diff options
| author | Chris Wilson <chris@chris-wilson.co.uk> | 2011-05-22 10:34:12 +0100 |
|---|---|---|
| committer | Chris Wilson <chris@chris-wilson.co.uk> | 2011-05-24 22:08:48 +0100 |
| commit | 719ffef7c39386c453614bc71e1af97904d3671b (patch) | |
| tree | b441f3febb142a910be8e33e2318360146e6a21b | |
| parent | d73cdde45afc41fb7b36829eda1af763af682ada (diff) | |
gem_stress: Add render copyfunc for SandyBridge
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
| -rw-r--r-- | lib/intel_batchbuffer.c | 7 | ||||
| -rw-r--r-- | lib/intel_batchbuffer.h | 13 | ||||
| -rw-r--r-- | lib/intel_reg.h | 260 | ||||
| -rw-r--r-- | tests/Makefile.am | 5 | ||||
| -rw-r--r-- | tests/gem_stress.c | 51 | ||||
| -rw-r--r-- | tests/gem_stress.h | 8 | ||||
| -rw-r--r-- | tests/gem_stress_gen6.c | 595 | ||||
| -rw-r--r-- | tests/gen6_render.h | 1547 |
8 files changed, 2209 insertions, 277 deletions
diff --git a/lib/intel_batchbuffer.c b/lib/intel_batchbuffer.c index 111f65d4..ffdc36c2 100644 --- a/lib/intel_batchbuffer.c +++ b/lib/intel_batchbuffer.c @@ -197,3 +197,10 @@ intel_copy_bo(struct intel_batchbuffer *batch, intel_batchbuffer_flush(batch); } + +void +intel_batchbuffer_emit_mi_flush(struct intel_batchbuffer *batch) +{ + intel_batchbuffer_require_space(batch, 4); + intel_batchbuffer_emit_dword(batch, MI_FLUSH); +} diff --git a/lib/intel_batchbuffer.h b/lib/intel_batchbuffer.h index 7ed110e3..2beccc4e 100644 --- a/lib/intel_batchbuffer.h +++ b/lib/intel_batchbuffer.h @@ -3,13 +3,11 @@ #include <assert.h> #include "intel_bufmgr.h" -#include "intel_reg.h" #define BATCH_SZ 4096 #define BATCH_RESERVED 16 -struct intel_batchbuffer -{ +struct intel_batchbuffer { drm_intel_bufmgr *bufmgr; uint32_t devid; @@ -93,13 +91,8 @@ intel_batchbuffer_require_space(struct intel_batchbuffer *batch, #define ADVANCE_BATCH() do { \ } while(0) - -static inline void -intel_batchbuffer_emit_mi_flush(struct intel_batchbuffer *batch) -{ - intel_batchbuffer_require_space(batch, 4); - intel_batchbuffer_emit_dword(batch, MI_FLUSH); -} +void +intel_batchbuffer_emit_mi_flush(struct intel_batchbuffer *batch); void intel_copy_bo(struct intel_batchbuffer *batch, drm_intel_bo *dst_bo, drm_intel_bo *src_bo, diff --git a/lib/intel_reg.h b/lib/intel_reg.h index eb09a89f..48d8f66d 100644 --- a/lib/intel_reg.h +++ b/lib/intel_reg.h @@ -2694,27 +2694,17 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #define XY_MONO_SRC_BLT_WRITE_ALPHA (1<<21) #define XY_MONO_SRC_BLT_WRITE_RGB (1<<20) -#define STATE3D_FOG_MODE ((3<<29)|(0x1d<<24)|(0x89<<16)|2) -#define FOG_MODE_VERTEX (1<<31) - -#define DISABLE_TEX_TRANSFORM (1<<28) -#define TEXTURE_SET(x) (x<<29) - -#define STATE3D_VERTEX_TRANSFORM ((3<<29)|(0x1d<<24)|(0x8b<<16)) -#define DISABLE_VIEWPORT_TRANSFORM (1<<31) -#define DISABLE_PERSPECTIVE_DIVIDE (1<<29) - #define MI_STORE_DWORD_IMM ((0x20<<23)|2) #define MI_MEM_VIRTUAL (1 << 22) /* 965+ only */ #define MI_SET_CONTEXT (0x18<<23) -#define CTXT_NO_RESTORE (1) +#define CTXT_NO_RESTORE (1) #define CTXT_PALETTE_SAVE_DISABLE (1<<3) #define CTXT_PALETTE_RESTORE_DISABLE (1<<2) /* Dword 0 */ #define MI_VERTEX_BUFFER (0x17<<23) -#define MI_VERTEX_BUFFER_IDX(x) (x<<20) +#define MI_VERTEX_BUFFER_IDX(x) (x<<20) #define MI_VERTEX_BUFFER_PITCH(x) (x<<13) #define MI_VERTEX_BUFFER_WIDTH(x) (x<<6) /* Dword 1 */ @@ -2763,252 +2753,6 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #define MI_BATCH_NON_SECURE (1) #define MI_BATCH_NON_SECURE_I965 (1 << 8) -/* STATE3D_FOG_MODE stuff */ -#define ENABLE_FOG_SOURCE (1<<27) -#define ENABLE_FOG_CONST (1<<24) -#define ENABLE_FOG_DENSITY (1<<23) - -/* - * New regs for broadwater -- we need to split this file up sensibly somehow. - */ -#define BRW_3D(Pipeline,Opcode,Subopcode) ((3 << 29) | \ - ((Pipeline) << 27) | \ - ((Opcode) << 24) | \ - ((Subopcode) << 16)) - -#define BRW_URB_FENCE BRW_3D(0, 0, 0) -#define BRW_CS_URB_STATE BRW_3D(0, 0, 1) -#define BRW_CONSTANT_BUFFER BRW_3D(0, 0, 2) -#define BRW_STATE_PREFETCH BRW_3D(0, 0, 3) - -#define BRW_STATE_BASE_ADDRESS BRW_3D(0, 1, 1) -#define BRW_STATE_SIP BRW_3D(0, 1, 2) -#define BRW_PIPELINE_SELECT BRW_3D(0, 1, 4) - -#define NEW_PIPELINE_SELECT BRW_3D(1, 1, 4) - -#define BRW_MEDIA_STATE_POINTERS BRW_3D(2, 0, 0) -#define BRW_MEDIA_OBJECT BRW_3D(2, 1, 0) - -#define BRW_3DSTATE_PIPELINED_POINTERS BRW_3D(3, 0, 0) -#define BRW_3DSTATE_BINDING_TABLE_POINTERS BRW_3D(3, 0, 1) -#define BRW_3DSTATE_VERTEX_BUFFERS BRW_3D(3, 0, 8) -#define BRW_3DSTATE_VERTEX_ELEMENTS BRW_3D(3, 0, 9) -#define BRW_3DSTATE_INDEX_BUFFER BRW_3D(3, 0, 0xa) -#define BRW_3DSTATE_VF_STATISTICS BRW_3D(3, 0, 0xb) - -#define BRW_3DSTATE_DRAWING_RECTANGLE BRW_3D(3, 1, 0) -#define BRW_3DSTATE_CONSTANT_COLOR BRW_3D(3, 1, 1) -#define BRW_3DSTATE_SAMPLER_PALETTE_LOAD BRW_3D(3, 1, 2) -#define BRW_3DSTATE_CHROMA_KEY BRW_3D(3, 1, 4) -#define BRW_3DSTATE_DEPTH_BUFFER BRW_3D(3, 1, 5) -#define BRW_3DSTATE_POLY_STIPPLE_OFFSET BRW_3D(3, 1, 6) -#define BRW_3DSTATE_POLY_STIPPLE_PATTERN BRW_3D(3, 1, 7) -#define BRW_3DSTATE_LINE_STIPPLE BRW_3D(3, 1, 8) -#define BRW_3DSTATE_GLOBAL_DEPTH_OFFSET_CLAMP BRW_3D(3, 1, 