diff options
| author | Chris Wilson <chris@chris-wilson.co.uk> | 2012-12-06 17:18:52 +0000 |
|---|---|---|
| committer | Chris Wilson <chris@chris-wilson.co.uk> | 2012-12-06 19:52:10 +0000 |
| commit | 1945e2a7f18dc2e4c46010ab8b7774679fc217d3 (patch) | |
| tree | 603e149c44acfa3cffcf4f9ab86821481213619a | |
| parent | 3ad92902e1ed51894ddece2ae2a698ff55ac8adf (diff) | |
Implement a working gen7 rendercopy
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=57825
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
| -rw-r--r-- | lib/gen7_render.h | 1496 | ||||
| -rw-r--r-- | lib/intel_batchbuffer.h | 1 | ||||
| -rw-r--r-- | lib/rendercopy_gen7.c | 941 |
3 files changed, 1674 insertions, 764 deletions
diff --git a/lib/gen7_render.h b/lib/gen7_render.h index 6cd9475..1661d4c 100644 --- a/lib/gen7_render.h +++ b/lib/gen7_render.h @@ -1,222 +1,1364 @@ #ifndef GEN7_RENDER_H #define GEN7_RENDER_H -#include "gen6_render.h" - -#define GEN7_3DSTATE_URB_VS (0x7830 << 16) -#define GEN7_3DSTATE_URB_HS (0x7831 << 16) -#define GEN7_3DSTATE_URB_DS (0x7832 << 16) -#define GEN7_3DSTATE_URB_GS (0x7833 << 16) - -#define GEN6_3DSTATE_SCISSOR_STATE_POINTERS GEN6_3D(3, 0, 0xf) -#define GEN7_3DSTATE_CLEAR_PARAMS GEN6_3D(3, 0, 0x04) -#define GEN7_3DSTATE_DEPTH_BUFFER GEN6_3D(3, 0, 0x05) -#define GEN7_3DSTATE_STENCIL_BUFFER GEN6_3D(3, 0, 0x06) -#define GEN7_3DSTATE_HIER_DEPTH_BUFFER GEN6_3D(3, 0, 0x07) - -#define GEN7_3DSTATE_GS GEN6_3D(3, 0, 0x11) -#define GEN7_3DSTATE_CONSTANT_GS GEN6_3D(3, 0, 0x16) -#define GEN7_3DSTATE_CONSTANT_HS GEN6_3D(3, 0, 0x19) -#define GEN7_3DSTATE_CONSTANT_DS GEN6_3D(3, 0, 0x1a) -#define GEN7_3DSTATE_HS GEN6_3D(3, 0, 0x1b) -#define GEN7_3DSTATE_TE GEN6_3D(3, 0, 0x1c) -#define GEN7_3DSTATE_DS GEN6_3D(3, 0, 0x1d) -#define GEN7_3DSTATE_STREAMOUT GEN6_3D(3, 0, 0x1e) -#define GEN7_3DSTATE_SBE GEN6_3D(3, 0, 0x1f) -#define GEN7_3DSTATE_PS GEN6_3D(3, 0, 0x20) -#define GEN7_3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP \ - GEN6_3D(3, 0, 0x21) -#define GEN7_3DSTATE_VIEWPORT_STATE_POINTERS_CC GEN6_3D(3, 0, 0x23) -#define GEN7_3DSTATE_BLEND_STATE_POINTERS GEN6_3D(3, 0, 0x24) -#define GEN7_3DSTATE_DS_STATE_POINTERS GEN6_3D(3, 0, 0x25) -#define GEN7_3DSTATE_BINDING_TABLE_POINTERS_VS GEN6_3D(3, 0, 0x26) -#define GEN7_3DSTATE_BINDING_TABLE_POINTERS_HS GEN6_3D(3, 0, 0x27) -#define GEN7_3DSTATE_BINDING_TABLE_POINTERS_DS GEN6_3D(3, 0, 0x28) -#define GEN7_3DSTATE_BINDING_TABLE_POINTERS_GS GEN6_3D(3, 0, 0x29) -#define GEN7_3DSTATE_BINDING_TABLE_POINTERS_PS GEN6_3D(3, 0, 0x2a) - -#define GEN7_3DSTATE_SAMPLER_STATE_POINTERS_VS GEN6_3D(3, 0, 0x2b) -#define GEN7_3DSTATE_SAMPLER_STATE_POINTERS_HS GEN6_3D(3, 0, 0x2c) -#define GEN7_3DSTATE_SAMPLER_STATE_POINTERS_DS GEN6_3D(3, 0, 0x2d) -#define GEN7_3DSTATE_SAMPLER_STATE_POINTERS_GS GEN6_3D(3, 0, 0x2e) -#define GEN7_3DSTATE_SAMPLER_STATE_POINTERS_PS GEN6_3D(3, 0, 0x2f) - -#define GEN7_3DSTATE_PUSH_CONSTANT_ALLOC_VS GEN6_3D(3, 1, 0x12) -#define GEN7_3DSTATE_PUSH_CONSTANT_ALLOC_HS GEN6_3D(3, 1, 0x13) -#define GEN7_3DSTATE_PUSH_CONSTANT_ALLOC_DS GEN6_3D(3, 1, 0x14) -#define GEN7_3DSTATE_PUSH_CONSTANT_ALLOC_GS GEN6_3D(3, 1, 0x15) -#define GEN7_3DSTATE_PUSH_CONSTANT_ALLOC_PS GEN6_3D(3, 1, 0x16) - -/* Some random bits that we care about */ -#define GEN7_VB0_BUFFER_ADDR_MOD_EN (1 << 14) -#define GEN7_WM_DISPATCH_ENABLE (1 << 29) -#define GEN7_3DSTATE_PS_PERSPECTIVE_PIXEL_BARYCENTRIC (1 << 11) -#define GEN7_3DSTATE_PS_ATTRIBUTE_ENABLED (1 << 10) - -/* Random shifts */ -#define GEN7_3DSTATE_WM_MAX_THREADS_SHIFT 24 -#define HSW_3DSTATE_WM_MAX_THREADS_SHIFT 23 - -/* Shamelessly ripped from mesa */ -struct gen7_surface_state -{ - struct { - uint32_t cube_pos_z:1; - uint32_t cube_neg_z:1; - uint32_t cube_pos_y:1; - uint32_t cube_neg_y:1; - uint32_t cube_pos_x:1; - uint32_t cube_neg_x:1; - uint32_t pad2:2; - uint32_t render_cache_read_write:1; +#define INTEL_MASK(high, low) (((1 << ((high) - (low) + 1)) - 1) << (low)) + +#define GEN7_3D(Pipeline,Opcode,Subopcode) ((3 << 29) | \ + ((Pipeline) << 27) | \ + ((Opcode) << 24) | \ + ((Subopcode) << 16)) + +#define GEN7_STATE_BASE_ADDRESS GEN7_3D(0, 1, 1) +#define GEN7_STATE_SIP GEN7_3D(0, 1, 2) + +#define GEN7_PIPELINE_SELECT GEN7_3D(1, 1, 4) + +#define GEN7_MEDIA_STATE_POINTERS GEN7_3D(2, 0, 0) +#define GEN7_MEDIA_OBJECT GEN7_3D(2, 1, 0) + +#define GEN7_3DSTATE_VERTEX_BUFFERS GEN7_3D(3, 0, 8) +#define GEN7_3DSTATE_VERTEX_ELEMENTS GEN7_3D(3, 0, 9) +#define GEN7_3DSTATE_INDEX_BUFFER GEN7_3D(3, 0, 0xa) +#define GEN7_3DSTATE_VF_STATISTICS GEN7_3D(3, 0, 0xb) + +#define GEN7_3DSTATE_DRAWING_RECTANGLE GEN7_3D(3, 1, 0) +#define GEN7_3DSTATE_CONSTANT_COLOR GEN7_3D(3, 1, 1) +#define GEN7_3DSTATE_SAMPLER_PALETTE_LOAD GEN7_3D(3, 1, 2) +#define GEN7_3DSTATE_CHROMA_KEY GEN7_3D(3, 1, 4) + +#define GEN7_3DSTATE_POLY_STIPPLE_OFFSET GEN7_3D(3, 1, 6) +#define GEN7_3DSTATE_POLY_STIPPLE_PATTERN GEN7_3D(3, 1, 7) +#define GEN7_3DSTATE_LINE_STIPPLE GEN7_3D(3, 1, 8) +#define GEN7_3DSTATE_GLOBAL_DEPTH_OFFSET_CLAMP GEN7_3D(3, 1, 9) +/* These two are BLC and CTG only, not BW or CL */ +#define GEN7_3DSTATE_AA_LINE_PARAMS GEN7_3D(3, 1, 0xa) +#define GEN7_3DSTATE_GS_SVB_INDEX GEN7_3D(3, 1, 0xb) + +#define GEN7_3DPRIMITIVE GEN7_3D(3, 3, 0) + +#define GEN7_3DSTATE_SAMPLER_STATE_POINTERS GEN7_3D(3, 0, 0x02) +# define GEN7_3DSTATE_SAMPLER_STATE_MODIFY_PS (1 << 12) +# define GEN7_3DSTATE_SAMPLER_STATE_MODIFY_GS (1 << 9) +# define GEN7_3DSTATE_SAMPLER_STATE_MODIFY_VS (1 << 8) + +#define GEN7_3DSTATE_URB GEN7_3D(3, 0, 0x05) +/* DW1 */ +# define GEN7_3DSTATE_URB_VS_SIZE_SHIFT 16 +# define GEN7_3DSTATE_URB_VS_ENTRIES_SHIFT 0 +/* DW2 */ +# define GEN7_3DSTATE_URB_GS_ENTRIES_SHIFT 8 +# define GEN7_3DSTATE_URB_GS_SIZE_SHIFT 0 + +#define GEN7_3DSTATE_VIEWPORT_STATE_POINTERS GEN7_3D(3, 0, 0x0d) +# define GEN7_3DSTATE_VIEWPORT_STATE_MODIFY_CC (1 << 12) +# define GEN7_3DSTATE_VIEWPORT_STATE_MODIFY_SF (1 << 11) +# define GEN7_3DSTATE_VIEWPORT_STATE_MODIFY_CLIP (1 << 10) + +#define GEN7_3DSTATE_CC_STATE_POINTERS GEN7_3D(3, 0, 0x0e) + +#define GEN7_3DSTATE_VS GEN7_3D(3, 0, 0x10) + +#define GEN7_3DSTATE_GS GEN7_3D(3, 0, 0x11) +/* DW4 */ +# define GEN7_3DSTATE_GS_DISPATCH_START_GRF_SHIFT 0 + +#define GEN7_3DSTATE_CLIP GEN7_3D(3, 0, 0x12) + +#define GEN7_3DSTATE_SF GEN7_3D(3, 0, 0x13) +/* DW1 */ +# define GEN7_3DSTATE_SF_NUM_OUTPUTS_SHIFT 22 +# define GEN7_3DSTATE_SF_URB_ENTRY_READ_LENGTH_SHIFT 11 +# define GEN7_3DSTATE_SF_URB_ENTRY_READ_OFFSET_SHIFT 4 +/* DW2 */ +/* DW3 */ +# define GEN7_3DSTATE_SF_CULL_BOTH (0 << 29) +# define GEN7_3DSTATE_SF_CULL_NONE (1 << 29) +# define GEN7_3DSTATE_SF_CULL_FRONT (2 << 29) +# define GEN7_3DSTATE_SF_CULL_BACK (3 << 29) +/* DW4 */ +# define GEN7_3DSTATE_SF_TRI_PROVOKE_SHIFT 29 +# define GEN7_3DSTATE_SF_LINE_PROVOKE_SHIFT 27 +# define GEN7_3DSTATE_SF_TRIFAN_PROVOKE_SHIFT 25 + +#define GEN7_3DSTATE_WM GEN7_3D(3, 0, 0x14) +/* DW1 */ +# define GEN7_WM_STATISTICS_ENABLE (1 << 31) +# define GEN7_WM_DEPTH_CLEAR (1 << 30) +# define GEN7_WM_DISPATCH_ENABLE (1 << 29) +# define GEN7_WM_DEPTH_RESOLVE (1 << 28) +# define GEN7_WM_HIERARCHICAL_DEPTH_RESOLVE (1 << 27) +# define GEN7_WM_KILL_ENABLE (1 << 25) +# define GEN7_WM_PSCDEPTH_OFF (0 << 23) +# define GEN7_WM_PSCDEPTH_ON (1 << 23) +# define GEN7_WM_PSCDEPTH_ON_GE (2 << 23) +# define GEN7_WM_PSCDEPTH_ON_LE (3 << 23) +# define GEN7_WM_USES_SOURCE_DEPTH (1 << 20) +# define GEN7_WM_USES_SOURCE_W (1 << 19) +# define GEN7_WM_POSITION_ZW_PIXEL (0 << 17) +# define GEN7_WM_POSITION_ZW_CENTROID (2 << 17) +# define GEN7_WM_POSITION_ZW_SAMPLE (3 << 17) +# define GEN7_WM_NONPERSPECTIVE_SAMPLE_BARYCENTRIC (1 << 16) +# define GEN7_WM_NONPERSPECTIVE_CENTROID_BARYCENTRIC (1 << 15) +# define GEN7_WM_NONPERSPECTIVE_PIXEL_BARYCENTRIC (1 << 14) +# define GEN7_WM_PERSPECTIVE_SAMPLE_BARYCENTRIC (1 << 13) +# define GEN7_WM_PERSPECTIVE_CENTROID_BARYCENTRIC (1 << 12) +# define GEN7_WM_PERSPECTIVE_PIXEL_BARYCENTRIC (1 << 11) +# define GEN7_WM_USES_INPUT_COVERAGE_MASK (1 << 10) +# define GEN7_WM_LINE_END_CAP_AA_WIDTH_0_5 (0 << 8) +# define GEN7_WM_LINE_END_CAP_AA_WIDTH_1_0 (1 << 8) +# define GEN7_WM_LINE_END_CAP_AA_WIDTH_2_0 (2 << 8) +# define GEN7_WM_LINE_END_CAP_AA_WIDTH_4_0 (3 << 8) +# define GEN7_WM_LINE_AA_WIDTH_0_5 (0 << 6) +# define GEN7_WM_LINE_AA_WIDTH_1_0 (1 << 6) +# define GEN7_WM_LINE_AA_WIDTH_2_0 (2 << 6) +# define GEN7_WM_LINE_AA_WIDTH_4_0 (3 << 6) +# define GEN7_WM_POLYGON_STIPPLE_ENABLE (1 << 4) +# define GEN7_WM_LINE_STIPPLE_ENABLE (1 << 3) +# define GEN7_WM_POINT_RASTRULE_UPPER_RIGHT (1 << 2) +# define GEN7_WM_MSRAST_OFF_PIXEL (0 << 0) +# define GEN7_WM_MSRAST_OFF_PATTERN (1 << 0) +# define GEN7_WM_MSRAST_ON_PIXEL (2 << 0) +# define GEN7_WM_MSRAST_ON_PATTERN (3 << 0) +/* DW2 */ +# define GEN7_WM_MSDISPMODE_PERPIXEL (1 << 31) + + +#define GEN7_3DSTATE_CONSTANT_VS GEN7_3D(3, 0, 0x15) +#define GEN7_3DSTATE_CONSTANT_GS GEN7_3D(3, 0, 0x16) +#define GEN7_3DSTATE_CONSTANT_PS GEN7_3D(3, 0, 0x17) + +#define GEN7_3DSTATE_SAMPLE_MASK GEN7_3D(3, 0, 0x18) + +#define GEN7_3DSTATE_MULTISAMPLE GEN7_3D(3, 1, 0x0d) +/* DW1 */ +# define GEN7_3DSTATE_MULTISAMPLE_PIXEL_LOCATION_CENTER (0 << 4) +# define GEN7_3DSTATE_MULTISAMPLE_PIXEL_LOCATION_UPPER_LEFT (1 << 4) +# define GEN7_3DSTATE_MULTISAMPLE_NUMSAMPLES_1 (0 << 1) +# define GEN7_3DSTATE_MULTISAMPLE_NUMSAMPLES_4 (2 << 1) +# define GEN7_3DSTATE_MULTISAMPLE_NUMSAMPLES_8 (3 << 1) + +#define PIPELINE_SELECT_3D 0 +#define PIPELINE_SELECT_MEDIA 1 + +/* for GEN7_STATE_BASE_ADDRESS */ +#define BASE_ADDRESS_MODIFY (1 << 0) + +/* for GEN7_PIPE_CONTROL */ +#define GEN7_PIPE_CONTROL GEN7_3D(3, 2, 0) +#define GEN7_PIPE_CONTROL_CS_STALL (1 << 20) +#define GEN7_PIPE_CONTROL_NOWRITE (0 << 14) +#define GEN7_PIPE_CONTROL_WRITE_QWORD (1 << 14) +#define GEN7_PIPE_CONTROL_WRITE_DEPTH (2 << 14) +#define GEN7_PIPE_CONTROL_WRITE_TIME (3 << 14) +#define GEN7_PIPE_CONTROL_DEPTH_STALL (1 << 13) +#define GEN7_PIPE_CONTROL_WC_FLUSH (1 << 12) +#define GEN7_PIPE_CONTROL_IS_FLUSH (1 << 11) +#define GEN7_PIPE_CONTROL_TC_FLUSH (1 << 10) +#define GEN7_PIPE_CONTROL_NOTIFY_ENABLE (1 << 8) +#define GEN7_PIPE_CONTROL_GLOBAL_GTT (1 << 2) +#define GEN7_PIPE_CONTROL_LOCAL_PGTT (0 << 2) +#define GEN7_PIPE_CONTROL_STALL_AT_SCOREBOARD (1 << 1) +#define GEN7_PIPE_CONTROL_DEPTH_CACHE_FLUSH (1 << 0) + +/* VERTEX_BUFFER_STATE Structure */ +#define GEN7_VB0_BUFFER_INDEX_SHIFT 26 +#define GEN7_VB0_VERTEXDATA (0 << 20) +#define GEN7_VB0_INSTANCEDATA (1 << 20) +#define GEN7_VB0_BUFFER_PITCH_SHIFT 0 +#define GEN7_VB0_ADDRESS_MODIFY_ENABLE (1 << 14) + +/* VERTEX_ELEMENT_STATE Structure */ +#define GEN7_VE0_VERTEX_BUFFER_INDEX_SHIFT 26 +#define GEN7_VE0_VALID (1 << 25) +#define GEN7_VE0_FORMAT_SHIFT 16 +#define GEN7_VE0_OFFSET_SHIFT 0 +#define GEN7_VE1_VFCOMPONENT_0_SHIFT 28 +#define GEN7_VE1_VFCOMPONENT_1_SHIFT 24 +#define GEN7_VE1_VFCOMPONENT_2_SHIFT 20 +#define GEN7_VE1_VFCOMPONENT_3_SHIFT 16 +#define GEN7_VE1_DESTINATION_ELEMENT_OFFSET_SHIFT 0 + +/* 3DPRIMITIVE bits */ +#define GEN7_3DPRIMITIVE_VERTEX_SEQUENTIAL (0 << 15) +#define GEN7_3DPRIMITIVE_VERTEX_RANDOM (1 << 15) + +#define GEN7_SVG_CTL 0x7400 + +#define GEN7_SVG_CTL_GS_BA (0 << 8) +#define GEN7_SVG_CTL_SS_BA (1 << 8) +#define GEN7_SVG_CTL_IO_BA (2 << 8) +#define GEN7_SVG_CTL_GS_AUB (3 << 8) +#define GEN7_SVG_CTL_IO_AUB (4 << 8) +#define GEN7_SVG_CTL_SIP (5 << 8) + +#define GEN7_VF_CTL_SNAPSHOT_COMPLETE (1 << 31) +#define GEN7_VF_CTL_SNAPSHOT_MUX_SELECT_THREADID (0 << 8) +#define GEN7_VF_CTL_SNAPSHOT_MUX_SELECT_VF_DEBUG (1 << 8) +#define GEN7_VF_CTL_SNAPSHOT_TYPE_VERTEX_SEQUENCE (0 << 4) +#define GEN7_VF_CTL_SNAPSHOT_TYPE_VERTEX_INDEX (1 << 4) +#define GEN7_VF_CTL_SKIP_INITIAL_PRIMITIVES (1 << 3) +#define GEN7_VF_CTL_MAX_PRIMITIVES_LIMIT_ENABLE (1 << 2) +#define GEN7_VF_CTL_VERTEX_RANGE_LIMIT_ENABLE (1 << 1) +#define GEN7_VF_CTL_SNAPSHOT_ENABLE (1 << 0) + +#define GEN7_VF_STRG_VAL 0x7504 +#define GEN7_VF_STR_VL_OVR 0x7508 +#define GEN7_VF_VC_OVR 0x750c +#define GEN7_VF_STR_PSKIP 0x7510 +#define GEN7_VF_MAX_PRIM 0x7514 +#define GEN7_VF_RDATA 0x7518 + +#define GEN7_VS_CTL 0x7600 +#define GEN7_VS_CTL_SNAPSHOT_COMPLETE (1 << 31) +#define GEN7_VS_CTL_SNAPSHOT_MUX_VERTEX_0 (0 << 8) +#define GEN7_VS_CTL_SNAPSHOT_MUX_VERTEX_1 (1 << 8) +#define GEN7_VS_CTL_SNAPSHOT_MUX_VALID_COUNT (2 << 8) +#define GEN7_VS_CTL_SNAPSHOT_MUX_VS_KERNEL_POINTER (3 << 8) +#define GEN7_VS_CTL_SNAPSHOT_ALL_THREADS (1 << 2) +#define GEN7_VS_CTL_THREAD_SNAPSHOT_ENABLE (1 << 1) +#define GEN7_VS_CTL_SNAPSHOT_ENABLE (1 << 0) + +#define GEN7_VS_STRG_VAL 0x7604 +#define GEN7_VS_RDATA 0x7608 + +#define GEN7_SF_CTL 0x7b00 +#define GEN7_SF_CTL_SNAPSHOT_COMPLETE (1 << 31) +#define GEN7_SF_CTL_SNAPSHOT_MUX_VERTEX_0_FF_ID (0 << 8) +#define GEN7_SF_CTL_SNAPSHOT_MUX_VERTEX_0_REL_COUNT (1 << 8) +#define GEN7_SF_CTL_SNAPSHOT_MUX_VERTEX_1_FF_ID (2 << 8) +#define GEN7_SF_CTL_SNAPSHOT_MUX_VERTEX_1_REL_COUNT (3 << 8) +#define GEN7_SF_CTL_SNAPSHOT_MUX_VERTEX_2_FF_ID (4 << 8) +#define GEN7_SF_CTL_SNAPSHOT_MUX_VERTEX_2_REL_COUNT (5 << 8) +#define GEN7_SF_CTL_SNAPSHOT_MUX_VERTEX_COUNT (6 << 8) +#define GEN7_SF_CTL_SNAPSHOT_MUX_SF_KERNEL_POINTER (7 << 8) +#define GEN7_SF_CTL_MIN_MAX_PRIMITIVE_RANGE_ENABLE (1 << 4) +#define GEN7_SF_CTL_DEBUG_CLIP_RECTANGLE_ENABLE (1 << 3) +#define GEN7_SF_CTL_SNAPSHOT_ALL_THREADS (1 << 2) +#define GEN7_SF_CTL_THREAD_SNAPSHOT_ENABLE (1 << 1) +#define GEN7_SF_CTL_SNAPSHOT_ENABLE (1 << 0) + +#define GEN7_SF_STRG_VAL 0x7b04 +#define GEN7_SF_RDATA 0x7b18 + +#define GEN7_WIZ_CTL 0x7c00 +#define GEN7_WIZ_CTL_SNAPSHOT_COMPLETE (1 << 31) +#define GEN7_WIZ_CTL_SUBSPAN_INSTANCE_SHIFT 16 +#define GEN7_WIZ_CTL_SNAPSHOT_MUX_WIZ_KERNEL_POINTER (0 << 8) +#define GEN7_WIZ_CTL_SNAPSHOT_MUX_SUBSPAN_INSTANCE (1 << 8) +#define GEN7_WIZ_CTL_SNAPSHOT_MUX_PRIMITIVE_SEQUENCE (2 << 8) +#define GEN7_WIZ_CTL_SINGLE_SUBSPAN_DISPATCH (1 << 6) +#define GEN7_WIZ_CTL_IGNORE_COLOR_SCOREBOARD_STALLS (1 << 5) +#define GEN7_WIZ_CTL_ENABLE_SUBSPAN_INSTANCE_COMPARE (1 << 4) +#define GEN7_WIZ_CTL_USE_UPSTREAM_SNAPSHOT_FLAG (1 << 3) +#define GEN7_WIZ_CTL_SNAPSHOT_ALL_THREADS (1 << 2) +#define GEN7_WIZ_CTL_THREAD_SNAPSHOT_ENABLE (1 << 1) +#define GEN7_WIZ_CTL_SNAPSHOT_ENABLE (1 << 0) + +#define GEN7_WIZ_STRG_VAL 0x7c04 +#define GEN7_WIZ_RDATA 0x7c18 + +#define GEN7_TS_CTL 0x7e00 +#define GEN7_TS_CTL_SNAPSHOT_COMPLETE (1 << 31) +#define GEN7_TS_CTL_SNAPSHOT_MESSAGE_ERROR (0 << 8) +#define GEN7_TS_CTL_SNAPSHOT_INTERFACE_DESCRIPTOR (3 << 8) +#define GEN7_TS_CTL_SNAPSHOT_ALL_CHILD_THREADS (1 << 2) +#define GEN7_TS_CTL_SNAPSHOT_ALL_ROOT_THREADS (1 << 1) +#define GEN7_TS_CTL_SNAPSHOT_ENABLE (1 << 0) + +#define GEN7_TS_STRG_VAL 0x7e04 +#define GEN7_TS_RDATA 0x7e08 + +#define GEN7_TD_CTL 0x8000 +#define GEN7_TD_CTL_MUX_SHIFT 8 +#define GEN7_TD_CTL_EXTERNAL_HALT_R0_DEBUG_MATCH (1 << 7) +#define GEN7_TD_CTL_FORCE_EXTERNAL_HALT (1 << 6) +#define GEN7_TD_CTL_EXCEPTION_MASK_OVERRIDE (1 << 5) +#define GEN7_TD_CTL_FORCE_THREAD_BREAKPOINT_ENABLE (1 << 4) +#define GEN7_TD_CTL_BREAKPOINT_ENABLE (1 << 2) +#define GEN7_TD_CTL2 0x8004 +#define GEN7_TD_CTL2_ILLEGAL_OPCODE_EXCEPTION_OVERRIDE (1 << 28) +#define GEN7_TD_CTL2_MASKSTACK_EXCEPTION_OVERRIDE (1 << 26) +#define GEN7_TD_CTL2_SOFTWARE_EXCEPTION_OVERRIDE (1 << 25) +#define GEN7_TD_CTL2_ACTIVE_THREAD_LIMIT_SHIFT 16 +#define GEN7_TD_CTL2_ACTIVE_THREAD_LIMIT_ENABLE (1 << 8) +#define GEN7_TD_CTL2_THREAD_SPAWNER_EXECUTION_MASK_ENABLE (1 << 7) +#define GEN7_TD_CTL2_WIZ_EXECUTION_MASK_ENABLE (1 << 6) +#define GEN7_TD_CTL2_SF_EXECUTION_MASK_ENABLE (1 << 5) +#define GEN7_TD_CTL2_CLIPPER_EXECUTION_MASK_ENABLE (1 << 4) +#define GEN7_TD_CTL2_GS_EXECUTION_MASK_ENABLE (1 << 3) +#define GEN7_TD_CTL2_VS_EXECUTION_MASK_ENABLE (1 << 0) +#define GEN7_TD_VF_VS_EMSK 0x8008 +#define GEN7_TD_GS_EMSK 0x800c +#define GEN7_TD_CLIP_EMSK 0x8010 +#define GEN7_TD_SF_EMSK 0x8014 +#define GEN7_TD_WIZ_EMSK 0x8018 +#define GEN7_TD_0_6_EHTRG_VAL 0x801c +#define GEN7_TD_0_7_EHTRG_VAL 0x8020 +#define GEN7_TD_0_6_EHTRG_MSK 0x8024 +#define GEN7_TD_0_7_EHTRG_MSK 0x8028 +#define GEN7_TD_RDATA 0x802c +#define GEN7_TD_TS_EMSK 0x8030 + +#define GEN7_EU_CTL 0x8800 +#define GEN7_EU_CTL_SELECT_SHIFT 16 +#define GEN7_EU_CTL_DATA_MUX_SHIFT 8 +#define GEN7_EU_ATT_0 0x8810 +#define GEN7_EU_ATT_1 0x8814 +#define GEN7_EU_ATT_DATA_0 0x8820 +#define GEN7_EU_ATT_DATA_1 0x8824 +#define GEN7_EU_ATT_CLR_0 0x8830 +#define GEN7_EU_ATT_CLR_1 0x8834 +#define GEN7_EU_RDATA 0x8840 + +#define _3DPRIM_POINTLIST 0x01 +#define _3DPRIM_LINELIST 0x02 +#define _3DPRIM_LINESTRIP 0x03 +#define _3DPRIM_TRILIST 0x04 +#define _3DPRIM_TRISTRIP 0x05 +#define _3DPRIM_TRIFAN 0x06 +#define _3DPRIM_QUADLIST 0x07 +#define _3DPRIM_QUADSTRIP 0x08 +#define _3DPRIM_LINELIST_ADJ 0x09 +#define _3DPRIM_LINESTRIP_ADJ 0x0A +#define _3DPRIM_TRILIST_ADJ 0x0B +#define _3DPRIM_TRISTRIP_ADJ 0x0C +#define _3DPRIM_TRISTRIP_REVERSE 0x0D +#define _3DPRIM_POLYGON 0x0E +#define _3DPRIM_RECTLIST 0x0F +#define _3DPRIM_LINELOOP 0x10 +#define _3DPRIM_POINTLIST_BF 0x11 +#define _3DPRIM_LINESTRIP_CONT 0x12 +#define _3DPRIM_LINESTRIP_BF 0x13 +#define _3DPRIM_LINESTRIP_CONT_BF 0x14 +#define _3DPRIM_TRIFAN_NOSTIPPLE 0x15 + +#define _3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL 0 +#define _3DPRIM_VERTEXBUFFER_ACCESS_RANDOM 1 + +#define GEN7_ANISORATIO_2 0 +#define GEN7_ANISORATIO_4 1 +#define GEN7_ANISORATIO_6 2 +#define GEN7_ANISORATIO_8 3 +#define GEN7_ANISORATIO_10 4 +#define GEN7_ANISORATIO_12 5 +#define GEN7_ANISORATIO_14 6 +#define GEN7_ANISORATIO_16 7 + +#define GEN7_BLENDFACTOR_ONE 0x1 +#define GEN7_BLENDFACTOR_SRC_COLOR 0x2 +#define GEN7_BLENDFACTOR_SRC_ALPHA 0x3 +#define GEN7_BLENDFACTOR_DST_ALPHA 0x4 +#define GEN7_BLENDFACTOR_DST_COLOR 0x5 +#define GEN7_BLENDFACTOR_SRC_ALPHA_SATURATE 0x6 +#define GEN7_BLENDFACTOR_CONST_COLOR 0x7 +#define GEN7_BLENDFACTOR_CONST_ALPHA 0x8 +#define GEN7_BLENDFACTOR_SRC1_COLOR 0x9 +#define GEN7_BLENDFACTOR_SRC1_ALPHA 0x0A +#define GEN7_BLENDFACTOR_ZERO 0x11 +#define GEN7_BLENDFACTOR_INV_SRC_COLOR 0x12 +#define GEN7_BLENDFACTOR_INV_SRC_ALPHA 0x13 +#define GEN7_BLENDFACTOR_INV_DST_ALPHA 0x14 +#define GEN7_BLENDFACTOR_INV_DST_COLOR 0x15 +#define GEN7_BLENDFACTOR_INV_CONST_COLOR 0x17 +#define GEN7_BLENDFACTOR_INV_CONST_ALPHA 0x18 +#define GEN7_BLENDFACTOR_INV_SRC1_COLOR 0x19 +#define GEN7_BLENDFACTOR_INV_SRC1_ALPHA 0x1A + +#define GEN7_BLENDFUNCTION_ADD 0 +#define GEN7_BLENDFUNCTION_SUBTRACT 1 +#define GEN7_BLENDFUNCTION_REVERSE_SUBTRACT 2 +#define GEN7_BLENDFUNCTION_MIN 3 +#define GEN7_BLENDFUNCTION_MAX 4 + +#define GEN7_ALPHATEST_FORMAT_UNORM8 0 +#define GEN7_ALPHATEST_FORMAT_FLOAT32 1 + +#define GEN7_CHROMAKEY_KILL_ON_ANY_MATCH 0 +#define GEN7_CHROMAKEY_REPLACE_BLACK 1 + +#define GEN7_CLIP_API_OGL 0 +#define GEN7_CLIP_API_DX 1 + +#define GEN7_CLIPMODE_NORMAL 0 +#define GEN7_CLIPMODE_CLIP_ALL 1 +#define GEN7_CLIPMODE_CLIP_NON_REJECTED 2 +#define GEN7_CLIPMODE_REJECT_ALL 3 +#define GEN7_CLIPMODE_ACCEPT_ALL 4 + +#define GEN7_CLIP_NDCSPACE 0 +#define GEN7_CLIP_SCREENSPACE 1 + +#define GEN7_COMPAREFUNCTION_ALWAYS 0 +#define GEN7_COMPAREFUNCTION_NEVER 1 +#define GEN7_COMPAREFUNCTION_LESS 2 +#define GEN7_COMPAREFUNCTION_EQUAL 3 +#define GEN7_COMPAREFUNCTION_LEQUAL 4 +#define GEN7_COMPAREFUNCTION_GREATER 5 +#define GEN7_COMPAREFUNCTION_NOTEQUAL 6 +#define GEN7_COMPAREFUNCTION_GEQUAL 7 + +#define GEN7_COVERAGE_PIXELS_HALF 0 +#define GEN7_COVERAGE_PIXELS_1 1 +#define GEN7_COVERAGE_PIXELS_2 2 +#define GEN7_COVERAGE_PIXELS_4 3 + +#define GEN7_CULLMODE_BOTH 0 +#define GEN7_CULLMODE_NONE 1 +#define GEN7_CULLMODE_FRONT 2 +#define GEN7_CULLMODE_BACK 3 + +#define GEN7_DEFAULTCOLOR_R8G8B8A8_UNORM 0 +#define GEN7_DEFAULTCOLOR_R32G32B32A32_FLOAT 1 + +#define GEN7_DEPTHFORMAT_D32_FLOAT_S8X24_UINT 0 +#define GEN7_DEPTHFORMAT_D32_FLOAT 1 +#define GEN7_DEPTHFORMAT_D24_UNORM_S8_UINT 2 +#define GEN7_DEPTHFORMAT_D16_UNORM 5 + +#define GEN7_FLOATING_POINT_IEEE_754 0 +#define GEN7_FLOATING_POINT_NON_IEEE_754 1 + +#define GEN7_FRONTWINDING_CW 0 +#define GEN7_FRONTWINDING_CCW 1 + +#define GEN7_INDEX_BYTE 0 +#define GEN7_INDEX_WORD 1 +#define GEN7_INDEX_DWORD 2 + +#define GEN7_LOGICOPFUNCTION_CLEAR 0 +#define GEN7_LOGICOPFUNCTION_NOR 1 +#define GEN7_LOGICOPFUNCTION_AND_INVERTED 2 +#define GEN7_LOGICOPFUNCTION_COPY_INVERTED 3 +#define GEN7_LOGICOPFUNCTION_AND_REVERSE 4 +#define GEN7_LOGICOPFUNCTION_INVERT 5 +#define GEN7_LOGICOPFUNCTION_XOR 6 +#define GEN7_LOGICOPFUNCTION_NAND 7 +#define GEN7_LOGICOPFUNCTION_AND 8 +#define GEN7_LOGICOPFUNCTION_EQUIV 9 +#define GEN7_LOGICOPFUNCTION_NOOP 10 +#define GEN7_LOGICOPFUNCTION_OR_INVERTED 11 +#define GEN7_LOGICOPFUNCTION_COPY 12 +#define GEN7_LOGICOPFUNCTION_OR_REVERSE 13 +#define GEN7_LOGICOPFUNCTION_OR 14 +#define GEN7_LOGICOPFUNCTION_SET 15 + +#define GEN7_MAPFILTER_NEAREST 0x0 +#define GEN7_MAPFILTER_LINEAR 0x1 +#define GEN7_MAPFILTER_ANISOTROPIC 0x2 + +#define GEN7_MIPFILTER_NONE 0 +#define GEN7_MIPFILTER_NEAREST 1 +#define GEN7_MIPFILTER_LINEAR 3 + +#define GEN7_POLYGON_FRONT_FACING 0 +#define GEN7_POLYGON_BACK_FACING 1 + +#define GEN7_PREFILTER_ALWAYS 0x0 +#define GEN7_PREFILTER_NEVER 0x1 +#define GEN7_PREFILTER_LESS 0x2 +#define GEN7_PREFILTER_EQUAL 0x3 +#define GEN7_PREFILTER_LEQUAL 0x4 +#define GEN7_PREFILTER_GREATER 0x5 +#define GEN7_PREFILTER_NOTEQUAL 0x6 +#define GEN7_PREFILTER_GEQUAL 0x7 + +#define GEN7_PROVOKING_VERTEX_0 0 +#define GEN7_PROVOKING_VERTEX_1 1 +#define GEN7_PROVOKING_VERTEX_2 2 + +#define GEN7_RASTRULE_UPPER_LEFT 0 +#define GEN7_RASTRULE_UPPER_RIGHT 1 + +#define GEN7_RENDERTARGET_CLAMPRANGE_UNORM 0 +#define GEN7_RENDERTARGET_CLAMPRANGE_SNORM 1 +#define GEN7_RENDERTARGET_CLAMPRANGE_FORMAT 2 + +#define GEN7_STENCILOP_KEEP 0 +#define GEN7_STENCILOP_ZERO 1 +#define GEN7_STENCILOP_REPLACE 2 +#define GEN7_STENCILOP_INCRSAT 3 +#define GEN7_STENCILOP_DECRSAT 4 +#define GEN7_STENCILOP_INCR 5 +#define GEN7_STENCILOP_DECR 6 +#define GEN7_STENCILOP_INVERT 7 + +#define GEN7_SURFACE_MIPMAPLAYOUT_BELOW 0 +#define GEN7_SURFACE_MIPMAPLAYOUT_RIGHT 1 + +#define GEN7_SURFACEFORMAT_R32G32B32A32_FLOAT 0x000 +#define GEN7_SURFACEFORMAT_R32G32B32A32_SINT 0x001 +#define