enable);
+
+struct si_fast_udiv_info32 {
+   unsigned multiplier; /* the "magic number" multiplier */
+   unsigned pre_shift;  /* shift for the dividend before multiplying */
+   unsigned post_shift; /* shift for the dividend after multiplying */
+   int increment;       /* 0 or 1; if set then increment the numerator, using one of
+                           the two strategies */
+};
+
+struct si_fast_udiv_info32 si_compute_fast_udiv_info32(uint32_t D, unsigned num_bits);
 
 /* si_state_binning.c */
-void si_emit_dpbb_state(struct si_context *sctx, struct r600_atom *state);
+void si_emit_dpbb_state(struct si_context *sctx);
 
 /* si_state_shaders.c */
+void si_get_ir_cache_key(struct si_shader_selector *sel, bool ngg, bool es,
+                         unsigned char ir_sha1_cache_key[20]);
+bool si_shader_cache_load_shader(struct si_screen *sscreen, unsigned char ir_sha1_cache_key[20],
+                                 struct si_shader *shader);
+void si_shader_cache_insert_shader(struct si_screen *sscreen, unsigned char ir_sha1_cache_key[20],
+                                   struct si_shader *shader, bool insert_into_disk_cache);
 bool si_update_shaders(struct si_context *sctx);
+void si_init_screen_live_shader_cache(struct si_screen *sscreen);
 void si_init_shader_functions(struct si_context *sctx);
 bool si_init_shader_cache(struct si_screen *sscreen);
 void si_destroy_shader_cache(struct si_screen *sscreen);
-void si_get_active_slot_masks(const struct tgsi_shader_info *info,
-                             uint32_t *const_and_shader_buffers,
-                             uint64_t *samplers_and_images);
-void *si_get_blit_vs(struct si_context *sctx, enum blitter_attrib_type type,
-                    unsigned num_layers);
+void si_schedule_initial_compile(struct si_context *sctx, gl_shader_stage stage,
+                                 struct util_queue_fence *ready_fence,
+                                 struct si_compiler_ctx_state *compiler_ctx_state, void *job,
+                                 util_queue_execute_func execute);
+void si_get_active_slot_masks(const struct si_shader_info *info, uint64_t *const_and_shader_buffers,
+                              uint64_t *samplers_and_images);
+int si_shader_select_with_key(struct si_screen *sscreen, struct si_shader_ctx_state *state,
+                              struct si_compiler_ctx_state *compiler_state,
+                              struct si_shader_key *key, int thread_index, bool optimized_or_none);
+void si_shader_selector_key_vs(struct si_context *sctx, struct si_shader_selector *vs,
+                               struct si_shader_key *key, struct si_vs_prolog_bits *prolog_key);
+unsigned si_get_input_prim(const struct si_shader_selector *gs);
+bool si_update_ngg(struct si_context *sctx);
 
 /* si_state_draw.c */
-void si_init_ia_multi_vgt_param_table(struct si_context *sctx);
+void si_emit_surface_sync(struct si_context *sctx, struct radeon_cmdbuf *cs,
+                          unsigned cp_coher_cntl);
+void si_prim_discard_signal_next_compute_ib_start(struct si_context *sctx);
+void gfx10_emit_cache_flush(struct si_context *sctx);
 void si_emit_cache_flush(struct si_context *sctx);
-void si_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *dinfo);
-void si_draw_rectangle(struct blitter_context *blitter,
-                      void *vertex_elements_cso,
-                      blitter_get_vs_func get_vs,
-                      int x1, int y1, int x2, int y2,
-                      float depth, unsigned num_instances,
-                      enum blitter_attrib_type type,
-                      const union blitter_attrib *attrib);
 void si_trace_emit(struct si_context *sctx);
+void si_init_draw_functions(struct si_context *sctx);
 
 /* si_state_msaa.c */
 void si_init_msaa_functions(struct si_context *sctx);
-void si_emit_sample_locations(struct radeon_winsys_cs *cs, int nr_samples);
+void si_emit_sample_locations(struct radeon_cmdbuf *cs, int nr_samples);
 
 /* si_state_streamout.c */
 void si_streamout_buffers_dirty(struct si_context *sctx);
 void si_emit_streamout_end(struct si_context *sctx);
-void si_update_prims_generated_query_state(struct si_context *sctx,
-                                          unsigned type, int diff);
+void si_update_prims_generated_query_state(struct si_context *sctx, unsigned type, int diff);
 void si_init_streamout_functions(struct si_context *sctx);
 
-
-static inline unsigned
-si_tile_mode_index(struct r600_texture *rtex, unsigned level, bool stencil)
-{
-       if (stencil)
-               return rtex->surface.u.legacy.stencil_tiling_index[level];
-       else
-               return rtex->surface.u.legacy.tiling_index[level];
-}
-
 static inline unsigned si_get_constbuf_slot(unsigned slot)
 {
-       /* Constant buffers are in slots [16..31], ascending */
-       return SI_NUM_SHADER_BUFFERS + slot;
+   /* Constant buffers are in slots [32..47], ascending */
+   return SI_NUM_SHADER_BUFFERS + slot;
 }
 
 static inline unsigned si_get_shaderbuf_slot(unsigned slot)
 {
-       /* shader buffers are in slots [15..0], descending */
-       return SI_NUM_SHADER_BUFFERS - 1 - slot;
+   /* shader buffers are in slots [31..0], descending */
+   return SI_NUM_SHADER_BUFFERS - 1 - slot;
 }
 
 static inline unsigned si_get_sampler_slot(unsigned slot)
 {
-       /* samplers are in slots [8..39], ascending */
-       return SI_NUM_IMAGES / 2 + slot;
+   /* 32 samplers are in sampler slots [16..47], 16 dw per slot, ascending */
+   /* those are equivalent to image slots [32..95], 8 dw per slot, ascending  */
+   return SI_NUM_IMAGE_SLOTS / 2 + slot;
 }
 
 static inline unsigned si_get_image_slot(unsigned slot)
 {
-       /* images are in slots [15..0] (sampler slots [7..0]), descending */
-       return SI_NUM_IMAGES - 1 - slot;
+   /* image slots are in [31..0] (sampler slots [15..0]), descending */
+   /* images are in slots [31..16], while FMASKs are in slots [15..0] */
+   return SI_NUM_IMAGE_SLOTS - 1 - slot;
 }
 
 #endif