9) -/* These two are BLC and CTG only, not BW or CL */ -#define BRW_3DSTATE_AA_LINE_PARAMS BRW_3D(3, 1, 0xa) -#define BRW_3DSTATE_GS_SVB_INDEX BRW_3D(3, 1, 0xb) - -#define BRW_PIPE_CONTROL BRW_3D(3, 2, 0) - -#define BRW_3DPRIMITIVE BRW_3D(3, 3, 0) - -#define PIPELINE_SELECT_3D 0 -#define PIPELINE_SELECT_MEDIA 1 - -#define UF0_CS_REALLOC (1 << 13) -#define UF0_VFE_REALLOC (1 << 12) -#define UF0_SF_REALLOC (1 << 11) -#define UF0_CLIP_REALLOC (1 << 10) -#define UF0_GS_REALLOC (1 << 9) -#define UF0_VS_REALLOC (1 << 8) -#define UF1_CLIP_FENCE_SHIFT 20 -#define UF1_GS_FENCE_SHIFT 10 -#define UF1_VS_FENCE_SHIFT 0 -#define UF2_CS_FENCE_SHIFT 20 -#define UF2_VFE_FENCE_SHIFT 10 -#define UF2_SF_FENCE_SHIFT 0 - -/* for BRW_STATE_BASE_ADDRESS */ -#define BASE_ADDRESS_MODIFY (1 << 0) - -/* for BRW_3DSTATE_PIPELINED_POINTERS */ -#define BRW_GS_DISABLE 0 -#define BRW_GS_ENABLE 1 -#define BRW_CLIP_DISABLE 0 -#define BRW_CLIP_ENABLE 1 - -/* for BRW_PIPE_CONTROL */ -#define BRW_PIPE_CONTROL_NOWRITE (0 << 14) -#define BRW_PIPE_CONTROL_WRITE_QWORD (1 << 14) -#define BRW_PIPE_CONTROL_WRITE_DEPTH (2 << 14) -#define BRW_PIPE_CONTROL_WRITE_TIME (3 << 14) -#define BRW_PIPE_CONTROL_DEPTH_STALL (1 << 13) -#define BRW_PIPE_CONTROL_WC_FLUSH (1 << 12) -#define BRW_PIPE_CONTROL_IS_FLUSH (1 << 11) -#define BRW_PIPE_CONTROL_NOTIFY_ENABLE (1 << 8) -#define BRW_PIPE_CONTROL_GLOBAL_GTT (1 << 2) -#define BRW_PIPE_CONTROL_LOCAL_PGTT (0 << 2) - -/* VERTEX_BUFFER_STATE Structure */ -#define VB0_BUFFER_INDEX_SHIFT 27 -#define VB0_VERTEXDATA (0 << 26) -#define VB0_INSTANCEDATA (1 << 26) -#define VB0_BUFFER_PITCH_SHIFT 0 - -/* VERTEX_ELEMENT_STATE Structure */ -#define VE0_VERTEX_BUFFER_INDEX_SHIFT 27 -#define VE0_VALID (1 << 26) -#define VE0_FORMAT_SHIFT 16 -#define VE0_OFFSET_SHIFT 0 -#define VE1_VFCOMPONENT_0_SHIFT 28 -#define VE1_VFCOMPONENT_1_SHIFT 24 -#define VE1_VFCOMPONENT_2_SHIFT 20 -#define VE1_VFCOMPONENT_3_SHIFT 16 -#define VE1_DESTINATION_ELEMENT_OFFSET_SHIFT 0 - -/* 3DPRIMITIVE bits */ -#define BRW_3DPRIMITIVE_VERTEX_SEQUENTIAL (0 << 15) -#define BRW_3DPRIMITIVE_VERTEX_RANDOM (1 << 15) -/* Primitive types are in brw_defines.h */ -#define BRW_3DPRIMITIVE_TOPOLOGY_SHIFT 10 - -#define BRW_SVG_CTL 0x7400 - -#define BRW_SVG_CTL_GS_BA (0 << 8) -#define BRW_SVG_CTL_SS_BA (1 << 8) -#define BRW_SVG_CTL_IO_BA (2 << 8) -#define BRW_SVG_CTL_GS_AUB (3 << 8) -#define BRW_SVG_CTL_IO_AUB (4 << 8) -#define BRW_SVG_CTL_SIP (5 << 8) - -#define BRW_SVG_RDATA 0x7404 -#define BRW_SVG_WORK_CTL 0x7408 - -#define BRW_VF_CTL 0x7500 - -#define BRW_VF_CTL_SNAPSHOT_COMPLETE (1 << 31) -#define BRW_VF_CTL_SNAPSHOT_MUX_SELECT_THREADID (0 << 8) -#define BRW_VF_CTL_SNAPSHOT_MUX_SELECT_VF_DEBUG (1 << 8) -#define BRW_VF_CTL_SNAPSHOT_TYPE_VERTEX_SEQUENCE (0 << 4) -#define BRW_VF_CTL_SNAPSHOT_TYPE_VERTEX_INDEX (1 << 4) -#define BRW_VF_CTL_SKIP_INITIAL_PRIMITIVES (1 << 3) -#define BRW_VF_CTL_MAX_PRIMITIVES_LIMIT_ENABLE (1 << 2) -#define BRW_VF_CTL_VERTEX_RANGE_LIMIT_ENABLE (1 << 1) -#define BRW_VF_CTL_SNAPSHOT_ENABLE (1 << 0) - -#define BRW_VF_STRG_VAL 0x7504 -#define BRW_VF_STR_VL_OVR 0x7508 -#define BRW_VF_VC_OVR 0x750c -#define BRW_VF_STR_PSKIP 0x7510 -#define BRW_VF_MAX_PRIM 0x7514 -#define BRW_VF_RDATA 0x7518 - -#define BRW_VS_CTL 0x7600 -#define BRW_VS_CTL_SNAPSHOT_COMPLETE (1 << 31) -#define BRW_VS_CTL_SNAPSHOT_MUX_VERTEX_0 (0 << 8) -#define BRW_VS_CTL_SNAPSHOT_MUX_VERTEX_1 (1 << 8) -#define BRW_VS_CTL_SNAPSHOT_MUX_VALID_COUNT (2 << 8) -#define BRW_VS_CTL_SNAPSHOT_MUX_VS_KERNEL_POINTER (3 << 8) -#define BRW_VS_CTL_SNAPSHOT_ALL_THREADS (1 << 2) -#define BRW_VS_CTL_THREAD_SNAPSHOT_ENABLE (1 << 1) -#define BRW_VS_CTL_SNAPSHOT_ENABLE (1 << 0) - -#define BRW_VS_STRG_VAL 0x7604 -#define BRW_VS_RDATA 0x7608 - -#define BRW_SF_CTL 0x7b00 -#define BRW_SF_CTL_SNAPSHOT_COMPLETE (1 << 31) -#define BRW_SF_CTL_SNAPSHOT_MUX_VERTEX_0_FF_ID (0 << 8) -#define BRW_SF_CTL_SNAPSHOT_MUX_VERTEX_0_REL_COUNT (1 << 8) -#define BRW_SF_CTL_SNAPSHOT_MUX_VERTEX_1_FF_ID (2 << 8) -#define BRW_SF_CTL_SNAPSHOT_MUX_VERTEX_1_REL_COUNT (3 << 8) -#define BRW_SF_CTL_SNAPSHOT_MUX_VERTEX_2_FF_ID (4 << 8) -#define BRW_SF_CTL_SNAPSHOT_MUX_VERTEX_2_REL_COUNT (5 << 8) -#define BRW_SF_CTL_SNAPSHOT_MUX_VERTEX_COUNT (6 << 8) -#define BRW_SF_CTL_SNAPSHOT_MUX_SF_KERNEL_POINTER (7 << 8) -#define BRW_SF_CTL_MIN_MAX_PRIMITIVE_RANGE_ENABLE (1 << 4) -#define BRW_SF_CTL_DEBUG_CLIP_RECTANGLE_ENABLE (1 << 3) -#define BRW_SF_CTL_SNAPSHOT_ALL_THREADS (1 << 2) -#define BRW_SF_CTL_THREAD_SNAPSHOT_ENABLE (1 << 1) -#define BRW_SF_CTL_SNAPSHOT_ENABLE (1 << 0) - -#define BRW_SF_STRG_VAL 0x7b04 -#define BRW_SF_RDATA 0x7b18 - -#define BRW_WIZ_CTL 0x7c00 -#define BRW_WIZ_CTL_SNAPSHOT_COMPLETE (1 << 31) -#define BRW_WIZ_CTL_SUBSPAN_INSTANCE_SHIFT 16 -#define BRW_WIZ_CTL_SNAPSHOT_MUX_WIZ_KERNEL_POINTER (0 << 8) -#define BRW_WIZ_CTL_SNAPSHOT_MUX_SUBSPAN_INSTANCE (1 << 8) -#define BRW_WIZ_CTL_SNAPSHOT_MUX_PRIMITIVE_SEQUENCE (2 << 8) -#define BRW_WIZ_CTL_SINGLE_SUBSPAN_DISPATCH (1 << 6) -#define BRW_WIZ_CTL_IGNORE_COLOR_SCOREBOARD_STALLS (1 << 5) -#define BRW_WIZ_CTL_ENABLE_SUBSPAN_INSTANCE_COMPARE (1 << 4) -#define