GEN7_SURFACEFORMAT_R32G32B32A32_UINT 0x002 +#define GEN7_SURFACEFORMAT_R32G32B32A32_UNORM 0x003 +#define GEN7_SURFACEFORMAT_R32G32B32A32_SNORM 0x004 +#define GEN7_SURFACEFORMAT_R64G64_FLOAT 0x005 +#define GEN7_SURFACEFORMAT_R32G32B32X32_FLOAT 0x006 +#define GEN7_SURFACEFORMAT_R32G32B32A32_SSCALED 0x007 +#define GEN7_SURFACEFORMAT_R32G32B32A32_USCALED 0x008 +#define GEN7_SURFACEFORMAT_R32G32B32_FLOAT 0x040 +#define GEN7_SURFACEFORMAT_R32G32B32_SINT 0x041 +#define GEN7_SURFACEFORMAT_R32G32B32_UINT 0x042 +#define GEN7_SURFACEFORMAT_R32G32B32_UNORM 0x043 +#define GEN7_SURFACEFORMAT_R32G32B32_SNORM 0x044 +#define GEN7_SURFACEFORMAT_R32G32B32_SSCALED 0x045 +#define GEN7_SURFACEFORMAT_R32G32B32_USCALED 0x046 +#define GEN7_SURFACEFORMAT_R16G16B16A16_UNORM 0x080 +#define GEN7_SURFACEFORMAT_R16G16B16A16_SNORM 0x081 +#define GEN7_SURFACEFORMAT_R16G16B16A16_SINT 0x082 +#define GEN7_SURFACEFORMAT_R16G16B16A16_UINT 0x083 +#define GEN7_SURFACEFORMAT_R16G16B16A16_FLOAT 0x084 +#define GEN7_SURFACEFORMAT_R32G32_FLOAT 0x085 +#define GEN7_SURFACEFORMAT_R32G32_SINT 0x086 +#define GEN7_SURFACEFORMAT_R32G32_UINT 0x087 +#define GEN7_SURFACEFORMAT_R32_FLOAT_X8X24_TYPELESS 0x088 +#define GEN7_SURFACEFORMAT_X32_TYPELESS_G8X24_UINT 0x089 +#define GEN7_SURFACEFORMAT_L32A32_FLOAT 0x08A +#define GEN7_SURFACEFORMAT_R32G32_UNORM 0x08B +#define GEN7_SURFACEFORMAT_R32G32_SNORM 0x08C +#define GEN7_SURFACEFORMAT_R64_FLOAT 0x08D +#define GEN7_SURFACEFORMAT_R16G16B16X16_UNORM 0x08E +#define GEN7_SURFACEFORMAT_R16G16B16X16_FLOAT 0x08F +#define GEN7_SURFACEFORMAT_A32X32_FLOAT 0x090 +#define GEN7_SURFACEFORMAT_L32X32_FLOAT 0x091 +#define GEN7_SURFACEFORMAT_I32X32_FLOAT 0x092 +#define GEN7_SURFACEFORMAT_R16G16B16A16_SSCALED 0x093 +#define GEN7_SURFACEFORMAT_R16G16B16A16_USCALED 0x094 +#define GEN7_SURFACEFORMAT_R32G32_SSCALED 0x095 +#define GEN7_SURFACEFORMAT_R32G32_USCALED 0x096 +#define GEN7_SURFACEFORMAT_B8G8R8A8_UNORM 0x0C0 +#define GEN7_SURFACEFORMAT_B8G8R8A8_UNORM_SRGB 0x0C1 +#define GEN7_SURFACEFORMAT_R10G10B10A2_UNORM 0x0C2 +#define GEN7_SURFACEFORMAT_R10G10B10A2_UNORM_SRGB 0x0C3 +#define GEN7_SURFACEFORMAT_R10G10B10A2_UINT 0x0C4 +#define GEN7_SURFACEFORMAT_R10G10B10_SNORM_A2_UNORM 0x0C5 +#define GEN7_SURFACEFORMAT_R8G8B8A8_UNORM 0x0C7 +#define GEN7_SURFACEFORMAT_R8G8B8A8_UNORM_SRGB 0x0C8 +#define GEN7_SURFACEFORMAT_R8G8B8A8_SNORM 0x0C9 +#define GEN7_SURFACEFORMAT_R8G8B8A8_SINT 0x0CA +#define GEN7_SURFACEFORMAT_R8G8B8A8_UINT 0x0CB +#define GEN7_SURFACEFORMAT_R16G16_UNORM 0x0CC +#define GEN7_SURFACEFORMAT_R16G16_SNORM 0x0CD +#define GEN7_SURFACEFORMAT_R16G16_SINT 0x0CE +#define GEN7_SURFACEFORMAT_R16G16_UINT 0x0CF +#define GEN7_SURFACEFORMAT_R16G16_FLOAT 0x0D0 +#define GEN7_SURFACEFORMAT_B10G10R10A2_UNORM 0x0D1 +#define GEN7_SURFACEFORMAT_B10G10R10A2_UNORM_SRGB 0x0D2 +#define GEN7_SURFACEFORMAT_R11G11B10_FLOAT 0x0D3 +#define GEN7_SURFACEFORMAT_R32_SINT 0x0D6 +#define GEN7_SURFACEFORMAT_R32_UINT 0x0D7 +#define GEN7_SURFACEFORMAT_R32_FLOAT 0x0D8 +#define GEN7_SURFACEFORMAT_R24_UNORM_X8_TYPELESS 0x0D9 +#define GEN7_SURFACEFORMAT_X24_TYPELESS_G8_UINT 0x0DA +#define GEN7_SURFACEFORMAT_L16A16_UNORM 0x0DF +#define GEN7_SURFACEFORMAT_I24X8_UNORM 0x0E0 +#define GEN7_SURFACEFORMAT_L24X8_UNORM 0x0E1 +#define GEN7_SURFACEFORMAT_A24X8_UNORM 0x0E2 +#define GEN7_SURFACEFORMAT_I32_FLOAT 0x0E3 +#define GEN7_SURFACEFORMAT_L32_FLOAT 0x0E4 +#define GEN7_SURFACEFORMAT_A32_FLOAT 0x0E5 +#define GEN7_SURFACEFORMAT_B8G8R8X8_UNORM 0x0E9 +#define GEN7_SURFACEFORMAT_B8G8R8X8_UNORM_SRGB 0x0EA +#define GEN7_SURFACEFORMAT_R8G8B8X8_UNORM 0x0EB +#define GEN7_SURFACEFORMAT_R8G8B8X8_UNORM_SRGB 0x0EC +#define GEN7_SURFACEFORMAT_R9G9B9E5_SHAREDEXP 0x0ED +#define GEN7_SURFACEFORMAT_B10G10R10X2_UNORM 0x0EE +#define GEN7_SURFACEFORMAT_L16A16_FLOAT 0x0F0 +#define GEN7_SURFACEFORMAT_R32_UNORM 0x0F1 +#define GEN7_SURFACEFORMAT_R32_SNORM 0x0F2 +#define GEN7_SURFACEFORMAT_R10G10B10X2_USCALED 0x0F3 +#define GEN7_SURFACEFORMAT_R8G8B8A8_SSCALED 0x0F4 +#define GEN7_SURFACEFORMAT_R8G8B8A8_USCALED 0x0F5 +#define GEN7_SURFACEFORMAT_R16G16_SSCALED 0x0F6 +#define GEN7_SURFACEFORMAT_R16G16_USCALED 0x0F7 +#define GEN7_SURFACEFORMAT_R32_SSCALED 0x0F8 +#define GEN7_SURFACEFORMAT_R32_USCALED 0x0F9 +#define GEN7_SURFACEFORMAT_B5G6R5_UNORM 0x100 +#define GEN7_SURFACEFORMAT_B5G6R5_UNORM_SRGB 0x101 +#define GEN7_SURFACEFORMAT_B5G5R5A1_UNORM 0x102 +#define GEN7_SURFACEFORMAT_B5G5R5A1_UNORM_SRGB 0x103 +#define GEN7_SURFACEFORMAT_B4G4R4A4_UNORM 0x104 +#define GEN7_SURFACEFORMAT_B4G4R4A4_UNORM_SRGB 0x105 +#define GEN7_SURFACEFORMAT_R8G8_UNORM 0x106 +#define GEN7_SURFACEFORMAT_R8G8_SNORM 0x107 +#define GEN7_SURFACEFORMAT_R8G8_SINT 0x108 +#define GEN7_SURFACEFORMAT_R8G8_UINT 0x109 +#define GEN7_SURFACEFORMAT_R16_UNORM 0x10A +#define GEN7_SURFACEFORMAT_R16_SNORM 0x10B +#define GEN7_SURFACEFORMAT_R16_SINT 0x10C +#define GEN7_SURFACEFORMAT_R16_UINT 0x10D +#define GEN7_SURFACEFORMAT_R16_FLOAT 0x10E +#define GEN7_SURFACEFORMAT_I16_UNORM 0x111 +#define GEN7_SURFACEFORMAT_L16_UNORM 0x112 +#define GEN7_SURFACEFORMAT_A16_UNORM 0x113 +#define GEN7_SURFACEFORMAT_L8A8_UNORM 0x114 +#define GEN7_SURFACEFORMAT_I16_FLOAT 0x115 +#define GEN7_SURFACEFORMAT_L16_FLOAT 0x116 +#define GEN7_SURFACEFORMAT_A16_FLOAT 0x117 +#define GEN7_SURFACEFORMAT_R5G5_SNORM_B6_UNORM 0x119 +#define GEN7_SURFACEFORMAT_B5G5R5X1_UNORM 0x11A +#define GEN7_SURFACEFORMAT_B5G5R5X1_UNORM_SRGB 0x11B +#define GEN7_SURFACEFORMAT_R8G8_SSCALED 0x11C +#define GEN7_SURFACEFORMAT_R8G8_USCALED 0x11D +#define GEN7_SURFACEFORMAT_R16_SSCALED 0x11E +#define GEN7_SURFACEFORMAT_R16_USCALED 0x11F +#define GEN7_SURFACEFORMAT_R8_UNORM 0x140 +#define GEN7_SURFACEFORMAT_R8_SNORM 0x141 +#define GEN7_SURFACEFORMAT_R8_SINT 0x142 +#define GEN7_SURFACEFORMAT_R8_UINT 0x143 +#define GEN7_SURFACEFORMAT_A8_UNORM 0x144 +#define GEN7_SURFACEFORMAT_I8_UNORM 0x145 +#define GEN7_SURFACEFORMAT_L8_UNORM 0x146 +#define GEN7_SURFACEFORMAT_P4A4_UNORM 0x147 +#define GEN7_SURFACEFORMAT_A4P4_UNORM 0x148 +#define GEN7_SURFACEFORMAT_R8_SSCALED 0x149 +#define GEN7_SURFACEFORMAT_R8_USCALED 0x14A +#define GEN7_SURFACEFORMAT_R1_UINT 0x181 +#define GEN7_SURFACEFORMAT_YCRCB_NORMAL 0x182 +#define GEN7_SURFACEFORMAT_YCRCB_SWAPUVY 0x183 +#define GEN7_SURFACEFORMAT_BC1_UNORM 0x186 +#define GEN7_SURFACEFORMAT_BC2_UNORM 0x187 +#define GEN7_SURFACEFORMAT_BC3_UNORM 0x188 +#define GEN7_SURFACEFORMAT_BC4_UNORM 0x189 +#define GEN7_SURFACEFORMAT_BC5_UNORM 0x18A +#define GEN7_SURFACEFORMAT_BC1_UNORM_SRGB 0x18B +#define GEN7_SURFACEFORMAT_BC2_UNORM_SRGB 0x18C +#define GEN7_SURFACEFORMAT_BC3_UNORM_SRGB 0x18D +#define GEN7_SURFACEFORMAT_MONO8 0x18E +#define GEN7_SURFACEFORMAT_YCRCB_SWAPUV 0x18F +#define GEN7_SURFACEFORMAT_YCRCB_SWAPY 0x190 +#define GEN7_SURFACEFORMAT_DXT1_RGB 0x191 +#define GEN7_SURFACEFORMAT_FXT1 0x192 +#define GEN7_SURFACEFORMAT_R8G8B8_UNORM 0x193 +#define GEN7_SURFACEFORMAT_R8G8B8_SNORM 0x194 +#define GEN7_SURFACEFORMAT_R8G8B8_SSCALED 0x195 +#define GEN7_SURFACEFORMAT_R8G8B8_USCALED 0x196 +#define GEN7_SURFACEFORMAT_R64G64B64A64_FLOAT 0x197 +#define GEN7_SURFACEFORMAT_R64G64B64_FLOAT 0x198 +#define GEN7_SURFACEFORMAT_BC4_SNORM 0x199 +#define GEN7_SURFACEFORMAT_BC5_SNORM 0x19A +#define GEN7_SURFACEFORMAT_R16G16B16_UNORM 0x19C +#define GEN7_SURFACEFORMAT_R16G16B16_SNORM 0x19D +#define GEN7_SURFACEFORMAT_R16G16B16_SSCALED 0x19E +#define GEN7_SURFACEFORMAT_R16G16B16_USCALED 0x19F + +#define GEN7_SURFACERETURNFORMAT_FLOAT32 0 +#define GEN7_SURFACERETURNFORMAT_S1 1 + +#define GEN7_SURFACE_1D 0 +#define GEN7_SURFACE_2D 1 +#define GEN7_SURFACE_3D 2 +#define GEN7_SURFACE_CUBE 3 +#define GEN7_SURFACE_BUFFER 4 +#define GEN7_SURFACE_NULL 7 + +#define GEN7_BORDER_COLOR_MODE_DEFAULT 0 +#define GEN7_BORDER_COLOR_MODE_LEGACY 1 + +#define GEN7_TEXCOORDMODE_WRAP 0 +#define GEN7_TEXCOORDMODE_MIRROR 1 +#define GEN7_TEXCOORDMODE_CLAMP 2 +#define GEN7_TEXCOORDMODE_CUBE 3 +#define GEN7_TEXCOORDMODE_CLAMP_BORDER 4 +#define GEN7_TEXCOORDMODE_MIRROR_ONCE 5 + +#define GEN7_THREAD_PRIORITY_NORMAL 0 +#define GEN7_THREAD_PRIORITY_HIGH 1 + +#define GEN7_TILEWALK_XMAJOR 0 +#define GEN7_TILEWALK_YMAJOR 1 + +#define GEN7_VERTEX_SUBPIXEL_PRECISION_8BITS 0 +#define GEN7_VERTEX_SUBPIXEL_PRECISION_4BITS 1 + +#define GEN7_VERTEXBUFFER_ACCESS_VERTEXDATA 0 +#define GEN7_VERTEXBUFFER_ACCESS_INSTANCEDATA 1 + +#define GEN7_VFCOMPONENT_NOSTORE 0 +#define GEN7_VFCOMPONENT_STORE_SRC 1 +#define GEN7_VFCOMPONENT_STORE_0 2 +#define GEN7_VFCOMPONENT_STORE_1_FLT 3 +#define GEN7_VFCOMPONENT_STORE_1_INT 4 +#define GEN7_VFCOMPONENT_STORE_VID 5 +#define GEN7_VFCOMPONENT_STORE_IID 6 +#define GEN7_VFCOMPONENT_STORE_PID 7 + + +/* Execution Unit (EU) defines + */ + +#define GEN7_ALIGN_1 0 +#define GEN7_ALIGN_16 1 + +#define GEN7_ADDRESS_DIRECT 0 +#define GEN7_ADDRESS_REGISTER_INDIRECT_REGISTER 1 + +#define GEN7_CHANNEL_X 0 +#define GEN7_CHANNEL_Y 1 +#define GEN7_CHANNEL_Z 2 +#define GEN7_CHANNEL_W 3 + +#define GEN7_COMPRESSION_NONE 0 +#define GEN7_COMPRESSION_2NDHALF 1 +#define GEN7_COMPRESSION_COMPRESSED 2 + +#define GEN7_CONDITIONAL_NONE 0 +#define GEN7_CONDITIONAL_Z 1 +#define GEN7_CONDITIONAL_NZ 2 +#define GEN7_CONDITIONAL_EQ 1 /* Z */ +#define GEN7_CONDITIONAL_NEQ 2 /* NZ */ +#define GEN7_CONDITIONAL_G 3 +#define GEN7_CONDITIONAL_GE 4 +#define GEN7_CONDITIONAL_L 5 +#define GEN7_CONDITIONAL_LE 6 +#define GEN7_CONDITIONAL_C 7 +#define GEN7_CONDITIONAL_O 8 + +#define GEN7_DEBUG_NONE 0 +#define GEN7_DEBUG_BREAKPOINT 1 + +#define GEN7_DEPENDENCY_NORMAL 0 +#define GEN7_DEPENDENCY_NOTCLEARED 1 +#define GEN7_DEPENDENCY_NOTCHECKED 2 +#define GEN7_DEPENDENCY_DISABLE 3 + +#define GEN7_EXECUTE_1 0 +#define GEN7_EXECUTE_2 1 +#define GEN7_EXECUTE_4 2 +#define GEN7_EXECUTE_8 3 +#define GEN7_EXECUTE_16 4 +#define GEN7_EXECUTE_32 5 + +#define GEN7_HORIZONTAL_STRIDE_0 0 +#define GEN7_HORIZONTAL_STRIDE_1 1 +#define GEN7_HORIZONTAL_STRIDE_2 2 +#define GEN7_HORIZONTAL_STRIDE_4 3 + +#define GEN7_INSTRUCTION_NORMAL 0 +#define GEN7_INSTRUCTION_SATURATE 1 + +#define INTEL_MASK_ENABLE 0 +#define INTEL_MASK_DISABLE 1 + +#define GEN7_OPCODE_MOV 1 +#define GEN7_OPCODE_SEL 2 +#define GEN7_OPCODE_NOT 4 +#define GEN7_OPCODE_AND 5 +#define GEN7_OPCODE_OR 6 +#define GEN7_OPCODE_XOR 7 +#define GEN7_OPCODE_SHR 8 +#define GEN7_OPCODE_SHL 9 +#define GEN7_OPCODE_RSR 10 +#define GEN7_OPCODE_RSL 11 +#define GEN7_OPCODE_ASR 12 +#define GEN7_OPCODE_CMP 16 +#define GEN7_OPCODE_JMPI 32 +#define GEN7_OPCODE_IF 34 +#define GEN7_OPCODE_IFF 35 +#define GEN7_OPCODE_ELSE 36 +#define GEN7_OPCODE_ENDIF 37 +#define GEN7_OPCODE_DO 38 +#define GEN7_OPCODE_WHILE 39 +#define GEN7_OPCODE_BREAK 40 +#define GEN7_OPCODE_CONTINUE 41 +#define GEN7_OPCODE_HALT 42 +#define GEN7_OPCODE_MSAVE 44 +#define GEN7_OPCODE_MRESTORE 45 +#define GEN7_OPCODE_PUSH 46 +#define GEN7_OPCODE_POP 47 +#define GEN7_OPCODE_WAIT 48 +#define GEN7_OPCODE_SEND 49 +#define GEN7_OPCODE_ADD 64 +#define GEN7_OPCODE_MUL 65 +#define GEN7_OPCODE_AVG 66 +#define GEN7_OPCODE_FRC 67 +#define GEN7_OPCODE_RNDU 68 +#define GEN7_OPCODE_RNDD 69 +#define GEN7_OPCODE_RNDE 70 +#define