BRW_WIZ_CTL_USE_UPSTREAM_SNAPSHOT_FLAG (1 << 3) -#define BRW_WIZ_CTL_SNAPSHOT_ALL_THREADS (1 << 2) -#define BRW_WIZ_CTL_THREAD_SNAPSHOT_ENABLE (1 << 1) -#define BRW_WIZ_CTL_SNAPSHOT_ENABLE (1 << 0) - -#define BRW_WIZ_STRG_VAL 0x7c04 -#define BRW_WIZ_RDATA 0x7c18 - -#define BRW_TS_CTL 0x7e00 -#define BRW_TS_CTL_SNAPSHOT_COMPLETE (1 << 31) -#define BRW_TS_CTL_SNAPSHOT_MESSAGE_ERROR (0 << 8) -#define BRW_TS_CTL_SNAPSHOT_INTERFACE_DESCRIPTOR (3 << 8) -#define BRW_TS_CTL_SNAPSHOT_ALL_CHILD_THREADS (1 << 2) -#define BRW_TS_CTL_SNAPSHOT_ALL_ROOT_THREADS (1 << 1) -#define BRW_TS_CTL_SNAPSHOT_ENABLE (1 << 0) - -#define BRW_TS_STRG_VAL 0x7e04 -#define BRW_TS_RDATA 0x7e08 - -#define BRW_TD_CTL 0x8000 -#define BRW_TD_CTL_MUX_SHIFT 8 -#define BRW_TD_CTL_EXTERNAL_HALT_R0_DEBUG_MATCH (1 << 7) -#define BRW_TD_CTL_FORCE_EXTERNAL_HALT (1 << 6) -#define BRW_TD_CTL_EXCEPTION_MASK_OVERRIDE (1 << 5) -#define BRW_TD_CTL_FORCE_THREAD_BREAKPOINT_ENABLE (1 << 4) -#define BRW_TD_CTL_BREAKPOINT_ENABLE (1 << 2) -#define BRW_TD_CTL2 0x8004 -#define BRW_TD_CTL2_ILLEGAL_OPCODE_EXCEPTION_OVERRIDE (1 << 28) -#define BRW_TD_CTL2_MASKSTACK_EXCEPTION_OVERRIDE (1 << 26) -#define BRW_TD_CTL2_SOFTWARE_EXCEPTION_OVERRIDE (1 << 25) -#define BRW_TD_CTL2_ACTIVE_THREAD_LIMIT_SHIFT 16 -#define BRW_TD_CTL2_ACTIVE_THREAD_LIMIT_ENABLE (1 << 8) -#define BRW_TD_CTL2_THREAD_SPAWNER_EXECUTION_MASK_ENABLE (1 << 7) -#define BRW_TD_CTL2_WIZ_EXECUTION_MASK_ENABLE (1 << 6) -#define BRW_TD_CTL2_SF_EXECUTION_MASK_ENABLE (1 << 5) -#define BRW_TD_CTL2_CLIPPER_EXECUTION_MASK_ENABLE (1 << 4) -#define BRW_TD_CTL2_GS_EXECUTION_MASK_ENABLE (1 << 3) -#define BRW_TD_CTL2_VS_EXECUTION_MASK_ENABLE (1 << 0) -#define BRW_TD_VF_VS_EMSK 0x8008 -#define BRW_TD_GS_EMSK 0x800c -#define BRW_TD_CLIP_EMSK 0x8010 -#define BRW_TD_SF_EMSK 0x8014 -#define BRW_TD_WIZ_EMSK 0x8018 -#define BRW_TD_0_6_EHTRG_VAL 0x801c -#define BRW_TD_0_7_EHTRG_VAL 0x8020 -#define BRW_TD_0_6_EHTRG_MSK 0x8024 -#define BRW_TD_0_7_EHTRG_MSK 0x8028 -#define BRW_TD_RDATA 0x802c -#define BRW_TD_TS_EMSK 0x8030 - -#define BRW_EU_CTL 0x8800 -#define BRW_EU_CTL_SELECT_SHIFT 16 -#define BRW_EU_CTL_DATA_MUX_SHIFT 8 -#define BRW_EU_ATT_0 0x8810 -#define BRW_EU_ATT_1 0x8814 -#define BRW_EU_ATT_DATA_0 0x8820 -#define BRW_EU_ATT_DATA_1 0x8824 -#define BRW_EU_ATT_CLR_0 0x8830 -#define BRW_EU_ATT_CLR_1 0x8834 -#define BRW_EU_RDATA 0x8840 - -/* End regs for broadwater */ - #define MAX_DISPLAY_PIPES 2 typedef enum { diff --git a/tests/Makefile.am b/tests/Makefile.am index fb9966bb..abf5dedd 100644 --- a/tests/Makefile.am +++ b/tests/Makefile.am @@ -7,7 +7,10 @@ gem_stress_SOURCES = \ gem_stress.c \ gem_stress.h \ gem_stress_i915.c \ - gem_stress_i830.c + gem_stress_i830.c \ + gen6_render.h \ + gem_stress_gen6.c \ + $(NULL) TESTS = getversion \ getclient \ diff --git a/tests/gem_stress.c b/tests/gem_stress.c index bf3e2200..46bf4e22 100644 --- a/tests/gem_stress.c +++ b/tests/gem_stress.c @@ -300,6 +300,10 @@ static void render_copyfunc(struct scratch_buf *src, unsigned src_x, unsigned sr gen3_render_copyfunc(src, src_x, src_y, dst, dst_x, dst_y, logical_tile_no); + else if (IS_GEN6(devid)) + gen6_render_copyfunc(src, src_x, src_y, + dst, dst_x, dst_y, + logical_tile_no); else blitter_copyfunc(src, src_x, src_y, dst, dst_x, dst_y, @@ -310,7 +314,7 @@ static void next_copyfunc(int tile) { if (fence_storm) { if (tile == options.trace_tile) - printf(" using fence storm\n"); + printf(" using fence storm\n"); return; } @@ -417,12 +421,12 @@ static void permute_array(void *array, unsigned size, void (*exchange_func)(void *array, unsigned i, unsigned j)) { int i; - long int l; for (i = size - 1; i > 1; i--) { - l = random(); - l %= i+1; /* yes, no perfectly uniform, who cares */ - exchange_func(array, i, l); + /* yes, not perfectly uniform, who cares */ + long l = random() % (i +1); + if (i != l) + exchange_func(array, i, l); } } @@ -437,7 +441,7 @@ static void exchange_buf(void *array, unsigned i, unsigned j) } -/* libdrm is to clever and prevents us from changin tiling of buffers already +/* libdrm is too clever and prevents us from changing tiling of buffers already * used in relocations. */ static void set_tiling(drm_intel_bo *bo, unsigned *tiling, unsigned stride) { @@ -510,7 +514,7 @@ static void init_set(unsigned set) buffers[set][i].stride); if (i == options.trace_tile/TILES_PER_BUF) - printf("changing buffer %i containing tile %i: tiling %i, stride %i\n", i, + printf("changing buffer %i containing tile %i: tiling %i, stride %i\n", i, options.trace_tile, buffers[set][i].tiling, buffers[set][i].stride); } @@ -533,7 +537,7 @@ static void copy_tiles(unsigned *permutation) struct scratch_buf *src_buf, *dst_buf; int i, idx; for (i = 0; i < num_total_tiles; i++) { - /* tile_permutation is independant of current_permutation, so + /* tile_permutation is independent of current_permutation, so * abuse it to randomize the order of the src bos */ idx = tile_permutation[i]; src_buf_idx = idx / TILES_PER_BUF; @@ -721,6 +725,35 @@ static void init(void) srandom(0xdeadbeef); } +static void check_render_copyfunc(void) +{ + struct scratch_buf src, dst; + uint32_t *ptr; + int i, j; + + init_buffer(&src, options.scratch_buf_size); + init_buffer(&dst, options.scratch_buf_size); + + memset(src.data, 0xff, options.scratch_buf_size); + for (j = 0; j < TILE_SIZE; j++) { + ptr = (uint32_t*)((char *)src.data + j * src.stride); + for (i = 0; i < TILE_SIZE; i++) + ptr[i] = j * TILE_SIZE + i; + } + + render_copyfunc(&src, 0, 0, &dst, 0, 0, 0); + + for (j = 0; j < TILE_SIZE; j++) { + ptr = (uint32_t*)((char *)dst.data + j * dst.stride); + for (i = 0; i < TILE_SIZE; i++) + if (ptr[i] != j * TILE_SIZE + i) { + printf("render copyfunc mismatch at (%d, %d): found %d, expected %d\n", + i, j, ptr[i], j*TILE_SIZE + i); + } + } +} + + int main(int argc, char **argv) { int i, j; @@ -730,6 +763,8 @@ int main(int argc, char **argv) init(); + check_render_copyfunc(); + tile_permutation = malloc(num_total_tiles*sizeof(uint32_t)); current_permutation = malloc(num_total_tiles*sizeof(uint32_t)); tmp_permutation = malloc(num_total_tiles*sizeof(uint32_t)); diff --git a/tests/gem_stress.h b/tests/gem_stress.h index d2455065..e91453be 100644 --- a/tests/gem_stress.h +++ b/tests/gem_stress.h @@ -52,6 +52,11 @@ extern int fence_storm; void keep_gpu_busy(void); +static inline void emit_vertex_2s(int16_t x, int16_t y) +{ + OUT_BATCH((uint16_t)y << 16 | (uint16_t)x); +} + static inline void emit_vertex(float f) { union { float f; uint32_t ui; } u; @@ -76,6 +81,9 @@ static inline unsigned buf_height(struct scratch_buf *buf) return options.scratch_buf_size/buf->stride; } +void gen6_render_copyfunc(struct scratch_buf *src, unsigned src_x, unsigned src_y, + struct scratch_buf *dst, unsigned dst_x, unsigned dst_y, + unsigned logical_tile_no); void gen3_render_copyfunc(struct scratch_buf *src, unsigned src_x, unsigned src_y, struct scratch_buf *dst, unsigned dst_x, unsigned dst_y, unsigned logical_tile_no); diff --git a/tests/gem_stress_gen6.c b/tests/gem_stress_gen6.c new file mode 100644 index 00000000..e7fc0a86 --- /dev/null +++ b/tests/gem_stress_gen6.c @@ -0,0 +1,595 @@ +#include "gem_stress.h" +#include "gen6_render.h" + +#include <assert.h> + +#define ALIGN(x, y) (((x) + (y)-1) & ~((y)-1)) +#define VERTEX_SIZE (3*4) + +static const uint32_t ps_kernel_nomask_affine[][4] = { + { 0x0060005a, 0x204077be, 0x000000c0, 0x008d0040 }, + { 0x0060005a, 0x206077be, 0x000000c0, 0x008d0080 }, + { 0x0060005a, 0x208077be, 0x000000d0, 0x008d0040 }, + { 0x0060005a, 0x20a077be, 0x000000d0, 0x008d0080 }, + { 0x00000201, 0x20080061, 0x00000000, 0x00000000 }, + { 0x00600001, 0x20200022, 0x008d0000, 0x00000000 }, + { 0x02800031, 0x21c01cc9, 0x00000020, 0x0a8a0001 }, + { 0x00600001, 0x204003be, 0x008d01c0, 0x00000000 }, + { 0x00600001, 0x206003be, 0x008d01e0, 0x00000000 }, + { 0x00600001, 0x208003be, 0x008d0200, 0x00000000 }, + { 0x00600001, 0x20a003be, 0x008d0220, 0x00000000 }, + { 0x00600001, 0x20c003be, 0x008d0240, 0x00000000 }, + { 0x00600001, 0x20e003be, 0x008d0260, 0x00000000 }, + { 0x00600001, 0x210003be, 0x008d0280, 0x00000000 }, + { 0x00600001, 0x212003be, 0x008d02a0, 0x00000000 }, + { 0x05800031, 0x24001cc8, 0x00000040, 0x90019000 }, + { 0x0000007e, 0x00000000, 0x00000000, 0x00000000 }, + { 0x0000007e, 0x00000000, 0x00000000, 0x00000000 }, + { 0x0000007e, 0x00000000, 0x00000000, 0x00000000 }, + { 0x0000007e, 0x00000000, 0x00000000, 0x00000000 }, + { 0x0000007e, 0x00000000, 0x00000000, 0x00000000 }, + { 0x0000007e, 0x00000000, 0x00000000, 0x00000000 }, + { 0x0000007e, 0x00000000, 0x00000000, 0x00000000 }, + { 0x0000007e, 0x00000000, 0x00000000, 0x00000000 }, +}; + +static uint32_t +batch_used(void) +{ + return batch->ptr - batch->buffer; +} + +static uint32_t +batch_align(uint32_t align) +{ + uint32_t offset = batch_used(); + offset = ALIGN(offset, align); + batch->ptr = batch->buffer + offset; + return offset; +} + +static uint32_t +batch_round_upto(uint32_t div) +{ + uint32_t offset = batch_used(); + offset = (offset + div-1) / div * div; + batch->ptr = batch->buffer + offset; + return offset; +} + +static void * +batch_alloc(uint32_t size, uint32_t align) +{ + uint32_t offset = batch_align(align); + batch->ptr += size; + return memset(batch->buffer + offset, 0, size); +} + +static uint32_t +batch_offset(void *ptr) +{ + return (uint8_t *)ptr - batch->buffer; +} + +static uint32_t +batch_copy(const void *ptr, uint32_t size, uint32_t align) +{ + return batch_offset(memcpy(batch_alloc(size, align), ptr, size)); +} + +static void +gen6_render_flush(uint32_t batch_end) +{ + int ret; + + ret = drm_intel_bo_subdata(batch->bo, 0, 4096, batch->buffer); + if (ret == 0) + ret = drm_intel_bo_mrb_exec(batch->bo, batch_end, + NULL, 0, 0, 0); + assert(ret == 0); +} + +static uint32_t +gen6_bind_buf(struct scratch_buf *buf, + uint32_t format, int is_dst) +{ + struct gen6_surface_state *ss; + uint32_t write_domain, read_domain; + int ret; + + if (is_dst) { + write_domain = read_domain = I915_GEM_DOMAIN_RENDER; + } else { + write_domain = 0; + read_domain = I915_GEM_DOMAIN_SAMPLER; + } + + ss = batch_alloc(sizeof(*ss), 32); + ss->ss0.surface_type = GEN6_SURFACE_2D; + ss->ss0.surface_format = format; + + ss->ss0.data_return_format = GEN6_SURFACERETURNFORMAT_FLOAT32; + ss->ss0.color_blend = 1; + ss->ss1.base_addr = buf->bo->offset; + + ret = drm_intel_bo_emit_reloc(batch->bo, + batch_offset(ss) + 4, + buf->bo, 0, + 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; + ss->ss3.tiled_surface = buf->tiling != I915_TILING_NONE; + ss->ss3.tile_walk = buf->tiling == I915_TILING_Y; + + return batch_offset(ss); +} + +static uint32_t +gen6_bind_surfaces(struct scratch_buf *src, + struct scratch_buf *dst) +{ + uint32_t *binding_table; + + binding_table = batch_alloc(32, 32); + + binding_table[0] = + gen6_bind_buf(dst, GEN6_SURFACEFORMAT_B8G8R8A8_UNORM, 1); + binding_table[1] = + gen6_bind_buf(src, GEN6_SURFACEFORMAT_B8G8R8A8_UNORM, 0); + + return batch_offset(binding_table); +} + +static void +gen6_emit_sip(void) +{ + OUT_BATCH(GEN6_STATE_SIP | 0); + OUT_BATCH(0); +} + +static void +gen6_emit_urb(void) +{ + OUT_BATCH(GEN6_3DSTATE_URB | (3 - 2)); + OUT_BATCH((1 - 1) << GEN6_3DSTATE_URB_VS_SIZE_SHIFT | + 24 << GEN6_3DSTATE_URB_VS_ENTRIES_SHIFT); /* at least 24 on GEN6 */ + OUT_BATCH(0 << GEN6_3DSTATE_URB_GS_SIZE_SHIFT | + 0 << GEN6_3DSTATE_URB_GS_ENTRIES_SHIFT); /* no GS thread */ +} + +static void +gen6_emit_state_base_address(void) +{ + OUT_BATCH(GEN6_STATE_BASE_ADDRESS | (10 - 2)); + OUT_BATCH(0); /* general */ + OUT_RELOC(batch->bo, /* surface */ + I915_GEM_DOMAIN_INSTRUCTION, 0, + BASE_ADDRESS_MODIFY); + OUT_RELOC(batch->bo, /* instruction */ + I915_GEM_DOMAIN_INSTRUCTION, 0, + BASE_ADDRESS_MODIFY); + OUT_BATCH(0); /* indirect */ + OUT_RELOC(batch->bo, /* dynamic */ + I915_GEM_DOMAIN_INSTRUCTION, 0, + BASE_ADDRESS_MODIFY); + + /* upper bounds, disable */ + OUT_BATCH(0); + OUT_BATCH(BASE_ADDRESS_MODIFY); + OUT_BATCH(0); + OUT_BATCH(BASE_ADDRESS_MODIFY); +} + +static void +gen6_emit_viewports(uint32_t cc_vp) +{ + OUT_BATCH(GEN6_3DSTATE_VIEWPORT_STATE_POINTERS | + GEN6_3DSTATE_VIEWPORT_STATE_MODIFY_CC | + (4 - 2)); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(cc_vp); +} + +static void +gen6_emit_vs(void) +{ + /* disable VS constant buffer */ + OUT_BATCH(GEN6_3DSTATE_CONSTANT_VS | (5 - 2)); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + + OUT_BATCH(GEN6_3DSTATE_VS | (6 - 2)); + OUT_BATCH(0); /* no VS kernel */ + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); /* pass-through */ +} + +static void +gen6_emit_gs(void) +{ + /* disable GS constant buffer */ + OUT_BATCH(GEN6_3DSTATE_CONSTANT_GS | (5 - 2)); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + + OUT_BATCH(GEN6_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 +gen6_emit_clip(void) +{ + OUT_BATCH(GEN6_3DSTATE_CLIP | (4 - 2)); + OUT_BATCH(0); + OUT_BATCH(0); /* pass-through */ + OUT_BATCH(0); +} + +static void +gen6_emit_wm_constants(void) +{ + /* disable WM constant buffer */ + OUT_BATCH(GEN6_3DSTATE_CONSTANT_PS | (5 - 2)); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); +} + +static void +gen6_emit_null_depth_buffer(void) +{ + OUT_BATCH(GEN6_3DSTATE_DEPTH_BUFFER | (7 - 2)); + OUT_BATCH(GEN6_SURFACE_NULL << GEN6_3DSTATE_DEPTH_BUFFER_TYPE_SHIFT | + GEN6_DEPTHFORMAT_D32_FLOAT << GEN6_3DSTATE_DEPTH_BUFFER_FORMAT_SHIFT); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + + OUT_BATCH(GEN6_3DSTATE_CLEAR_PARAMS | (2 - 2)); + OUT_BATCH(0); +} + +static void +gen6_emit_invariant(void) +{ + OUT_BATCH(GEN6_PIPELINE_SELECT | PIPELINE_SELECT_3D); + + OUT_BATCH(GEN6_3DSTATE_MULTISAMPLE | (3 - 2)); + OUT_BATCH(GEN6_3DSTATE_MULTISAMPLE_PIXEL_LOCATION_CENTER | + GEN6_3DSTATE_MULTISAMPLE_NUMSAMPLES_1); /* 1 sample/pixel */ + OUT_BATCH(0); + + OUT_BATCH(GEN6_3DSTATE_SAMPLE_MASK | (2 - 2)); + OUT_BATCH(1); +} + +static void +gen6_emit_cc(uint32_t blend) +{ + OUT_BATCH(GEN6_3DSTATE_CC_STATE_POINTERS | (4 - 2)); + OUT_BATCH(blend | 1); + OUT_BATCH(1024 | 1); + OUT_BATCH(1024 | 1); +} + +static void +gen6_emit_sampler(uint32_t state) +{ + OUT_BATCH(GEN6_3DSTATE_SAMPLER_STATE_POINTERS | + GEN6_3DSTATE_SAMPLER_STATE_MODIFY_PS | + (4 - 2)); + OUT_BATCH(0); /* VS */ + OUT_BATCH(0); /* GS */ + OUT_BATCH(state); +} + +static void +gen6_emit_sf(void) +{ + OUT_BATCH(GEN6_3DSTATE_SF | (20 - 2)); + OUT_BATCH(1 << GEN6_3DSTATE_SF_NUM_OUTPUTS_SHIFT | + 1 << GEN6_3DSTATE_SF_URB_ENTRY_READ_LENGTH_SHIFT | + 1 << GEN6_3DSTATE_SF_URB_ENTRY_READ_OFFSET_SHIFT); + OUT_BATCH(0); + OUT_BATCH(GEN6_3DSTATE_SF_CULL_NONE); + OUT_BATCH(2 << GEN6_3DSTATE_SF_TRIFAN_PROVOKE_SHIFT); /* DW4 */ + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); /* DW9 */ + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); /* DW14 */ + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); /* DW19 */ +} + +static void +gen6_emit_wm(int kernel) +{ + OUT_BATCH(GEN6_3DSTATE_WM | (9 - 2)); + OUT_BATCH(kernel); + OUT_BATCH(1 << GEN6_3DSTATE_WM_SAMPLER_COUNT_SHIFT | + 2 << GEN6_3DSTATE_WM_BINDING_TABLE_ENTRY_COUNT_SHIFT); + OUT_BATCH(0); + OUT_BATCH(6 << GEN6_3DSTATE_WM_DISPATCH_START_GRF_0_SHIFT); /* DW4 */ + OUT_BATCH((40 - 1) << GEN6_3DSTATE_WM_MAX_THREADS_SHIFT | + GEN6_3DSTATE_WM_DISPATCH_ENABLE | + GEN6_3DSTATE_WM_16_DISPATCH_ENABLE); + OUT_BATCH(1 << GEN6_3DSTATE_WM_NUM_SF_OUTPUTS_SHIFT | + GEN6_3DSTATE_WM_PERSPECTIVE_PIXEL_BARYCENTRIC); + OUT_BATCH(0); + OUT_BATCH(0); +} + +static void +gen6_emit_binding_table(uint32_t wm_table) +{ + OUT_BATCH(GEN6_3DSTATE_BINDING_TABLE_POINTERS | + GEN6_3DSTATE_BINDING_TABLE_MODIFY_PS | + (4 - 2)); + OUT_BATCH(0); /* vs */ + OUT_BATCH(0); /* gs */ + OUT_BATCH(wm_table); +} + +static void +gen6_emit_drawing_rectangle(struct scratch_buf *dst) +{ + 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); +} + +static void +gen6_emit_vertex_elements(void) +{ + /* The VUE layout + * dword 0-3: pad (0.0, 0.0, 0.0. 