GEN7_OPCODE_RNDZ 71 +#define GEN7_OPCODE_MAC 72 +#define GEN7_OPCODE_MACH 73 +#define GEN7_OPCODE_LZD 74 +#define GEN7_OPCODE_SAD2 80 +#define GEN7_OPCODE_SADA2 81 +#define GEN7_OPCODE_DP4 84 +#define GEN7_OPCODE_DPH 85 +#define GEN7_OPCODE_DP3 86 +#define GEN7_OPCODE_DP2 87 +#define GEN7_OPCODE_DPA2 88 +#define GEN7_OPCODE_LINE 89 +#define GEN7_OPCODE_NOP 126 + +#define GEN7_PREDICATE_NONE 0 +#define GEN7_PREDICATE_NORMAL 1 +#define GEN7_PREDICATE_ALIGN1_ANYV 2 +#define GEN7_PREDICATE_ALIGN1_ALLV 3 +#define GEN7_PREDICATE_ALIGN1_ANY2H 4 +#define GEN7_PREDICATE_ALIGN1_ALL2H 5 +#define GEN7_PREDICATE_ALIGN1_ANY4H 6 +#define GEN7_PREDICATE_ALIGN1_ALL4H 7 +#define GEN7_PREDICATE_ALIGN1_ANY8H 8 +#define GEN7_PREDICATE_ALIGN1_ALL8H 9 +#define GEN7_PREDICATE_ALIGN1_ANY16H 10 +#define GEN7_PREDICATE_ALIGN1_ALL16H 11 +#define GEN7_PREDICATE_ALIGN16_REPLICATE_X 2 +#define GEN7_PREDICATE_ALIGN16_REPLICATE_Y 3 +#define GEN7_PREDICATE_ALIGN16_REPLICATE_Z 4 +#define GEN7_PREDICATE_ALIGN16_REPLICATE_W 5 +#define GEN7_PREDICATE_ALIGN16_ANY4H 6 +#define GEN7_PREDICATE_ALIGN16_ALL4H 7 + +#define GEN7_ARCHITECTURE_REGISTER_FILE 0 +#define GEN7_GENERAL_REGISTER_FILE 1 +#define GEN7_MESSAGE_REGISTER_FILE 2 +#define GEN7_IMMEDIATE_VALUE 3 + +#define GEN7_REGISTER_TYPE_UD 0 +#define GEN7_REGISTER_TYPE_D 1 +#define GEN7_REGISTER_TYPE_UW 2 +#define GEN7_REGISTER_TYPE_W 3 +#define GEN7_REGISTER_TYPE_UB 4 +#define GEN7_REGISTER_TYPE_B 5 +#define GEN7_REGISTER_TYPE_VF 5 /* packed float vector, immediates only? */ +#define GEN7_REGISTER_TYPE_HF 6 +#define GEN7_REGISTER_TYPE_V 6 /* packed int vector, immediates only, uword dest only */ +#define GEN7_REGISTER_TYPE_F 7 + +#define GEN7_ARF_NULL 0x00 +#define GEN7_ARF_ADDRESS 0x10 +#define GEN7_ARF_ACCUMULATOR 0x20 +#define GEN7_ARF_FLAG 0x30 +#define GEN7_ARF_MASK 0x40 +#define GEN7_ARF_MASK_STACK 0x50 +#define GEN7_ARF_MASK_STACK_DEPTH 0x60 +#define GEN7_ARF_STATE 0x70 +#define GEN7_ARF_CONTROL 0x80 +#define GEN7_ARF_NOTIFICATION_COUNT 0x90 +#define GEN7_ARF_IP 0xA0 + +#define GEN7_AMASK 0 +#define GEN7_IMASK 1 +#define GEN7_LMASK 2 +#define GEN7_CMASK 3 + +#define GEN7_THREAD_NORMAL 0 +#define GEN7_THREAD_ATOMIC 1 +#define GEN7_THREAD_SWITCH 2 + +#define GEN7_VERTICAL_STRIDE_0 0 +#define GEN7_VERTICAL_STRIDE_1 1 +#define GEN7_VERTICAL_STRIDE_2 2 +#define GEN7_VERTICAL_STRIDE_4 3 +#define GEN7_VERTICAL_STRIDE_8 4 +#define GEN7_VERTICAL_STRIDE_16 5 +#define GEN7_VERTICAL_STRIDE_32 6 +#define GEN7_VERTICAL_STRIDE_64 7 +#define GEN7_VERTICAL_STRIDE_128 8 +#define GEN7_VERTICAL_STRIDE_256 9 +#define GEN7_VERTICAL_STRIDE_ONE_DIMENSIONAL 0xF + +#define GEN7_WIDTH_1 0 +#define GEN7_WIDTH_2 1 +#define GEN7_WIDTH_4 2 +#define GEN7_WIDTH_8 3 +#define GEN7_WIDTH_16 4 + +#define GEN7_STATELESS_BUFFER_BOUNDARY_1K 0 +#define GEN7_STATELESS_BUFFER_BOUNDARY_2K 1 +#define GEN7_STATELESS_BUFFER_BOUNDARY_4K 2 +#define GEN7_STATELESS_BUFFER_BOUNDARY_8K 3 +#define GEN7_STATELESS_BUFFER_BOUNDARY_16K 4 +#define GEN7_STATELESS_BUFFER_BOUNDARY_32K 5 +#define GEN7_STATELESS_BUFFER_BOUNDARY_64K 6 +#define GEN7_STATELESS_BUFFER_BOUNDARY_128K 7 +#define GEN7_STATELESS_BUFFER_BOUNDARY_256K 8 +#define GEN7_STATELESS_BUFFER_BOUNDARY_512K 9 +#define GEN7_STATELESS_BUFFER_BOUNDARY_1M 10 +#define GEN7_STATELESS_BUFFER_BOUNDARY_2M 11 + +#define GEN7_POLYGON_FACING_FRONT 0 +#define GEN7_POLYGON_FACING_BACK 1 + +#define GEN7_MESSAGE_TARGET_NULL 0 +#define GEN7_MESSAGE_TARGET_MATH 1 +#define GEN7_MESSAGE_TARGET_SAMPLER 2 +#define GEN7_MESSAGE_TARGET_GATEWAY 3 +#define GEN7_MESSAGE_TARGET_DATAPORT_READ 4 +#define GEN7_MESSAGE_TARGET_DATAPORT_WRITE 5 +#define GEN7_MESSAGE_TARGET_URB 6 +#define GEN7_MESSAGE_TARGET_THREAD_SPAWNER 7 + +#define GEN7_SAMPLER_RETURN_FORMAT_FLOAT32 0 +#define GEN7_SAMPLER_RETURN_FORMAT_UINT32 2 +#define GEN7_SAMPLER_RETURN_FORMAT_SINT32 3 + +#define GEN7_SAMPLER_MESSAGE_SIMD8_SAMPLE 0 +#define GEN7_SAMPLER_MESSAGE_SIMD16_SAMPLE 0 +#define GEN7_SAMPLER_MESSAGE_SIMD16_SAMPLE_BIAS 0 +#define GEN7_SAMPLER_MESSAGE_SIMD8_KILLPIX 1 +#define GEN7_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_LOD 1 +#define GEN7_SAMPLER_MESSAGE_SIMD16_SAMPLE_LOD 1 +#define GEN7_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_GRADIENTS 2 +#define GEN7_SAMPLER_MESSAGE_SIMD8_SAMPLE_GRADIENTS 2 +#define GEN7_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_COMPARE 0 +#define GEN7_SAMPLER_MESSAGE_SIMD16_SAMPLE_COMPARE 2 +#define GEN7_SAMPLER_MESSAGE_SIMD4X2_RESINFO 2 +#define GEN7_SAMPLER_MESSAGE_SIMD8_RESINFO 2 +#define GEN7_SAMPLER_MESSAGE_SIMD16_RESINFO 2 +#define GEN7_SAMPLER_MESSAGE_SIMD4X2_LD 3 +#define GEN7_SAMPLER_MESSAGE_SIMD8_LD 3 +#define GEN7_SAMPLER_MESSAGE_SIMD16_LD 3 + +#define GEN7_DATAPORT_OWORD_BLOCK_1_OWORDLOW 0 +#define GEN7_DATAPORT_OWORD_BLOCK_1_OWORDHIGH 1 +#define GEN7_DATAPORT_OWORD_BLOCK_2_OWORDS 2 +#define GEN7_DATAPORT_OWORD_BLOCK_4_OWORDS 3 +#define GEN7_DATAPORT_OWORD_BLOCK_8_OWORDS 4 + +#define GEN7_DATAPORT_OWORD_DUAL_BLOCK_1OWORD 0 +#define GEN7_DATAPORT_OWORD_DUAL_BLOCK_4OWORDS 2 + +#define GEN7_DATAPORT_DWORD_SCATTERED_BLOCK_8DWORDS 2 +#define GEN7_DATAPORT_DWORD_SCATTERED_BLOCK_16DWORDS 3 + +#define GEN7_DATAPORT_READ_MESSAGE_OWORD_BLOCK_READ 0 +#define GEN7_DATAPORT_READ_MESSAGE_OWORD_DUAL_BLOCK_READ 1 +#define GEN7_DATAPORT_READ_MESSAGE_DWORD_BLOCK_READ 2 +#define GEN7_DATAPORT_READ_MESSAGE_DWORD_SCATTERED_READ 3 + +#define GEN7_DATAPORT_READ_TARGET_DATA_CACHE 0 +#define GEN7_DATAPORT_READ_TARGET_RENDER_CACHE 1 +#define GEN7_DATAPORT_READ_TARGET_SAMPLER_CACHE 2 + +#define GEN7_DATAPORT_RENDER_TARGET_WRITE_SIMD16_SINGLE_SOURCE 0 +#define GEN7_DATAPORT_RENDER_TARGET_WRITE_SIMD16_SINGLE_SOURCE_REPLICATED 1 +#define GEN7_DATAPORT_RENDER_TARGET_WRITE_SIMD8_DUAL_SOURCE_SUBSPAN01 2 +#define GEN7_DATAPORT_RENDER_TARGET_WRITE_SIMD8_DUAL_SOURCE_SUBSPAN23 3 +#define GEN7_DATAPORT_RENDER_TARGET_WRITE_SIMD8_SINGLE_SOURCE_SUBSPAN01 4 + +#define GEN7_DATAPORT_WRITE_MESSAGE_OWORD_BLOCK_WRITE 0 +#define GEN7_DATAPORT_WRITE_MESSAGE_OWORD_DUAL_BLOCK_WRITE 1 +#define GEN7_DATAPORT_WRITE_MESSAGE_DWORD_BLOCK_WRITE 2 +#define GEN7_DATAPORT_WRITE_MESSAGE_DWORD_SCATTERED_WRITE 3 +#define GEN7_DATAPORT_WRITE_MESSAGE_RENDER_TARGET_WRITE 4 +#define GEN7_DATAPORT_WRITE_MESSAGE_STREAMED_VERTEX_BUFFER_WRITE 5 +#define GEN7_DATAPORT_WRITE_MESSAGE_FLUSH_RENDER_CACHE 7 + +#define GEN7_MATH_FUNCTION_INV 1 +#define GEN7_MATH_FUNCTION_LOG 2 +#define GEN7_MATH_FUNCTION_EXP 3 +#define GEN7_MATH_FUNCTION_SQRT 4 +#define GEN7_MATH_FUNCTION_RSQ 5 +#define GEN7_MATH_FUNCTION_SIN 6 /* was 7 */ +#define GEN7_MATH_FUNCTION_COS 7 /* was 8 */ +#define GEN7_MATH_FUNCTION_SINCOS 8 /* was 6 */ +#define GEN7_MATH_FUNCTION_TAN 9 +#define GEN7_MATH_FUNCTION_POW 10 +#define GEN7_MATH_FUNCTION_INT_DIV_QUOTIENT_AND_REMAINDER 11 +#define GEN7_MATH_FUNCTION_INT_DIV_QUOTIENT 12 +#define GEN7_MATH_FUNCTION_INT_DIV_REMAINDER 13 + +#define GEN7_MATH_INTEGER_UNSIGNED 0 +#define GEN7_MATH_INTEGER_SIGNED 1 + +#define GEN7_MATH_PRECISION_FULL 0 +#define GEN7_MATH_PRECISION_PARTIAL 1 + +#define GEN7_MATH_SATURATE_NONE 0 +#define GEN7_MATH_SATURATE_SATURATE 1 + +#define GEN7_MATH_DATA_VECTOR 0 +#define GEN7_MATH_DATA_SCALAR 1 + +#define GEN7_URB_OPCODE_WRITE 0 + +#define GEN7_URB_SWIZZLE_NONE 0 +#define GEN7_URB_SWIZZLE_INTERLEAVE 1 +#define GEN7_URB_SWIZZLE_TRANSPOSE 2 + +#define GEN7_SCRATCH_SPACE_SIZE_1K 0 +#define GEN7_SCRATCH_SPACE_SIZE_2K 1 +#define GEN7_SCRATCH_SPACE_SIZE_4K 2 +#define GEN7_SCRATCH_SPACE_SIZE_8K 3 +#define GEN7_SCRATCH_SPACE_SIZE_16K 4 +#define GEN7_SCRATCH_SPACE_SIZE_32K 5 +#define GEN7_SCRATCH_SPACE_SIZE_64K 6 +#define GEN7_SCRATCH_SPACE_SIZE_128K 7 +#define GEN7_SCRATCH_SPACE_SIZE_256K 8 +#define GEN7_SCRATCH_SPACE_SIZE_512K 9 +#define GEN7_SCRATCH_SPACE_SIZE_1M 10 +#define GEN7_SCRATCH_SPACE_SIZE_2M 11 + +/* The hardware supports two different modes for border color. The + * default (OpenGL) mode uses floating-point color channels, while the + * legacy mode uses 4 bytes. + * + * More significantly, the legacy mode respects the components of the + * border color for channels not present in the source, (whereas the + * default mode will ignore the border color's alpha channel and use + * alpha==1 for an RGB source, for example). + * + * The legacy mode matches the semantics specified by the Render + * extension. + */ +struct gen7_sampler_default_border_color { + float color[4]; +}; + +struct gen7_sampler_legacy_border_color { + uint8_t color[4]; +}; + +struct gen7_blend_state { + struct { + uint32_t dest_blend_factor:5; + uint32_t source_blend_factor:5; + uint32_t pad3:1; + uint32_t blend_func:3; + uint32_t pad2:1; + uint32_t ia_dest_blend_factor:5; + uint32_t ia_source_blend_factor:5; uint32_t pad1:1; - uint32_t surface_array_spacing:1; - uint32_t vert_line_stride_ofs:1; - uint32_t vert_line_stride:1; - uint32_t tile_walk:1; - uint32_t tiled_surface:1; - uint32_t horizontal_alignment:1; - uint32_t vertical_alignment:2; - uint32_t surface_format:9; /**< BRW_SURFACEFORMAT_x */ + uint32_t ia_blend_func:3; uint32_t pad0:1; - uint32_t is_array:1; - uint32_t surface_type:3; /**< BRW_SURFACE_1D/2D/3D/CUBE */ + uint32_t ia_blend_enable:1; + uint32_t blend_enable:1; + } blend0; + + struct { + uint32_t post_blend_clamp_enable:1; + uint32_t pre_blend_clamp_enable:1; + uint32_t clamp_range:2; + uint32_t pad0:4; + uint32_t x_dither_offset:2; + uint32_t y_dither_offset:2; + uint32_t dither_enable:1; + uint32_t alpha_test_func:3; + uint32_t alpha_test_enable:1; + uint32_t pad1:1; + uint32_t logic_op_func:4; + uint32_t logic_op_enable:1; + uint32_t pad2:1; + uint32_t write_disable_b:1; + uint32_t write_disable_g:1; + uint32_t write_disable_r:1; + uint32_t write_disable_a:1; + uint32_t pad3:1; + uint32_t alpha_to_coverage_dither:1; + uint32_t alpha_to_one:1; + uint32_t alpha_to_coverage:1; + } blend1; +}; + +struct gen7_color_calc_state { + struct { + uint32_t alpha_test_format:1; + uint32_t pad0:14; + uint32_t round_disable:1; + uint32_t bf_stencil_ref:8; + uint32_t stencil_ref:8; + } cc0; + + union { + float alpha_ref_f; + struct { + uint32_t ui:8; + uint32_t pad0:24; + } alpha_ref_fi; + } cc1; + + float constant_r; + float constant_g; + float constant_b; + float constant_a; +}; + +struct gen7_depth_stencil_state { + struct { + uint32_t pad0:3; + uint32_t bf_stencil_pass_depth_pass_op:3; + uint32_t bf_stencil_pass_depth_fail_op:3; + uint32_t bf_stencil_fail_op:3; + uint32_t bf_stencil_func:3; + uint32_t bf_stencil_enable:1; + uint32_t pad1:2; + uint32_t stencil_write_enable:1; + uint32_t stencil_pass_depth_pass_op:3; + uint32_t stencil_pass_depth_fail_op:3; + uint32_t stencil_fail_op:3; + uint32_t stencil_func:3; + uint32_t stencil_enable:1; + } ds0; + + struct { + uint32_t bf_stencil_write_mask:8; + uint32_t bf_stencil_test_mask:8; + uint32_t stencil_write_mask:8; + uint32_t stencil_test_mask:8; + } ds1; + + struct { + uint32_t pad0:26; + uint32_t depth_write_enable:1; + uint32_t depth_test_func:3; + uint32_t pad1:1; + uint32_t