0.0) + * dword 4-7: position (x, y, 1.0, 1.0), + * dword 8-11: texture coordinate 0 (u0, v0, 0, 0) + * + * dword 4-11 are fetched from vertex buffer + */ + OUT_BATCH(GEN6_3DSTATE_VERTEX_ELEMENTS | (2 * 3 + 1 - 2)); + + OUT_BATCH(0 << VE0_VERTEX_BUFFER_INDEX_SHIFT | VE0_VALID | + GEN6_SURFACEFORMAT_R32G32B32A32_FLOAT << VE0_FORMAT_SHIFT | + 0 << VE0_OFFSET_SHIFT); + 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); + + /* x,y */ + 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_1_FLT << VE1_VFCOMPONENT_2_SHIFT | + GEN6_VFCOMPONENT_STORE_1_FLT << VE1_VFCOMPONENT_3_SHIFT); + + /* u0, v0 */ + 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_0 << VE1_VFCOMPONENT_3_SHIFT); +} + +static uint32_t +gen6_create_cc_viewport(void) +{ + struct gen6_cc_viewport *vp; + + vp = batch_alloc(sizeof(*vp), 32); + + vp->min_depth = -1.e35; + vp->max_depth = 1.e35; + + return batch_offset(vp); +} + +static uint32_t +gen6_create_cc_blend(void) +{ + struct gen6_blend_state *blend; + + blend = batch_alloc(sizeof(*blend), 64); + + blend->blend0.dest_blend_factor = GEN6_BLENDFACTOR_ZERO; + blend->blend0.source_blend_factor = GEN6_BLENDFACTOR_ONE; + blend->blend0.blend_func = GEN6_BLENDFUNCTION_ADD; + blend->blend0.blend_enable = 1; + + blend->blend1.post_blend_clamp_enable = 1; + blend->blend1.pre_blend_clamp_enable = 1; + + return batch_offset(blend); +} + +static uint32_t +gen6_create_kernel(void) +{ + return batch_copy(ps_kernel_nomask_affine, + sizeof(ps_kernel_nomask_affine), + 64); +} + +static uint32_t +gen6_create_sampler(sampler_filter_t filter, + sampler_extend_t extend) +{ + struct gen6_sampler_state *ss; + + ss = batch_alloc(sizeof(*ss), 32); + ss->ss0.lod_preclamp = 1; /* GL mode */ + + /* We use the legacy mode to get the semantics specified by + * the Render extension. */ + ss->ss0.border_color_mode = GEN6_BORDER_COLOR_MODE_LEGACY; + + switch (filter) { + default: + case SAMPLER_FILTER_NEAREST: + ss->ss0.min_filter = GEN6_MAPFILTER_NEAREST; + ss->ss0.mag_filter = GEN6_MAPFILTER_NEAREST; + break; + case SAMPLER_FILTER_BILINEAR: + ss->ss0.min_filter = GEN6_MAPFILTER_LINEAR; + ss->ss0.mag_filter = GEN6_MAPFILTER_LINEAR; + break; + } + + switch (extend) { + default: + case SAMPLER_EXTEND_NONE: + ss->ss1.r_wrap_mode = GEN6_TEXCOORDMODE_CLAMP_BORDER; + ss->ss1.s_wrap_mode = GEN6_TEXCOORDMODE_CLAMP_BORDER; + ss->ss1.t_wrap_mode = GEN6_TEXCOORDMODE_CLAMP_BORDER; + break; + case SAMPLER_EXTEND_REPEAT: + ss->ss1.r_wrap_mode = GEN6_TEXCOORDMODE_WRAP; + ss->ss1.s_wrap_mode = GEN6_TEXCOORDMODE_WRAP; + ss->ss1.t_wrap_mode = GEN6_TEXCOORDMODE_WRAP; + break; + case SAMPLER_EXTEND_PAD: + ss->ss1.r_wrap_mode = GEN6_TEXCOORDMODE_CLAMP; + ss->ss1.s_wrap_mode = GEN6_TEXCOORDMODE_CLAMP; + ss->ss1.t_wrap_mode = GEN6_TEXCOORDMODE_CLAMP; + break; + case SAMPLER_EXTEND_REFLECT: + ss->ss1.r_wrap_mode = GEN6_TEXCOORDMODE_MIRROR; + ss->ss1.s_wrap_mode = GEN6_TEXCOORDMODE_MIRROR; + ss->ss1.t_wrap_mode = GEN6_TEXCOORDMODE_MIRROR; + break; + } + + return batch_offset(ss); +} + +static void gen6_emit_vertex_buffer(void) +{ + OUT_BATCH(GEN6_3DSTATE_VERTEX_BUFFERS | 3); + OUT_BATCH(VB0_VERTEXDATA | + 0 << VB0_BUFFER_INDEX_SHIFT | + VERTEX_SIZE << VB0_BUFFER_PITCH_SHIFT); + OUT_RELOC(batch->bo, I915_GEM_DOMAIN_VERTEX, 0, 0); + OUT_RELOC(batch->bo, I915_GEM_DOMAIN_VERTEX, 0, batch->bo->size-1); + OUT_BATCH(0); +} + +static uint32_t gen6_emit_primitive(void) +{ + uint32_t offset; + + OUT_BATCH(GEN6_3DPRIMITIVE | + GEN6_3DPRIMITIVE_VERTEX_SEQUENTIAL | + _3DPRIM_RECTLIST << GEN6_3DPRIMITIVE_TOPOLOGY_SHIFT | + 0 << 9 | + 4); + OUT_BATCH(3); /* vertex count */ + offset = batch_used(); + OUT_BATCH(0); /* vertex_index */ + OUT_BATCH(1); /* single instance */ + OUT_BATCH(0); /* start instance location */ + OUT_BATCH(0); /* index buffer offset, ignored */ + + return offset; +} + +void gen6_render_copyfunc(struct scratch_buf *src, unsigned src_x, unsigned src_y, + struct scratch_buf *dst, unsigned dst_x, unsigned dst_y, + unsigned logical_tile_no) +{ + uint32_t wm_state, wm_kernel, wm_table; + uint32_t cc_vp, cc_blend, offset; + uint32_t batch_end; + + intel_batchbuffer_flush(batch); + + batch->ptr = batch->buffer + 1024; + batch_alloc(64, 64); + wm_table = gen6_bind_surfaces(src, dst); + wm_kernel = gen6_create_kernel(); + wm_state = gen6_create_sampler(SAMPLER_FILTER_NEAREST, + SAMPLER_EXTEND_NONE); + + cc_vp = gen6_create_cc_viewport(); + cc_blend = gen6_create_cc_blend(); + + batch->ptr = batch->buffer; + + gen6_emit_invariant(); + gen6_emit_state_base_address(); + + gen6_emit_sip(); + gen6_emit_urb(); + + gen6_emit_viewports(cc_vp); + gen6_emit_vs(); + gen6_emit_gs(); + gen6_emit_clip(); + gen6_emit_wm_constants(); + gen6_emit_null_depth_buffer(); + + gen6_emit_drawing_rectangle(dst); + gen6_emit_cc(cc_blend); + gen6_emit_sampler(wm_state); + gen6_emit_sf(); + gen6_emit_wm(wm_kernel); + gen6_emit_vertex_elements(); + gen6_emit_binding_table(wm_table); + + gen6_emit_vertex_buffer(); + offset = gen6_emit_primitive(); + + OUT_BATCH(MI_BATCH_BUFFER_END); + batch_end = batch_align(8); + + *(uint32_t*)(batch->buffer + offset) = + batch_round_upto(VERTEX_SIZE)/VERTEX_SIZE; + + emit_vertex_2s(dst_x + TILE_SIZE, dst_y + TILE_SIZE); + emit_vertex_normalized(src_x + TILE_SIZE, buf_width(src)); + emit_vertex_normalized(src_y + TILE_SIZE, buf_height(src)); + + emit_vertex_2s(dst_x, dst_y + TILE_SIZE); + emit_vertex_normalized(src_x, buf_width(src)); + emit_vertex_normalized(src_y + TILE_SIZE, buf_height(src)); + + emit_vertex_2s(dst_x, dst_y); + emit_vertex_normalized(src_x, buf_width(src)); + emit_vertex_normalized(src_y, buf_height(src)); + + gen6_render_flush(batch_end); + intel_batchbuffer_reset(batch); +} diff --git a/tests/gen6_render.h b/tests/gen6_render.h new file mode 100644 index 00000000..fa940d9d --- /dev/null +++ b/tests/gen6_render.h @@ -0,0 +1,1547 @@ +#ifndef GEN6_RENDER_H +#define GEN6_RENDER_H + +#define GEN6_3D(Pipeline,Opcode,Subopcode) ((3 << 29) | \ + ((Pipeline) << 27) | \ + ((Opcode) << 24) | \ + ((Subopcode) << 16)) + +#define GEN6_STATE_BASE_ADDRESS GEN6_3D(0, 1, 1) +#define GEN6_STATE_SIP GEN6_3D(0, 1, 2) + +#define GEN6_PIPELINE_SELECT GEN6_3D(1, 1, 4) + +#define GEN6_MEDIA_STATE_POINTERS GEN6_3D(2, 0, 0) +#define GEN6_MEDIA_OBJECT GEN6_3D(2, 1, 0) + +#define GEN6_3DSTATE_BINDING_TABLE_POINTERS GEN6_3D(3, 0, 1) +# define GEN6_3DSTATE_BINDING_TABLE_MODIFY_PS (1 << 12)/* for GEN6 */ +# define GEN6_3DSTATE_BINDING_TABLE_MODIFY_GS (1 << 9) /* for GEN6 */ +# define GEN6_3DSTATE_BINDING_TABLE_MODIFY_VS (1 << 8) /* for GEN6 */ + +#define GEN6_3DSTATE_VERTEX_BUFFERS GEN6_3D(3, 0, 8) +#define GEN6_3DSTATE_VERTEX_ELEMENTS GEN6_3D(3, 0, 9) +#define GEN6_3DSTATE_INDEX_BUFFER GEN6_3D(3, 0, 0xa) +#define GEN6_3DSTATE_VF_STATISTICS GEN6_3D(3, 0, 0xb) + +#define GEN6_3DSTATE_DRAWING_RECTANGLE GEN6_3D(3, 1, 0) +#define GEN6_3DSTATE_CONSTANT_COLOR GEN6_3D(3, 1, 1) +#define GEN6_3DSTATE_SAMPLER_PALETTE_LOAD GEN6_3D(3, 1, 2) +#define GEN6_3DSTATE_CHROMA_KEY GEN6_3D(3, 1, 4) +#define GEN6_3DSTATE_DEPTH_BUFFER GEN6_3D(3, 1, 5) +# define GEN6_3DSTATE_DEPTH_BUFFER_TYPE_SHIFT 29 +# define GEN6_3DSTATE_DEPTH_BUFFER_FORMAT_SHIFT 18 + +#define GEN6_3DSTATE_POLY_STIPPLE_OFFSET GEN6_3D(3, 1, 6) +#define GEN6_3DSTATE_POLY_STIPPLE_PATTERN GEN6_3D(3, 1, 7) +#define GEN6_3DSTATE_LINE_STIPPLE GEN6_3D(3, 1, 8) +#define GEN6_3DSTATE_GLOBAL_DEPTH_OFFSET_CLAMP GEN6_3D(3, 1, 9) +/* These two are BLC and CTG only, not BW or CL */ +#define GEN6_3DSTATE_AA_LINE_PARAMS GEN6_3D(3, 1, 0xa) +#define GEN6_3DSTATE_GS_SVB_INDEX GEN6_3D(3, 1, 0xb) + +#define GEN6_3DPRIMITIVE GEN6_3D(3, 3, 0) + +#define GEN6_3DSTATE_CLEAR_PARAMS GEN6_3D(3, 1, 0x10) +/* DW1 */ +# define GEN6_3DSTATE_DEPTH_CLEAR_VALID (1 << 15) + +#define GEN6_3DSTATE_SAMPLER_STATE_POINTERS GEN6_3D(3, 0, 0x02) +# define GEN6_3DSTATE_SAMPLER_STATE_MODIFY_PS (1 << 12) +# define GEN6_3DSTATE_SAMPLER_STATE_MODIFY_GS (1 << 9) +# define GEN6_3DSTATE_SAMPLER_STATE_MODIFY_VS (1 << 8) + +#define GEN6_3DSTATE_URB GEN6_3D(3, 0, 0x05) +/* DW1 */ +# define GEN6_3DSTATE_URB_VS_SIZE_SHIFT 16 +# define GEN6_3DSTATE_URB_VS_ENTRIES_SHIFT 0 +/* DW2 */ +# define GEN6_3DSTATE_URB_GS_ENTRIES_SHIFT 8 +# define GEN6_3DSTATE_URB_GS_SIZE_SHIFT 0 + +#define GEN6_3DSTATE_VIEWPORT_STATE_POINTERS GEN6_3D(3, 0, 0x0d) +# define GEN6_3DSTATE_VIEWPORT_STATE_MODIFY_CC (1 << 12) +# define GEN6_3DSTATE_VIEWPORT_STATE_MODIFY_SF (1 << 11) +# define GEN6_3DSTATE_VIEWPORT_STATE_MODIFY_CLIP (1 << 10) + +#define GEN6_3DSTATE_CC_STATE_POINTERS GEN6_3D(3, 0, 0x0e) + +#define GEN6_3DSTATE_VS GEN6_3D(3, 0, 0x10) + +#define GEN6_3DSTATE_GS GEN6_3D(3, 0, 0x11) +/* DW4 */ +# define GEN6_3DSTATE_GS_DISPATCH_START_GRF_SHIFT 0 + +#define GEN6_3DSTATE_CLIP GEN6_3D(3, 0, 0x12) + +#define GEN6_3DSTATE_SF GEN6_3D(3, 0, 0x13) +/* DW1 */ +# define GEN6_3DSTATE_SF_NUM_OUTPUTS_SHIFT 22 +# define GEN6_3DSTATE_SF_URB_ENTRY_READ_LENGTH_SHIFT 11 +# define GEN6_3DSTATE_SF_URB_ENTRY_READ_OFFSET_SHIFT 4 +/* DW2 */ +/* DW3 */ +# define GEN6_3DSTATE_SF_CULL_BOTH (0 << 29) +# define GEN6_3DSTATE_SF_CULL_NONE (1 << 29) +# define GEN6_3DSTATE_SF_CULL_FRONT (2 << 29) +# define GEN6_3DSTATE_SF_CULL_BACK (3 << 29) +/* DW4 */ +# define GEN6_3DSTATE_SF_TRI_PROVOKE_SHIFT 29 +# define GEN6_3DSTATE_SF_LINE_PROVOKE_SHIFT 27 +# define GEN6_3DSTATE_SF_TRIFAN_PROVOKE_SHIFT 25 + +#define GEN6_3DSTATE_WM GEN6_3D(3, 0, 0x14) +/* DW2 */ +# define GEN6_3DSTATE_WM_SAMPLER_COUNT_SHIFT 27 +# define GEN6_3DSTATE_WM_BINDING_TABLE_ENTRY_COUNT_SHIFT 18 +/* DW4 */ +# define GEN6_3DSTATE_WM_DISPATCH_START_GRF_0_SHIFT 16 +/* DW5 */ +# define GEN6_3DSTATE_WM_MAX_THREADS_SHIFT 