depth_test_enable:1; + } ds2; +}; + +struct gen7_surface_state { + struct { + unsigned int cube_pos_z:1; + unsigned int cube_neg_z:1; + unsigned int cube_pos_y:1; + unsigned int cube_neg_y:1; + unsigned int cube_pos_x:1; + unsigned int cube_neg_x:1; + unsigned int pad2:2; + unsigned int render_cache_read_write:1; + unsigned int pad1:1; + unsigned int surface_array_spacing:1; + unsigned int vert_line_stride_ofs:1; + unsigned int vert_line_stride:1; + unsigned int tile_walk:1; + unsigned int tiled_surface:1; + unsigned int horizontal_alignment:1; + unsigned int vertical_alignment:2; + unsigned int surface_format:9; /**< BRW_SURFACEFORMAT_x */ + unsigned int pad0:1; + unsigned int is_array:1; + unsigned int surface_type:3; /**< BRW_SURFACE_1D/2D/3D/CUBE */ } ss0; struct { - uint32_t base_addr; + unsigned int base_addr; } ss1; struct { - uint32_t width:14; - uint32_t pad1:2; - uint32_t height:14; - uint32_t pad0:2; + unsigned int width:14; + unsigned int pad1:2; + unsigned int height:14; + unsigned int pad0:2; } ss2; struct { - uint32_t pitch:18; - uint32_t pad:3; - uint32_t depth:11; + unsigned int pitch:18; + unsigned int pad:3; + unsigned int depth:11; } ss3; struct { - uint32_t multisample_position_palette_index:3; - uint32_t num_multisamples:3; - uint32_t multisampled_surface_storage_format:1; - uint32_t render_target_view_extent:11; - uint32_t min_array_elt:11; - uint32_t rotation:2; - uint32_t pad0:1; + unsigned int multisample_position_palette_index:3; + unsigned int num_multisamples:3; + unsigned int multisampled_surface_storage_format:1; + unsigned int render_target_view_extent:11; + unsigned int min_array_elt:11; + unsigned int rotation:2; + unsigned int pad0:1; } ss4; struct { - uint32_t mip_count:4; - uint32_t min_lod:4; - uint32_t pad1:12; - uint32_t y_offset:4; - uint32_t pad0:1; - uint32_t x_offset:7; + unsigned int mip_count:4; + unsigned int min_lod:4; + unsigned int pad1:12; + unsigned int y_offset:4; + unsigned int pad0:1; + unsigned int x_offset:7; } ss5; struct { - uint32_t pad; /* Multisample Control Surface stuff */ + unsigned int pad; /* Multisample Control Surface stuff */ } ss6; struct { - uint32_t resource_min_lod:12; - - /* Only on Haswell */ - uint32_t pad0:4; - uint32_t shader_chanel_select_a:3; - uint32_t shader_chanel_select_b:3; - uint32_t shader_chanel_select_g:3; - uint32_t shader_chanel_select_r:3; - - uint32_t alpha_clear_color:1; - uint32_t blue_clear_color:1; - uint32_t green_clear_color:1; - uint32_t red_clear_color:1; + unsigned int resource_min_lod:12; + unsigned int pad0:16; + unsigned int alpha_clear_color:1; + unsigned int blue_clear_color:1; + unsigned int green_clear_color:1; + unsigned int red_clear_color:1; } ss7; }; -struct gen7_sampler_state -{ - struct - { - uint32_t aniso_algorithm:1; - uint32_t lod_bias:13; - uint32_t min_filter:3; - uint32_t mag_filter:3; - uint32_t mip_filter:2; - uint32_t base_level:5; - uint32_t pad1:1; - uint32_t lod_preclamp:1; - uint32_t default_color_mode:1; - uint32_t pad0:1; - uint32_t disable:1; +struct gen7_sampler_state { + struct { + unsigned int aniso_algorithm:1; + unsigned int lod_bias:13; + unsigned int min_filter:3; + unsigned int mag_filter:3; + unsigned int mip_filter:2; + unsigned int base_level:5; + unsigned int pad1:1; + unsigned int lod_preclamp:1; + unsigned int default_color_mode:1; + unsigned int pad0:1; + unsigned int disable:1; } ss0; - struct - { - uint32_t cube_control_mode:1; - uint32_t shadow_function:3; - uint32_t pad:4; - uint32_t max_lod:12; - uint32_t min_lod:12; + struct { + unsigned int cube_control_mode:1; + unsigned int shadow_function:3; + unsigned int pad:4; + unsigned int max_lod:12; + unsigned int min_lod:12; } ss1; - struct - { - uint32_t pad:5; - uint32_t default_color_pointer:27; + struct { + unsigned int pad:5; + unsigned int default_color_pointer:27; } ss2; - struct - { - uint32_t r_wrap_mode:3; - uint32_t t_wrap_mode:3; - uint32_t s_wrap_mode:3; - uint32_t pad:1; - uint32_t non_normalized_coord:1; - uint32_t trilinear_quality:2; - uint32_t address_round:6; - uint32_t max_aniso:3; - uint32_t chroma_key_mode:1; - uint32_t chroma_key_index:2; - uint32_t chroma_key_enable:1; - uint32_t pad0:6; + struct { + unsigned int r_wrap_mode:3; + unsigned int t_wrap_mode:3; + unsigned int s_wrap_mode:3; + unsigned int pad:1; + unsigned int non_normalized_coord:1; + unsigned int trilinear_quality:2; + unsigned int address_round:6; + unsigned int max_aniso:3; + unsigned int chroma_key_mode:1; + unsigned int chroma_key_index:2; + unsigned int chroma_key_enable:1; + unsigned int pad0:6; } ss3; }; -struct gen7_sf_clip_viewport { - struct { - float m00; - float m11; - float m22; - float m30; - float m31; - float m32; - } viewport; +/* Surface state DW0 */ +#define GEN7_SURFACE_RC_READ_WRITE (1 << 8) +#define GEN7_SURFACE_TILED (1 << 14) +#define GEN7_SURFACE_TILED_Y (1 << 13) +#define GEN7_SURFACE_FORMAT_SHIFT 18 +#define GEN7_SURFACE_TYPE_SHIFT 29 - uint32_t pad0[2]; +/* Surface state DW2 */ +#define GEN7_SURFACE_HEIGHT_SHIFT 16 +#define GEN7_SURFACE_WIDTH_SHIFT 0 - struct { - float xmin; - float xmax; - float ymin; - float ymax; - } guardband; +/* Surface state DW3 */ +#define GEN7_SURFACE_DEPTH_SHIFT 21 +#define GEN7_SURFACE_PITCH_SHIFT 0 - float pad1[4]; -}; +#define HSW_SWIZZLE_ZERO 0 +#define HSW_SWIZZLE_ONE 1 +#define HSW_SWIZZLE_RED 4 +#define HSW_SWIZZLE_GREEN 5 +#define HSW_SWIZZLE_BLUE 6 +#define HSW_SWIZZLE_ALPHA 7 +#define __HSW_SURFACE_SWIZZLE(r,g,b,a) \ + ((a) << 16 | (b) << 19 | (g) << 22 | (r) << 25) +#define HSW_SURFACE_SWIZZLE(r,g,b,a) \ + __HSW_SURFACE_SWIZZLE(HSW_SWIZZLE_##r, HSW_SWIZZLE_##g, HSW_SWIZZLE_##b, HSW_SWIZZLE_##a) + +/* _3DSTATE_VERTEX_BUFFERS on GEN7*/ +/* DW1 */ +#define GEN7_VB0_ADDRESS_MODIFYENABLE (1 << 14) + +/* _3DPRIMITIVE on GEN7 */ +/* DW1 */ +# define GEN7_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL (0 << 8) +# define GEN7_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM (1 << 8) + +#define GEN7_3DSTATE_CLEAR_PARAMS GEN7_3D(3, 0, 0x04) +#define GEN7_3DSTATE_DEPTH_BUFFER GEN7_3D(3, 0, 0x05) +# define GEN7_3DSTATE_DEPTH_BUFFER_TYPE_SHIFT 29 +# define GEN7_3DSTATE_DEPTH_BUFFER_FORMAT_SHIFT 18 +/* DW1 */ +# define GEN7_3DSTATE_DEPTH_CLEAR_VALID (1 << 15) + +#define GEN7_3DSTATE_CONSTANT_HS GEN7_3D(3, 0, 0x19) +#define GEN7_3DSTATE_CONSTANT_DS GEN7_3D(3, 0, 0x1a) -struct gen6_scissor_rect -{ - uint32_t xmin:16; - uint32_t ymin:16; - uint32_t xmax:16; - uint32_t ymax:16; +#define GEN7_3DSTATE_HS GEN7_3D(3, 0, 0x1b) +#define GEN7_3DSTATE_TE GEN7_3D(3, 0, 0x1c) +#define GEN7_3DSTATE_DS GEN7_3D(3, 0, 0x1d) +#define GEN7_3DSTATE_STREAMOUT GEN7_3D(3, 0, 0x1e) +#define GEN7_3DSTATE_SBE GEN7_3D(3, 0, 0x1f) + +/* DW1 */ +# define GEN7_SBE_SWIZZLE_CONTROL_MODE (1 << 28) +# define GEN7_SBE_NUM_OUTPUTS_SHIFT 22 +# define GEN7_SBE_SWIZZLE_ENABLE (1 << 21) +# define GEN7_SBE_POINT_SPRITE_LOWERLEFT (1 << 20) +# define GEN7_SBE_URB_ENTRY_READ_LENGTH_SHIFT 11 +# define GEN7_SBE_URB_ENTRY_READ_OFFSET_SHIFT 4 + +#define GEN7_3DSTATE_PS GEN7_3D(3, 0, 0x20) +/* DW1: kernel pointer */ +/* DW2 */ +# define GEN7_PS_SPF_MODE (1 << 31) +# define GEN7_PS_VECTOR_MASK_ENABLE (1 << 30) +# define GEN7_PS_SAMPLER_COUNT_SHIFT 27 +# define GEN7_PS_BINDING_TABLE_ENTRY_COUNT_SHIFT 18 +# define GEN7_PS_FLOATING_POINT_MODE_IEEE_754 (0 << 16) +# define GEN7_PS_FLOATING_POINT_MODE_ALT (1 << 16) +/* DW3: scratch space */ +/* DW4 */ +# define IVB_PS_MAX_THREADS_SHIFT 24 +# define HSW_PS_MAX_THREADS_SHIFT 23 +# define HSW_PS_SAMPLE_MASK_SHIFT 12 +# define GEN7_PS_PUSH_CONSTANT_ENABLE (1 << 11) +# define GEN7_PS_ATTRIBUTE_ENABLE (1 << 10) +# define GEN7_PS_OMASK_TO_RENDER_TARGET (1 << 9) +# define GEN7_PS_DUAL_SOURCE_BLEND_ENABLE (1 << 7) +# define GEN7_PS_POSOFFSET_NONE (0 << 3) +# define GEN7_PS_POSOFFSET_CENTROID (2 << 3) +# define GEN7_PS_POSOFFSET_SAMPLE (3 << 3) +# define GEN7_PS_32_DISPATCH_ENABLE (1 << 2) +# define GEN7_PS_16_DISPATCH_ENABLE (1 << 1) +# define GEN7_PS_8_DISPATCH_ENABLE (1 << 0) +/* DW5 */ +# define GEN7_PS_DISPATCH_START_GRF_SHIFT_0 16 +# define GEN7_PS_DISPATCH_START_GRF_SHIFT_1 8 +# define GEN7_PS_DISPATCH_START_GRF_SHIFT_2 0 +/* DW6: kernel 1 pointer */ +/* DW7: kernel 2 pointer */ + +#define GEN7_3DSTATE_VIEWPORT_STATE_POINTERS_SF_CL GEN7_3D(3, 0, 0x21) +#define GEN7_3DSTATE_VIEWPORT_STATE_POINTERS_CC GEN7_3D(3, 0, 0x23) + +#define GEN7_3DSTATE_BLEND_STATE_POINTERS GEN7_3D(3, 0, 0x24) +#define GEN7_3DSTATE_DEPTH_STENCIL_STATE_POINTERS GEN7_3D(3, 0, 0x25) + +#define GEN7_3DSTATE_BINDING_TABLE_POINTERS_VS GEN7_3D(3, 0, 0x26) +#define GEN7_3DSTATE_BINDING_TABLE_POINTERS_HS GEN7_3D(3, 0, 0x27) +#define GEN7_3DSTATE_BINDING_TABLE_POINTERS_DS GEN7_3D(3, 0, 0x28) +#define GEN7_3DSTATE_BINDING_TABLE_POINTERS_GS GEN7_3D(3, 0, 0x29) +#define GEN7_3DSTATE_BINDING_TABLE_POINTERS_PS GEN7_3D(3, 0, 0x2a) + +#define GEN7_3DSTATE_SAMPLER_STATE_POINTERS_VS GEN7_3D(3, 0, 0x2b) +#define GEN7_3DSTATE_SAMPLER_STATE_POINTERS_GS GEN7_3D(3, 0, 0x2e) +#define GEN7_3DSTATE_SAMPLER_STATE_POINTERS_PS GEN7_3D(3, 0, 0x2f) + +#define GEN7_3DSTATE_URB_VS GEN7_3D(3, 0, 0x30) +#define GEN7_3DSTATE_URB_HS GEN7_3D(3, 0, 0x31) +#define GEN7_3DSTATE_URB_DS GEN7_3D(3, 0, 0x32) +#define GEN7_3DSTATE_URB_GS GEN7_3D(3, 0, 0x33) +/* DW1 */ +# define GEN7_URB_ENTRY_NUMBER_SHIFT 0 +# define GEN7_URB_ENTRY_SIZE_SHIFT 16 +# define GEN7_URB_STARTING_ADDRESS_SHIFT 25 + +#define GEN7_3DSTATE_PUSH_CONSTANT_ALLOC_VS GEN7_3D(3, 1, 0x12) +#define GEN7_3DSTATE_PUSH_CONSTANT_ALLOC_PS GEN7_3D(3, 1, 0x16) +/* DW1 */ +# define GEN7_PUSH_CONSTANT_BUFFER_OFFSET_SHIFT 16 + +struct gen7_cc_viewport { + float min_depth; + float max_depth; }; +typedef enum { + SAMPLER_FILTER_NEAREST = 0, + SAMPLER_FILTER_BILINEAR, + FILTER_COUNT +} sampler_filter_t; + +typedef enum { + SAMPLER_EXTEND_NONE = 0, + SAMPLER_EXTEND_REPEAT, + SAMPLER_EXTEND_PAD, + SAMPLER_EXTEND_REFLECT, + EXTEND_COUNT +} sampler_extend_t; + #endif diff --git a/lib/intel_batchbuffer.h b/lib/intel_batchbuffer.h index ba0934c..9f45ac6 100644 --- a/lib/intel_batchbuffer.h +++ b/lib/intel_batchbuffer.h @@ -15,6 +15,7 @@ struct intel_batchbuffer { uint8_t buffer[BATCH_SZ]; uint8_t *ptr; + uint8_t *state; }; struct intel_batchbuffer *intel_batchbuffer_alloc(drm_intel_bufmgr *bufmgr, diff --git a/lib/rendercopy_gen7.c b/lib/rendercopy_gen7.c index 56181ab..fdf741f 100644 --- a/lib/rendercopy_gen7.c +++ b/lib/rendercopy_gen7.c @@ -4,53 +4,22 @@ #include <assert.h> #define ALIGN(x, y) (((x) + (y)-1) & ~((y)-1)) -#define VERTEX_SIZE (3*4) - -#if DEBUG_RENDERCPY -static void dump_batch(struct intel_batchbuffer *batch) -#else -#define dump_batch(x) do { } while(0) -#endif - -struct { - uint32_t cc_state; - uint32_t blend_state; - uint32_t ds_state; -} cc; - -struct { - uint32_t cc_state; - uint32_t sf_clip_state; -} viewport; - -/* see shaders/ps/blit.g7a */ + static const uint32_t ps_kernel[][4] = { -#if 1 - { 0x0060005a, 0x214077bd, 0x000000c0, 0x008d0040 }, - { 0x0060005a, 0x216077bd, 0x000000c0, 0x008d0080 }, - { 0x0060005a, 0x218077bd, 0x000000d0, 0x008d0040 }, - { 0x0060005a, 0x21a077bd, 0x000000d0, 0x008d0080 }, - { 0x02800031, 0x2e001e3d, 0x00000140, 0x08840001 }, - { 0x05800031, 0x20001e3c, 0x00000e00, 0x90031000 }, - -#else - /* Write all -1 */ - { 0x00600001, 0x2e000061, 0x00000000, 0x3f800000 }, - { 0x00600001, 0x2e200061, 0x00000000, 0x3f800000 }, - { 0x00600001, 0x2e400061, 0x00000000, 0x3f800000 }, - { 0x00600001, 0x2e600061, 0x00000000, 0x3f800000 }, - { 0x00600001, 0x2e800061, 0x00000000, 0x3f800000 }, - { 0x00600001, 0x2ea00061, 0x00000000, 0x3f800000 }, - { 0x00600001, 0x2ec00061, 0x00000000, 0x3f800000 }, - { 0x00600001, 0x2ee00061, 0x00000000, 0x3f800000 }, - { 0x05800031, 0x20001e3c, 0x00000e00, 0x90031000 }, -#endif + { 0x0080005a, 0x2e2077bd, 0x000000c0, 0x008d0040 }, + { 0x0080005a, 0x2e6077bd, 0x000000d0, 0x008d0040 }, + { 0x02800031, 0x21801fa9, 0x008d0e20, 0x08840001 }, + { 0x00800001, 0x2e2003bd, 0x008d0180, 0x00000000 }, + { 0x00800001, 0x2e6003bd, 0x008d01c0, 0x00000000 }, + { 0x00800001, 0x2ea003bd, 0x008d0200, 0x00000000 }, + { 0x00800001, 0x2ee003bd, 0x008d0240, 0x00000000 }, + { 0x05800031, 0x20001fa8, 0x008d0e20, 0x90031000 }, }; static uint32_t batch_used(struct intel_batchbuffer *batch) { - return batch->ptr - batch->buffer; + return batch->state - batch->buffer; } static uint32_t @@ -58,7 +27,7 @@ batch_align(struct intel_batchbuffer *batch, uint32_t align) { uint32_t offset = batch_used(batch); offset = ALIGN(offset, align); - batch->ptr = batch->buffer + offset; + batch->state = batch->buffer + offset; return offset; } @@ -66,7 +35,7 @@ static void * batch_alloc(struct intel_batchbuffer *batch, uint32_t size, uint32_t align) { uint32_t offset = batch_align(batch, align); - batch->ptr += size; + batch->state += size; return memset(batch->buffer + offset, 0, size); } @@ -83,7 +52,7 @@ batch_copy(struct intel_batchbuffer *batch, const void *ptr, uint32_t size, uint } static void -gen6_render_flush(struct intel_batchbuffer *batch, uint32_t batch_end) +gen7_render_flush(struct intel_batchbuffer *batch, uint32_t batch_end) { int ret; @@ -94,11 +63,24 @@ gen6_render_flush(struct intel_batchbuffer *batch, uint32_t batch_end) assert(ret == 0); } -/* Mostly copy+paste from gen6, except height, width, pitch moved */ static uint32_t -gen7_bind_buf(struct intel_batchbuffer *batch, struct scratch_buf *buf, - uint32_t format, int is_dst) { - struct gen7_surface_state *ss; +gen7_tiling_bits(uint32_t tiling) +{ + switch (tiling) { + default: assert(0); + case I915_TILING_NONE: return 0; + case I915_TILING_X: return GEN7_SURFACE_TILED; + case I915_TILING_Y: return GEN7_SURFACE_TILED | GEN7_SURFACE_TILED_Y; + } +} + +static uint32_t +gen7_bind_buf(struct intel_batchbuffer *batch, + struct scratch_buf *buf, + uint32_t format, + int is_dst) +{ + uint32_t *ss; uint32_t write_domain, read_domain; int ret; @@ -110,13 +92,20 @@ gen7_bind_buf(struct intel_batchbuffer *batch, struct scratch_buf *buf, } ss = batch_alloc(batch, sizeof(*ss), 32); - ss->ss0.surface_type = GEN6_SURFACE_2D; - ss->ss0.surface_format = format; - ss->ss0.render_cache_read_write = 1; /* GEN7+ */ - ss->ss0.tiled_surface = buf->tiling != I915_TILING_NONE; - ss->ss0.tile_walk = buf->tiling == I915_TILING_Y; - ss->ss1.base_addr = buf->bo->offset; + ss[0] = (GEN7_SURFACE_2D << GEN7_SURFACE_TYPE_SHIFT | + gen7_tiling_bits(buf->tiling) | + format << GEN7_SURFACE_FORMAT_SHIFT); + ss[1] = buf->bo->offset; + ss[2] = ((buf_width(buf) - 1) << GEN7_SURFACE_WIDTH_SHIFT | + (buf_height(buf) - 1) << GEN7_SURFACE_HEIGHT_SHIFT); + ss[3] = (buf->stride - 1) << GEN7_SURFACE_PITCH_SHIFT; + ss[4] = 0; + ss[5] = 0; + ss[6] = 0; + ss[7] = 0; + if (IS_HASWELL(batch->devid)) + ss[7] |= HSW_SURFACE_SWIZZLE(RED, GREEN, BLUE, ALPHA); ret = drm_intel_bo_emit_reloc(batch->bo, batch_offset(batch, ss) + 4, @@ -124,574 +113,403 @@ gen7_bind_buf(struct intel_batchbuffer *batch, struct scratch_buf *buf, read_domain, write_domain); assert(ret == 0); - ss->ss2.height = buf_height(buf) - 1; - ss->ss2.width = buf_width(buf) - 1; - ss->ss3.pitch = buf->stride - 1; - - if (IS_HASWELL(batch->devid)) { - ss->ss7.shader_chanel_select_a = 4; - ss->ss7.shader_chanel_select_g = 5; - ss->ss7.shader_chanel_select_b = 6; - ss->ss7.shader_chanel_select_a = 7; - } - return batch_offset(batch, ss); } -static uint32_t -gen7_bind_surfaces(struct intel_batchbuffer *batch, - struct scratch_buf *src, - struct scratch_buf *dst) { - uint32_t *binding_table; - - binding_table = batch_alloc(batch, 8, 32); - - binding_table[0] = - gen7_bind_buf(batch, dst, GEN6_SURFACEFORMAT_B8G8R8A8_UNORM, 1); - binding_table[1] = - gen7_bind_buf(batch, src, GEN6_SURFACEFORMAT_B8G8R8A8_UNORM, 0); - - return batch_offset(batch, binding_table); +static void +gen7_emit_vertex_elements(struct intel_batchbuffer *batch) +{ + OUT_BATCH(GEN7_3DSTATE_VERTEX_ELEMENTS | + ((2 * (1 + 2)) + 1 - 2)); + + OUT_BATCH(0 << GEN7_VE0_VERTEX_BUFFER_INDEX_SHIFT | GEN7_VE0_VALID | + GEN7_SURFACEFORMAT_R32G32B32A32_FLOAT << GEN7_VE0_FORMAT_SHIFT | + 0 << GEN7_VE0_OFFSET_SHIFT); + + OUT_BATCH(GEN7_VFCOMPONENT_STORE_0 << GEN7_VE1_VFCOMPONENT_0_SHIFT | + GEN7_VFCOMPONENT_STORE_0 << GEN7_VE1_VFCOMPONENT_1_SHIFT | + GEN7_VFCOMPONENT_STORE_0 << GEN7_VE1_VFCOMPONENT_2_SHIFT | + GEN7_VFCOMPONENT_STORE_0 << GEN7_VE1_VFCOMPONENT_3_SHIFT); + + /* x,y */ + OUT_BATCH(0 << GEN7_VE0_VERTEX_BUFFER_INDEX_SHIFT | GEN7_VE0_VALID | + GEN7_SURFACEFORMAT_R16G16_SSCALED << GEN7_VE0_FORMAT_SHIFT | + 0 << GEN7_VE0_OFFSET_SHIFT); /* offsets vb in bytes */ + OUT_BATCH(GEN7_VFCOMPONENT_STORE_SRC << GEN7_VE1_VFCOMPONENT_0_SHIFT | + GEN7_VFCOMPONENT_STORE_SRC << GEN7_VE1_VFCOMPONENT_1_SHIFT | + GEN7_VFCOMPONENT_STORE_0 << GEN7_VE1_VFCOMPONENT_2_SHIFT | + GEN7_VFCOMPONENT_STORE_1_FLT << GEN7_VE1_VFCOMPONENT_3_SHIFT); + + /* s,t */ + OUT_BATCH(0 << GEN7_VE0_VERTEX_BUFFER_INDEX_SHIFT | GEN7_VE0_VALID | + GEN7_SURFACEFORMAT_R16G16_SSCALED << GEN7_VE0_FORMAT_SHIFT | + 4 << GEN7_VE0_OFFSET_SHIFT); /* offset vb in bytes */ + OUT_BATCH(GEN7_VFCOMPONENT_STORE_SRC << GEN7_VE1_VFCOMPONENT_0_SHIFT | + GEN7_VFCOMPONENT_STORE_SRC << GEN7_VE1_VFCOMPONENT_1_SHIFT | + GEN7_VFCOMPONENT_STORE_0 << GEN7_VE1_VFCOMPONENT_2_SHIFT | + GEN7_VFCOMPONENT_STORE_1_FLT << GEN7_VE1_VFCOMPONENT_3_SHIFT); } -/* Mostly copy+paste from gen6, except wrap modes moved */ static uint32_t -gen7_create_sampler(struct intel_batchbuffer *batch) { - struct gen7_sampler_state *ss; +gen7_create_vertex_buffer(struct intel_batchbuffer *batch, + uint32_t src_x, uint32_t src_y, + uint32_t dst_x, uint32_t dst_y, + uint32_t width, uint32_t height) +{ + uint16_t *v; - ss = batch_alloc(batch, sizeof(*ss), 32); + v = batch_alloc(batch, 12*sizeof(*v), 8); - ss->ss0.min_filter = GEN6_MAPFILTER_NEAREST; - ss->ss0.mag_filter = GEN6_MAPFILTER_NEAREST; - ss->ss3.r_wrap_mode = GEN6_TEXCOORDMODE_CLAMP; - ss->ss3.s_wrap_mode = GEN6_TEXCOORDMODE_CLAMP; - ss->ss3.t_wrap_mode = GEN6_TEXCOORDMODE_CLAMP; + v[0] = dst_x + width; + v[1] = dst_y + height; + v[2] = src_x + width; + v[3] = src_y + height; - /* I've experimented with non-normalized coordinates and using the LD - * sampler fetch, but couldn't make it work. */ - ss->ss3.non_normalized_coord = 0; + v[4] = dst_x; + v[5] = dst_y + height; + v[6] = src_x; + v[7] = src_y + height; - return batch_offset(batch, ss); + v[8] = dst_x; + v[9] = dst_y; + v[10] = src_x; + v[11] = src_y; + + return batch_offset(batch, v); } -/** - * gen7_fill_vertex_buffer_data populate vertex buffer with data. - * - * The vertex buffer consists of 3 vertices to construct a RECTLIST. The 4th - * vertex is implied (automatically derived by the HW). Each element has the - * destination offset, and the normalized texture offset (src). The rectangle - * itself will span the entire subsurface to be copied. - * - * see gen6_emit_vertex_elements - */ -static uint32_t -gen7_fill_vertex_buffer_data(struct intel_batchbuffer *batch, - struct scratch_buf *src, - uint32_t src_x, uint32_t src_y, - uint32_t dst_x, uint32_t dst_y, - uint32_t width, uint32_t height) { - void *ret; - - ret = batch->ptr; - - emit_vertex_2s(batch, dst_x + width, dst_y + height); - emit_vertex_normalized(batch, src_x + width, buf_width(src)); - emit_vertex_normalized(batch, src_y + height, buf_height(src)); - - emit_vertex_2s(batch, dst_x, dst_y + height); - emit_vertex_normalized(batch, src_x, buf_width(src)); - emit_vertex_normalized(batch, src_y + height, buf_height(src)); - - emit_vertex_2s(batch, dst_x, dst_y); - emit_vertex_normalized(batch, src_x, buf_width(src)); - emit_vertex_normalized(batch, src_y, buf_height(src)); - - return batch_offset(batch, ret); -} - -/** - * gen6_emit_vertex_elements - The vertex elements describe the contents of the - * vertex buffer. We pack the vertex buffer in a semi weird way, conforming to - * what gen6_rendercopy did. The most straightforward would be to store - * everything as floats. - * - * see gen7_fill_vertex_buffer_data() for where the corresponding elements are - * packed. - */ -static void -gen6_emit_vertex_elements(struct intel_batchbuffer *batch) { - /* - * The VUE layout - * dword 0-3: pad (0, 0, 0. 0) - * dword 4-7: position (x, y, 0, 1.0), - * dword 8-11: texture coordinate 0 (u0, v0, 0, 1.0) - */ - OUT_BATCH(GEN6_3DSTATE_VERTEX_ELEMENTS | (3 * 2 + 1 - 2)); - - /* Element state 0. These are 4 dwords of 0 required for the VUE format. - * We don't really know or care what they do. - */ - OUT_BATCH(0 << VE0_VERTEX_BUFFER_INDEX_SHIFT | VE0_VALID | - GEN6_SURFACEFORMAT_R32G32B32A32_FLOAT << VE0_FORMAT_SHIFT | - 0 << VE0_OFFSET_SHIFT); /* we specify 0, but it's really does not exist */ - OUT_BATCH(GEN6_VFCOMPONENT_STORE_0 << VE1_VFCOMPONENT_0_SHIFT | - GEN6_VFCOMPONENT_STORE_0 << VE1_VFCOMPONENT_1_SHIFT | - GEN6_VFCOMPONENT_STORE_0 << VE1_VFCOMPONENT_2_SHIFT | - GEN6_VFCOMPONENT_STORE_0 << VE1_VFCOMPONENT_3_SHIFT); - - /* Element state 1 - Our "destination" vertices. These are passed down - * through the pipeline, and eventually make it to the pixel shader as - * the offsets in the destination surface. It's packed as the 16 - * signed/scaled because of gen6 rendercopy. I see no particular reason - * for doing this though. - */ - OUT_BATCH(0 << VE0_VERTEX_BUFFER_INDEX_SHIFT | VE0_VALID | - GEN6_SURFACEFORMAT_R16G16_SSCALED << VE0_FORMAT_SHIFT | - 0 << VE0_OFFSET_SHIFT); /* offsets vb in bytes */ - OUT_BATCH(GEN6_VFCOMPONENT_STORE_SRC << VE1_VFCOMPONENT_0_SHIFT | - GEN6_VFCOMPONENT_STORE_SRC << VE1_VFCOMPONENT_1_SHIFT | - GEN6_VFCOMPONENT_STORE_0 << VE1_VFCOMPONENT_2_SHIFT | - GEN6_VFCOMPONENT_STORE_1_FLT << VE1_VFCOMPONENT_3_SHIFT); - - /* Element state 2. Last but not least we store the U,V components as - * normalized floats. These will be used in the pixel shader to sample - * from the source buffer. - */ - OUT_BATCH(0 << VE0_VERTEX_BUFFER_INDEX_SHIFT | VE0_VALID | - GEN6_SURFACEFORMAT_R32G32_FLOAT << VE0_FORMAT_SHIFT | - 4 << VE0_OFFSET_SHIFT); /* offset vb in bytes */ - OUT_BATCH(GEN6_VFCOMPONENT_STORE_SRC << VE1_VFCOMPONENT_0_SHIFT | - GEN6_VFCOMPONENT_STORE_SRC << VE1_VFCOMPONENT_1_SHIFT | - GEN6_VFCOMPONENT_STORE_0 << VE1_VFCOMPONENT_2_SHIFT | - GEN6_VFCOMPONENT_STORE_1_FLT << VE1_VFCOMPONENT_3_SHIFT); -} - -/** - * gen7_emit_vertex_buffer emit the vertex buffers command - * - * @batch - * @offset - bytw offset within the @batch where the vertex buffer starts. - */ static void gen7_emit_vertex_buffer(struct intel_batchbuffer *batch, - uint32_t offset) { - OUT_BATCH(GEN6_3DSTATE_VERTEX_BUFFERS | (4 * 1 - 1)); - OUT_BATCH(0 << VB0_BUFFER_INDEX_SHIFT | /* VB 0th index */ - VB0_VERTEXDATA | - GEN7_VB0_BUFFER_ADDR_MOD_EN | /* Address Modify Enable */ - VERTEX_SIZE << VB0_BUFFER_PITCH_SHIFT); + int src_x, int src_y, + int dst_x, int dst_y, + int width, int height) +{ + uint32_t offset; + + offset = gen7_create_vertex_buffer(batch, + src_x, src_y, + dst_x, dst_y, + width, height); + + OUT_BATCH(GEN7_3DSTATE_VERTEX_BUFFERS | (5 - 2)); + OUT_BATCH(0 << GEN7_VB0_BUFFER_INDEX_SHIFT | + GEN7_VB0_VERTEXDATA | + GEN7_VB0_ADDRESS_MODIFY_ENABLE | + 4*2 << GEN7_VB0_BUFFER_PITCH_SHIFT); + OUT_RELOC(batch->bo, I915_GEM_DOMAIN_VERTEX, 0, offset); - OUT_RELOC(batch->bo, I915_GEM_DOMAIN_VERTEX, 0, offset + (VERTEX_SIZE * 3) - 1); + OUT_BATCH(~0); OUT_BATCH(0); } static uint32_t -gen6_create_cc_state(struct intel_batchbuffer *batch) +gen7_bind_surfaces(struct intel_batchbuffer *batch, + struct scratch_buf *src, + struct scratch_buf *dst) +{ + uint32_t *binding_table; + + binding_table = batch_alloc(batch, 8, 32); + + binding_table[0] = + gen7_bind_buf(batch, dst, GEN7_SURFACEFORMAT_B8G8R8A8_UNORM, 1); + binding_table[1] = + gen7_bind_buf(batch, src, GEN7_SURFACEFORMAT_B8G8R8A8_UNORM, 0); + + return batch_offset(batch, binding_table); +} + +static void +gen7_emit_binding_table(struct intel_batchbuffer *batch, + struct scratch_buf *src, + struct scratch_buf *dst) { - struct gen6_color_calc_state *cc_state; - cc_state = batch_alloc(batch, sizeof(*cc_state), 64); - return batch_offset(batch, cc_state); + OUT_BATCH(GEN7_3DSTATE_BINDING_TABLE_POINTERS_PS | (2 - 2)); + OUT_BATCH(gen7_bind_surfaces(batch, src, dst)); } -static uint32_t -gen6_create_depth_stencil_state(struct intel_batchbuffer *batch) +static void +gen7_emit_drawing_rectangle(struct intel_batchbuffer *batch, struct scratch_buf *dst) { - struct gen6_depth_stencil_state *depth; - depth = batch_alloc(batch, sizeof(*depth), 64); - depth->ds0.stencil_enable = 0; - return batch_offset(batch, depth); + OUT_BATCH(GEN7_3DSTATE_DRAWING_RECTANGLE | (4 - 2)); + OUT_BATCH(0); + OUT_BATCH((buf_height(dst) - 1) << 16 | (buf_width(dst) - 1)); + OUT_BATCH(0); } static uint32_t -gen6_create_blend_state(struct intel_batchbuffer *batch) +gen7_create_blend_state(struct intel_batchbuffer *batch) { - struct gen6_blend_state *blend; + struct gen7_blend_state *blend; + blend = batch_alloc(batch, sizeof(*blend), 64); - blend->blend0.blend_enable = 0; + + blend->blend0.dest_blend_factor = GEN7_BLENDFACTOR_ZERO; + blend->blend0.source_blend_factor = GEN7_BLENDFACTOR_ONE; + blend->blend0.blend_func = GEN7_BLENDFUNCTION_ADD; + blend->blend1.post_blend_clamp_enable = 1; blend->blend1.pre_blend_clamp_enable = 1; + return batch_offset(batch, blend); } +static void +gen7_emit_state_base_address(struct intel_batchbuffer *batch) +{ + OUT_BATCH(GEN7_STATE_BASE_ADDRESS | (10 - 2)); + OUT_BATCH(0); + OUT_RELOC(batch->bo, I915_GEM_DOMAIN_INSTRUCTION, 0, BASE_ADDRESS_MODIFY); + OUT_RELOC(batch->bo, I915_GEM_DOMAIN_INSTRUCTION, 0, BASE_ADDRESS_MODIFY); + OUT_BATCH(0); + OUT_RELOC(batch->bo, I915_GEM_DOMAIN_INSTRUCTION, 0, BASE_ADDRESS_MODIFY); + + OUT_BATCH(0); + OUT_BATCH(0 | BASE_ADDRESS_MODIFY); + OUT_BATCH(0); + OUT_BATCH(0 | BASE_ADDRESS_MODIFY); +} + static uint32_t -gen6_create_cc_viewport(struct intel_batchbuffer *batch) +gen7_create_cc_viewport(struct intel_batchbuffer *batch) { - struct gen6_cc_viewport *vp; + struct gen7_cc_viewport *vp; vp = batch_alloc(batch, sizeof(*vp), 32); - /* XXX I don't understand this */ vp->min_depth = -1.e35; vp->max_depth = 1.e35; + return batch_offset(batch, vp); } -static uint32_t -gen7_create_sf_clip_viewport(struct intel_batchbuffer *batch) { - /* XXX these are likely not needed */ - struct gen7_sf_clip_viewport *scv_state; - scv_state = batch_alloc(batch, sizeof(*scv_state), 64); - scv_state->guardband.xmin = 0; - scv_state->guardband.xmax = 1.0f; - scv_state->guardband.ymin = 0; - scv_state->guardband.ymax = 1.0f; - return batch_offset(batch, scv_state); +static void +gen7_emit_cc(struct intel_batchbuffer *batch) +{ + OUT_BATCH(GEN7_3DSTATE_BLEND_STATE_POINTERS | (2 - 2)); + OUT_BATCH(gen7_create_blend_state(batch)); + + OUT_BATCH(GEN7_3DSTATE_VIEWPORT_STATE_POINTERS_CC | (2 - 2)); + OUT_BATCH(gen7_create_cc_viewport(batch)); } static uint32_t -gen6_create_scissor_rect(struct intel_batchbuffer *batch) +gen7_create_sampler(struct intel_batchbuffer *batch) { - struct gen6_scissor_rect *scissor; - scissor = batch_alloc(batch, sizeof(*scissor), 64); - return batch_offset(batch, scissor); -} + struct gen7_sampler_state *ss; + ss = batch_alloc(batch, sizeof(*ss), 32); + ss->ss0.min_filter = GEN7_MAPFILTER_NEAREST; + ss->ss0.mag_filter = GEN7_MAPFILTER_NEAREST; + ss->ss3.r_wrap_mode = GEN7_TEXCOORDMODE_CLAMP; + ss->ss3.s_wrap_mode = GEN7_TEXCOORDMODE_CLAMP; + ss->ss3.t_wrap_mode = GEN7_TEXCOORDMODE_CLAMP; + ss->ss3.non_normalized_coord = 1; -static void -gen6_emit_sip(struct intel_batchbuffer *batch) { - OUT_BATCH(GEN6_STATE_SIP | 0); - OUT_BATCH(0); + return batch_offset(batch, ss); } static void -gen7_emit_push_constants(struct intel_batchbuffer *batch) { - OUT_BATCH(GEN7_3DSTATE_PUSH_CONSTANT_ALLOC_VS); - OUT_BATCH(0); - OUT_BATCH(GEN7_3DSTATE_PUSH_CONSTANT_ALLOC_HS); - OUT_BATCH(0); - OUT_BATCH(GEN7_3DSTATE_PUSH_CONSTANT_ALLOC_DS); - OUT_BATCH(0); - OUT_BATCH(GEN7_3DSTATE_PUSH_CONSTANT_ALLOC_GS); - OUT_BATCH(0); - OUT_BATCH(GEN7_3DSTATE_PUSH_CONSTANT_ALLOC_PS); - OUT_BATCH(0); +gen7_emit_sampler(struct intel_batchbuffer *batch) +{ + OUT_BATCH(GEN7_3DSTATE_SAMPLER_STATE_POINTERS_PS | (2 - 2)); + OUT_BATCH(gen7_create_sampler(batch)); } static void -gen7_emit_state_base_address(struct intel_batchbuffer *batch) { - OUT_BATCH(GEN6_STATE_BASE_ADDRESS | (10 - 2)); - /* general (stateless) */ - /* surface */ - /* instruction */ - /* indirect */ - /* dynamic */ - OUT_BATCH(0 | BASE_ADDRESS_MODIFY); - OUT_RELOC(batch->bo, I915_GEM_DOMAIN_SAMPLER, 0, BASE_ADDRESS_MODIFY); - OUT_RELOC(batch->bo, I915_GEM_DOMAIN_RENDER | I915_GEM_DOMAIN_INSTRUCTION, - 0, BASE_ADDRESS_MODIFY); - OUT_BATCH(0 | BASE_ADDRESS_MODIFY); - OUT_RELOC(batch->bo, I915_GEM_DOMAIN_INSTRUCTION, 0, BASE_ADDRESS_MODIFY); +gen7_emit_multisample(struct intel_batchbuffer *batch) +{ + OUT_BATCH(GEN7_3DSTATE_MULTISAMPLE | (4 - 2)); + OUT_BATCH(GEN7_3DSTATE_MULTISAMPLE_PIXEL_LOCATION_CENTER | + GEN7_3DSTATE_MULTISAMPLE_NUMSAMPLES_1); /* 1 sample/pixel */ + OUT_BATCH(0); + OUT_BATCH(0); - OUT_BATCH(0 | BASE_ADDRESS_MODIFY); - OUT_BATCH(0xfffff000 | BASE_ADDRESS_MODIFY); // copied from mesa - OUT_BATCH(0 | BASE_ADDRESS_MODIFY); - OUT_BATCH(0 | BASE_ADDRESS_MODIFY); + OUT_BATCH(GEN7_3DSTATE_SAMPLE_MASK | (2 - 2)); + OUT_BATCH(1); } static void -gen7_emit_urb(struct intel_batchbuffer *batch) { - /* XXX: Min valid values from mesa */ - const int vs_entries = 32; - const int vs_size = 2; - const int vs_start = 2; - - OUT_BATCH(GEN7_3DSTATE_URB_VS); - OUT_BATCH(vs_entries | ((vs_size - 1) << 16) | (vs_start << 25)); - OUT_BATCH(GEN7_3DSTATE_URB_GS); - OUT_BATCH(vs_start << 25); - OUT_BATCH(GEN7_3DSTATE_URB_HS); - OUT_BATCH(vs_start << 25); - OUT_BATCH(GEN7_3DSTATE_URB_DS); - OUT_BATCH(vs_start << 25); -} +gen7_emit_urb(struct intel_batchbuffer *batch) +{ + OUT_BATCH(GEN7_3DSTATE_PUSH_CONSTANT_ALLOC_PS | (2 - 2)); + OUT_BATCH(8); /* in 1KBs */ -static void -gen7_emit_cc(struct intel_batchbuffer *batch) { - OUT_BATCH(GEN7_3DSTATE_BLEND_STATE_POINTERS); - OUT_BATCH(cc.blend_state | 1); + /* num of VS entries must be divisible by 8 if size < 9 */ + OUT_BATCH(GEN7_3DSTATE_URB_VS | (2 - 2)); + OUT_BATCH((64 << GEN7_URB_ENTRY_NUMBER_SHIFT) | + (2 - 1) << GEN7_URB_ENTRY_SIZE_SHIFT | + (1 << GEN7_URB_STARTING_ADDRESS_SHIFT)); - OUT_BATCH(GEN6_3DSTATE_CC_STATE_POINTERS); - OUT_BATCH(cc.cc_state | 1); + OUT_BATCH(GEN7_3DSTATE_URB_HS | (2 - 2)); + OUT_BATCH((0 << GEN7_URB_ENTRY_SIZE_SHIFT) | + (2 << GEN7_URB_STARTING_ADDRESS_SHIFT)); - OUT_BATCH(GEN7_3DSTATE_DS_STATE_POINTERS); - OUT_BATCH(cc.ds_state | 1); + OUT_BATCH(GEN7_3DSTATE_URB_DS | (2 - 2)); + OUT_BATCH((0 << GEN7_URB_ENTRY_SIZE_SHIFT) | + (2 << GEN7_URB_STARTING_ADDRESS_SHIFT)); + + OUT_BATCH(GEN7_3DSTATE_URB_GS | (2 - 2)); + OUT_BATCH((0 << GEN7_URB_ENTRY_SIZE_SHIFT) | + (1 << GEN7_URB_STARTING_ADDRESS_SHIFT)); } static void -gen7_emit_multisample(struct intel_batchbuffer *batch) { - OUT_BATCH(GEN6_3DSTATE_MULTISAMPLE | 2); +gen7_emit_vs(struct intel_batchbuffer *batch) +{ + OUT_BATCH(GEN7_3DSTATE_VS | (6 - 2)); + OUT_BATCH(0); /* no VS kernel */ OUT_BATCH(0); OUT_BATCH(0); OUT_BATCH(0); - - OUT_BATCH(GEN6_3DSTATE_SAMPLE_MASK); - OUT_BATCH(1); + OUT_BATCH(0); /* pass-through */ } static void -gen7_emit_vs(struct intel_batchbuffer *batch) { - OUT_BATCH(GEN7_3DSTATE_BINDING_TABLE_POINTERS_VS); - OUT_BATCH(0); - - OUT_BATCH(GEN7_3DSTATE_SAMPLER_STATE_POINTERS_VS); - OUT_BATCH(0); - - OUT_BATCH(GEN6_3DSTATE_CONSTANT_VS | (7-2)); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - - OUT_BATCH(GEN6_3DSTATE_VS | (6-2)); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); +gen7_emit_hs(struct intel_batchbuffer *batch) +{ + OUT_BATCH(GEN7_3DSTATE_HS | (7 - 2)); + OUT_BATCH(0); /* no HS kernel */ + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); /* pass-through */ } static void -gen7_emit_hs(struct intel_batchbuffer *batch) { - OUT_BATCH(GEN7_3DSTATE_CONSTANT_HS | (7-2)); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - - OUT_BATCH(GEN7_3DSTATE_HS | (7-2)); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - - OUT_BATCH(GEN7_3DSTATE_BINDING_TABLE_POINTERS_HS); - OUT_BATCH(0); - - OUT_BATCH(GEN7_3DSTATE_SAMPLER_STATE_POINTERS_HS); - OUT_BATCH(0); +gen7_emit_te(struct intel_batchbuffer *batch) +{ + OUT_BATCH(GEN7_3DSTATE_TE | (4 - 2)); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); } static void -gen7_emit_gs(struct intel_batchbuffer *batch) { - OUT_BATCH(GEN7_3DSTATE_CONSTANT_GS | (7-2)); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - - OUT_BATCH(GEN7_3DSTATE_GS | (7-2)); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - - OUT_BATCH(GEN7_3DSTATE_BINDING_TABLE_POINTERS_GS); - OUT_BATCH(0); - - OUT_BATCH(GEN7_3DSTATE_SAMPLER_STATE_POINTERS_GS); - OUT_BATCH(0); +gen7_emit_ds(struct intel_batchbuffer *batch) +{ + OUT_BATCH(GEN7_3DSTATE_DS | (6 - 2)); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); } static void -gen7_emit_ds(struct intel_batchbuffer *batch) { - OUT_BATCH(GEN7_3DSTATE_CONSTANT_DS | (7-2)); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - - OUT_BATCH(GEN7_3DSTATE_DS | (6-2)); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - - OUT_BATCH(GEN7_3DSTATE_BINDING_TABLE_POINTERS_DS); - OUT_BATCH(0); - - OUT_BATCH(GEN7_3DSTATE_SAMPLER_STATE_POINTERS_DS); - OUT_BATCH(0); +gen7_emit_gs(struct intel_batchbuffer *batch) +{ + OUT_BATCH(GEN7_3DSTATE_GS | (7 - 2)); + OUT_BATCH(0); /* no GS kernel */ + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); /* pass-through */ } static void -gen7_emit_null_state(struct intel_batchbuffer *batch) { - gen7_emit_hs(batch); - OUT_BATCH(GEN7_3DSTATE_TE | (4-2)); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - gen7_emit_gs(batch); - gen7_emit_ds(batch); - gen7_emit_vs(batch); +gen7_emit_streamout(struct intel_batchbuffer *batch) +{ + OUT_BATCH(GEN7_3DSTATE_STREAMOUT | (3 - 2)); + OUT_BATCH(0); + OUT_BATCH(0); } static void -gen7_emit_clip(struct intel_batchbuffer *batch) { - OUT_BATCH(GEN6_3DSTATE_CLIP | (4 - 2)); - OUT_BATCH(0); - OUT_BATCH(0); /* pass-through */ - OUT_BATCH(0); +gen7_emit_sf(struct intel_batchbuffer *batch) +{ + OUT_BATCH(GEN7_3DSTATE_SF | (7 - 2)); + OUT_BATCH(0); + OUT_BATCH(GEN7_3DSTATE_SF_CULL_NONE); + OUT_BATCH(2 << GEN7_3DSTATE_SF_TRIFAN_PROVOKE_SHIFT); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); } static void -gen7_emit_sf(struct intel_batchbuffer *batch) { +gen7_emit_sbe(struct intel_batchbuffer *batch) +{ OUT_BATCH(GEN7_3DSTATE_SBE | (14 - 2)); -#ifdef GPU_HANG - OUT_BATCH(0 << 22 | 1 << 11 | 1 << 4); -#else - OUT_BATCH(1 << 22 | 1 << 11 | 1 << 4); -#endif - OUT_BATCH(0); - OUT_BATCH(0); + OUT_BATCH(1 << GEN7_SBE_NUM_OUTPUTS_SHIFT | + 1 << GEN7_SBE_URB_ENTRY_READ_LENGTH_SHIFT | + 1 << GEN7_SBE_URB_ENTRY_READ_OFFSET_SHIFT); OUT_BATCH(0); + OUT_BATCH(0); /* dw4 */ OUT_BATCH(0); OUT_BATCH(0); OUT_BATCH(0); + OUT_BATCH(0); /* dw8 */ OUT_BATCH(0); OUT_BATCH(0); OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - - OUT_BATCH(GEN6_3DSTATE_SF | (7 - 2)); - OUT_BATCH(0); - OUT_BATCH(GEN6_3DSTATE_SF_CULL_NONE); -// OUT_BATCH(2 << GEN6_3DSTATE_SF_TRIFAN_PROVOKE_SHIFT); - OUT_BATCH(0); - OUT_BATCH(0); + OUT_BATCH(0); /* dw12 */ OUT_BATCH(0); OUT_BATCH(0); } static void -gen7_emit_ps(struct intel_batchbuffer *batch, uint32_t kernel) { - const int max_threads = 86; - - OUT_BATCH(GEN6_3DSTATE_WM | (3 - 2)); - OUT_BATCH(GEN7_WM_DISPATCH_ENABLE | - /* XXX: I don't understand the BARYCENTRIC stuff, but it - * appears we need it to put our setup data in the place we - * expect (g6, see below) */ - GEN7_3DSTATE_PS_PERSPECTIVE_PIXEL_BARYCENTRIC); - OUT_BATCH(0); +gen7_emit_ps(struct intel_batchbuffer *batch) +{ + int threads; - OUT_BATCH(GEN6_3DSTATE_CONSTANT_PS | (7-2)); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); + if (IS_HASWELL(batch->devid)) + threads = 40 << HSW_PS_MAX_THREADS_SHIFT | 1 << HSW_PS_SAMPLE_MASK_SHIFT; + else + threads = 40 << IVB_PS_MAX_THREADS_SHIFT; + + OUT_BATCH(GEN7_3DSTATE_PS | (8 - 2)); + OUT_BATCH(batch_copy(batch, ps_kernel, sizeof(ps_kernel), 64)); + OUT_BATCH(1 << GEN7_PS_SAMPLER_COUNT_SHIFT | + 2 << GEN7_PS_BINDING_TABLE_ENTRY_COUNT_SHIFT); + OUT_BATCH(0); /* scratch address */ + OUT_BATCH(threads | + GEN7_PS_16_DISPATCH_ENABLE | + GEN7_PS_ATTRIBUTE_ENABLE); + OUT_BATCH(6 << GEN7_PS_DISPATCH_START_GRF_SHIFT_0); OUT_BATCH(0); OUT_BATCH(0); - - OUT_BATCH(GEN7_3DSTATE_PS | (8-2)); - OUT_BATCH(kernel); - OUT_BATCH(1 << GEN6_3DSTATE_WM_SAMPLER_COUNT_SHITF | - 2 << GEN6_3DSTATE_WM_BINDING_TABLE_ENTRY_COUNT_SHIFT); - OUT_BATCH(0); /* scratch space stuff */ - if (IS_HASWELL(batch->devid)) { - OUT_BATCH((max_threads - 1) << GEN7_3DSTATE_WM_MAX_THREADS_SHIFT | - GEN7_3DSTATE_PS_ATTRIBUTE_ENABLED | - GEN6_3DSTATE_WM_16_DISPATCH_ENABLE); - } else { - OUT_BATCH((max_threads - 1) << HSW_3DSTATE_WM_MAX_THREADS_SHIFT | - GEN7_3DSTATE_PS_ATTRIBUTE_ENABLED | - GEN6_3DSTATE_WM_16_DISPATCH_ENABLE); - } - OUT_BATCH(6 << GEN6_3DSTATE_WM_DISPATCH_START_GRF_0_SHIFT); - OUT_BATCH(0); // kernel 1 - OUT_BATCH(0); // kernel 2 } static void -gen7_emit_depth(struct intel_batchbuffer *batch) { - OUT_BATCH(GEN7_3DSTATE_DEPTH_BUFFER | (7-2)); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - OUT_BATCH(0); - - OUT_BATCH(GEN7_3DSTATE_HIER_DEPTH_BUFFER | (3-2)); - OUT_BATCH(0); - OUT_BATCH(0); +gen7_emit_clip(struct intel_batchbuffer *batch) +{ + OUT_BATCH(GEN7_3DSTATE_CLIP | (4 - 2)); + OUT_BATCH(0); + OUT_BATCH(0); /* pass-through */ + OUT_BATCH(0); - OUT_BATCH(GEN7_3DSTATE_STENCIL_BUFFER | (3-2)); - OUT_BATCH(0); - OUT_BATCH(0); + OUT_BATCH(GEN7_3DSTATE_VIEWPORT_STATE_POINTERS_SF_CL | (2 - 2)); + OUT_BATCH(0); } static void -gen7_emit_clear(struct intel_batchbuffer *batch) { - OUT_BATCH(GEN7_3DSTATE_CLEAR_PARAMS | (3-2)); - OUT_BATCH(0); - OUT_BATCH(1); // clear valid +gen7_emit_wm(struct intel_batchbuffer *batch) +{ + OUT_BATCH(GEN7_3DSTATE_WM | (3 - 2)); + OUT_BATCH(GEN7_WM_DISPATCH_ENABLE | + GEN7_WM_PERSPECTIVE_PIXEL_BARYCENTRIC); + OUT_BATCH(0); } static void -gen6_emit_drawing_rectangle(struct intel_batchbuffer *batch, struct scratch_buf *dst) +gen7_emit_null_depth_buffer(struct intel_batchbuffer *batch) { - OUT_BATCH(GEN6_3DSTATE_DRAWING_RECTANGLE | (4 - 2)); - OUT_BATCH(0); - OUT_BATCH((buf_height(dst) - 1) << 16 | (buf_width(dst) - 1)); - OUT_BATCH(0); -} + OUT_BATCH(GEN7_3DSTATE_DEPTH_BUFFER | (7 - 2)); + OUT_BATCH(GEN7_SURFACE_NULL << GEN7_3DSTATE_DEPTH_BUFFER_TYPE_SHIFT | + GEN7_DEPTHFORMAT_D32_FLOAT << GEN7_3DSTATE_DEPTH_BUFFER_FORMAT_SHIFT); + OUT_BATCH(0); /* disable depth, stencil and hiz */ + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); -/* Vertex elements MUST be defined before this according to spec */ -static void gen7_emit_primitive(struct intel_batchbuffer *batch, uint32_t offset) -{ - OUT_BATCH(GEN6_3DPRIMITIVE | (7-2)); - OUT_BATCH(_3DPRIM_RECTLIST); - OUT_BATCH(3); /* vertex count */ - OUT_BATCH(0); /* We're specifying this instead with offset in GEN6_3DSTATE_VERTEX_BUFFERS */ - OUT_BATCH(1); /* single instance */ - OUT_BATCH(0); /* start instance location */ - OUT_BATCH(0); /* index buffer offset, ignored */ -} - -/* The general rule is if it's named gen6 it is directly copied from - * gen6_render_copyfunc. - * - * This sets up most of the 3d pipeline, and most of that to NULL state. The - * docs aren't specific about exactly what must be set up NULL, but the general - * rule is we could be run at any time, and so the most state we set to NULL, - * the better our odds of success. - * - * +---------------+ <---- 4096 - * | ^ | - * | | | - * | various | - * | state | - * | | | - * |_______|_______| <---- 2048 + ? - * | ^ | - * | | | - * | batch | - * | commands | - * | | | - * | | | - * +---------------+ <---- 0 + ? - * - * The batch commands point to state within tthe batch, so all state offsets should be - * 0 < offset < 4096. Both commands and state build upwards, and are constructed - * in that order. This means too many batch commands can delete state if not - * careful. - * - */ + OUT_BATCH(GEN7_3DSTATE_CLEAR_PARAMS | (3 - 2)); + OUT_BATCH(0); + OUT_BATCH(0); +} #define BATCH_STATE_SPLIT 2048 void gen7_render_copyfunc(struct intel_batchbuffer *batch, @@ -699,103 +517,52 @@ void gen7_render_copyfunc(struct intel_batchbuffer *batch, unsigned width, unsigned height, struct scratch_buf *dst, unsigned dst_x, unsigned dst_y) { - uint32_t ps_sampler_state, ps_kernel_off, ps_binding_table; - uint32_t scissor_state; - uint32_t vertex_buffer; uint32_t batch_end; intel_batchbuffer_flush(batch); - batch_align(batch, 8); - - batch->ptr = &batch->buffer[BATCH_STATE_SPLIT]; - - ps_binding_table = gen7_bind_surfaces(batch, src, dst); - ps_sampler_state = gen7_create_sampler(batch); - ps_kernel_off = batch_copy(batch, ps_kernel, sizeof(ps_kernel), 64); - vertex_buffer = gen7_fill_vertex_buffer_data(batch, src, src_x, src_y, dst_x, dst_y, width, height); - cc.cc_state = gen6_create_cc_state(batch); - cc.ds_state = gen6_create_depth_stencil_state(batch); - cc.blend_state = gen6_create_blend_state(batch); - viewport.cc_state = gen6_create_cc_viewport(batch); - viewport.sf_clip_state = gen7_create_sf_clip_viewport(batch); - scissor_state = gen6_create_scissor_rect(batch); - /* TODO: theree is other state which isn't setup */ - - assert(batch->ptr < &batch->buffer[4095]); - - batch->ptr = batch->buffer; + batch->state = &batch->buffer[BATCH_STATE_SPLIT]; - /* Start emitting the commands. The order roughly follows the mesa blorp - * order */ - OUT_BATCH(GEN6_PIPELINE_SELECT | PIPELINE_SELECT_3D); - - gen6_emit_sip(batch); - - gen7_emit_push_constants(batch); + OUT_BATCH(GEN7_PIPELINE_SELECT | PIPELINE_SELECT_3D); gen7_emit_state_base_address(batch); - - OUT_BATCH(GEN7_3DSTATE_VIEWPORT_STATE_POINTERS_CC); - OUT_BATCH(viewport.cc_state); - OUT_BATCH(GEN7_3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP); - OUT_BATCH(viewport.sf_clip_state); - - gen7_emit_urb(batch); - - gen7_emit_cc(batch); - gen7_emit_multisample(batch); - - gen7_emit_null_state(batch); - - OUT_BATCH(GEN7_3DSTATE_STREAMOUT | 1); - OUT_BATCH(0); - OUT_BATCH(0); - + gen7_emit_urb(batch); + gen7_emit_vs(batch); + gen7_emit_hs(batch); + gen7_emit_te(batch); + gen7_emit_ds(batch); + gen7_emit_gs(batch); gen7_emit_clip(batch); - gen7_emit_sf(batch); + gen7_emit_wm(batch); + gen7_emit_streamout(batch); + gen7_emit_null_depth_buffer(batch); - OUT_BATCH(GEN7_3DSTATE_BINDING_TABLE_POINTERS_PS); - OUT_BATCH(ps_binding_table); - - OUT_BATCH(GEN7_3DSTATE_SAMPLER_STATE_POINTERS_PS); - OUT_BATCH(ps_sampler_state); - - gen7_emit_ps(batch, ps_kernel_off); - - OUT_BATCH(GEN6_3DSTATE_SCISSOR_STATE_POINTERS); - OUT_BATCH(scissor_state); - - gen7_emit_depth(batch); - - gen7_emit_clear(batch); - - gen6_emit_drawing_rectangle(batch, dst); - - gen7_emit_vertex_buffer(batch, vertex_buffer); - gen6_emit_vertex_elements(batch); - - gen7_emit_primitive(batch, vertex_buffer); + gen7_emit_cc(batch); + gen7_emit_sampler(batch); + gen7_emit_sbe(batch); + gen7_emit_ps(batch); + gen7_emit_vertex_elements(batch); + gen7_emit_vertex_buffer(batch, + src_x, src_y, dst_x, dst_y, width, height); + gen7_emit_binding_table(batch, src, dst); + gen7_emit_drawing_rectangle(batch, dst); + + OUT_BATCH(GEN7_3DPRIMITIVE | (7- 2)); + OUT_BATCH(GEN7_3DPRIMITIVE_VERTEX_SEQUENTIAL | _3DPRIM_RECTLIST); + OUT_BATCH(3); + OUT_BATCH(0); + OUT_BATCH(1); /* single instance */ + OUT_BATCH(0); /* start instance location */ + OUT_BATCH(0); /* index buffer offset, ignored */ OUT_BATCH(MI_BATCH_BUFFER_END); - batch_end = batch_align(batch, 8); + batch_end = batch->ptr - batch->buffer; + batch_end = ALIGN(batch_end, 8); assert(batch_end < BATCH_STATE_SPLIT); - dump_batch(batch); - - gen6_render_flush(batch, batch_end); + gen7_render_flush(batch, batch_end); intel_batchbuffer_reset(batch); } - -#if DEBUG_RENDERCPY -static void dump_batch(struct intel_batchbuffer *batch) { - int fd = open("/tmp/i965-batchbuffers.dump", O_WRONLY | O_CREAT, 0666); - if (fd != -1) { - write(fd, batch->buffer, 4096); - fd = close(fd); - } -} -#endif |