25 +# define GEN6_3DSTATE_WM_DISPATCH_ENABLE (1 << 19) +# define GEN6_3DSTATE_WM_16_DISPATCH_ENABLE (1 << 1) +# define GEN6_3DSTATE_WM_8_DISPATCH_ENABLE (1 << 0) +/* DW6 */ +# define GEN6_3DSTATE_WM_NUM_SF_OUTPUTS_SHIFT 20 +# define GEN6_3DSTATE_WM_NONPERSPECTIVE_SAMPLE_BARYCENTRIC (1 << 15) +# define GEN6_3DSTATE_WM_NONPERSPECTIVE_CENTROID_BARYCENTRIC (1 << 14) +# define GEN6_3DSTATE_WM_NONPERSPECTIVE_PIXEL_BARYCENTRIC (1 << 13) +# define GEN6_3DSTATE_WM_PERSPECTIVE_SAMPLE_BARYCENTRIC (1 << 12) +# define GEN6_3DSTATE_WM_PERSPECTIVE_CENTROID_BARYCENTRIC (1 << 11) +# define GEN6_3DSTATE_WM_PERSPECTIVE_PIXEL_BARYCENTRIC (1 << 10) + + +#define GEN6_3DSTATE_CONSTANT_VS GEN6_3D(3, 0, 0x15) +#define GEN6_3DSTATE_CONSTANT_GS GEN6_3D(3, 0, 0x16) +#define GEN6_3DSTATE_CONSTANT_PS GEN6_3D(3, 0, 0x17) + +#define GEN6_3DSTATE_SAMPLE_MASK GEN6_3D(3, 0, 0x18) + +#define GEN6_3DSTATE_MULTISAMPLE GEN6_3D(3, 1, 0x0d) +/* DW1 */ +# define GEN6_3DSTATE_MULTISAMPLE_PIXEL_LOCATION_CENTER (0 << 4) +# define GEN6_3DSTATE_MULTISAMPLE_PIXEL_LOCATION_UPPER_LEFT (1 << 4) +# define GEN6_3DSTATE_MULTISAMPLE_NUMSAMPLES_1 (0 << 1) +# define GEN6_3DSTATE_MULTISAMPLE_NUMSAMPLES_4 (2 << 1) +# define GEN6_3DSTATE_MULTISAMPLE_NUMSAMPLES_8 (3 << 1) + +#define PIPELINE_SELECT_3D 0 +#define PIPELINE_SELECT_MEDIA 1 + +/* for GEN6_STATE_BASE_ADDRESS */ +#define BASE_ADDRESS_MODIFY (1 << 0) + +/* for GEN6_PIPE_CONTROL */ +#define GEN6_PIPE_CONTROL_NOWRITE (0 << 14) +#define GEN6_PIPE_CONTROL_WRITE_QWORD (1 << 14) +#define GEN6_PIPE_CONTROL_WRITE_DEPTH (2 << 14) +#define GEN6_PIPE_CONTROL_WRITE_TIME (3 << 14) +#define GEN6_PIPE_CONTROL_DEPTH_STALL (1 << 13) +#define GEN6_PIPE_CONTROL_WC_FLUSH (1 << 12) +#define GEN6_PIPE_CONTROL_IS_FLUSH (1 << 11) +#define GEN6_PIPE_CONTROL_TC_FLUSH (1 << 10) +#define GEN6_PIPE_CONTROL_NOTIFY_ENABLE (1 << 8) +#define GEN6_PIPE_CONTROL_GLOBAL_GTT (1 << 2) +#define GEN6_PIPE_CONTROL_LOCAL_PGTT (0 << 2) +#define GEN6_PIPE_CONTROL_DEPTH_CACHE_FLUSH (1 << 0) + +/* VERTEX_BUFFER_STATE Structure */ +#define VB0_BUFFER_INDEX_SHIFT 26 +#define VB0_VERTEXDATA (0 << 20) +#define VB0_INSTANCEDATA (1 << 20) +#define VB0_BUFFER_PITCH_SHIFT 0 + +/* VERTEX_ELEMENT_STATE Structure */ +#define VE0_VERTEX_BUFFER_INDEX_SHIFT 26 /* for GEN6 */ +#define VE0_VALID (1 << 25) /* for GEN6 */ +#define VE0_FORMAT_SHIFT 16 +#define VE0_OFFSET_SHIFT 0 +#define VE1_VFCOMPONENT_0_SHIFT 28 +#define VE1_VFCOMPONENT_1_SHIFT 24 +#define VE1_VFCOMPONENT_2_SHIFT 20 +#define VE1_VFCOMPONENT_3_SHIFT 16 +#define VE1_DESTINATION_ELEMENT_OFFSET_SHIFT 0 + +/* 3DPRIMITIVE bits */ +#define GEN6_3DPRIMITIVE_VERTEX_SEQUENTIAL (0 << 15) +#define GEN6_3DPRIMITIVE_VERTEX_RANDOM (1 << 15) +/* Primitive types are in gen6_defines.h */ +#define GEN6_3DPRIMITIVE_TOPOLOGY_SHIFT 10 + +#define GEN6_SVG_CTL 0x7400 + +#define GEN6_SVG_CTL_GS_BA (0 << 8) +#define GEN6_SVG_CTL_SS_BA (1 << 8) +#define GEN6_SVG_CTL_IO_BA (2 << 8) +#define GEN6_SVG_CTL_GS_AUB (3 << 8) +#define GEN6_SVG_CTL_IO_AUB (4 << 8) +#define GEN6_SVG_CTL_SIP (5 << 8) + +#define GEN6_SVG_RDATA 0x7404 +#define GEN6_SVG_WORK_CTL 0x7408 + +#define GEN6_VF_CTL 0x7500 + +#define GEN6_VF_CTL_SNAPSHOT_COMPLETE (1 << 31) +#define GEN6_VF_CTL_SNAPSHOT_MUX_SELECT_THREADID (0 << 8) +#define GEN6_VF_CTL_SNAPSHOT_MUX_SELECT_VF_DEBUG (1 << 8) +#define GEN6_VF_CTL_SNAPSHOT_TYPE_VERTEX_SEQUENCE (0 << 4) +#define GEN6_VF_CTL_SNAPSHOT_TYPE_VERTEX_INDEX (1 << 4) +#define GEN6_VF_CTL_SKIP_INITIAL_PRIMITIVES (1 << 3) +#define GEN6_VF_CTL_MAX_PRIMITIVES_LIMIT_ENABLE (1 << 2) +#define GEN6_VF_CTL_VERTEX_RANGE_LIMIT_ENABLE (1 << 1) +#define GEN6_VF_CTL_SNAPSHOT_ENABLE (1 << 0) + +#define GEN6_VF_STRG_VAL 0x7504 +#define GEN6_VF_STR_VL_OVR 0x7508 +#define GEN6_VF_VC_OVR 0x750c +#define GEN6_VF_STR_PSKIP 0x7510 +#define GEN6_VF_MAX_PRIM 0x7514 +#define GEN6_VF_RDATA 0x7518 + +#define GEN6_VS_CTL 0x7600 +#define GEN6_VS_CTL_SNAPSHOT_COMPLETE (1 << 31) +#define GEN6_VS_CTL_SNAPSHOT_MUX_VERTEX_0 (0 << 8) +#define GEN6_VS_CTL_SNAPSHOT_MUX_VERTEX_1 (1 << 8) +#define GEN6_VS_CTL_SNAPSHOT_MUX_VALID_COUNT (2 << 8) +#define GEN6_VS_CTL_SNAPSHOT_MUX_VS_KERNEL_POINTER (3 << 8) +#define GEN6_VS_CTL_SNAPSHOT_ALL_THREADS (1 << 2) +#define GEN6_VS_CTL_THREAD_SNAPSHOT_ENABLE (1 << 1) +#define GEN6_VS_CTL_SNAPSHOT_ENABLE (1 << 0) + +#define GEN6_VS_STRG_VAL 0x7604 +#define GEN6_VS_RDATA 0x7608 + +#define GEN6_SF_CTL 0x7b00 +#define GEN6_SF_CTL_SNAPSHOT_COMPLETE (1 << 31) +#define GEN6_SF_CTL_SNAPSHOT_MUX_VERTEX_0_FF_ID (0 << 8) +#define GEN6_SF_CTL_SNAPSHOT_MUX_VERTEX_0_REL_COUNT (1 << 8) +#define GEN6_SF_CTL_SNAPSHOT_MUX_VERTEX_1_FF_ID (2 << 8) +#define GEN6_SF_CTL_SNAPSHOT_MUX_VERTEX_1_REL_COUNT (3 << 8) +#define GEN6_SF_CTL_SNAPSHOT_MUX_VERTEX_2_FF_ID (4 << 8) +#define GEN6_SF_CTL_SNAPSHOT_MUX_VERTEX_2_REL_COUNT (5 << 8) +#define GEN6_SF_CTL_SNAPSHOT_MUX_VERTEX_COUNT (6 << 8) +#define GEN6_SF_CTL_SNAPSHOT_MUX_SF_KERNEL_POINTER (7 << 8) +#define GEN6_SF_CTL_MIN_MAX_PRIMITIVE_RANGE_ENABLE (1 << 4) +#define GEN6_SF_CTL_DEBUG_CLIP_RECTANGLE_ENABLE (1 << 3) +#define GEN6_SF_CTL_SNAPSHOT_ALL_THREADS (1 << 2) +#define GEN6_SF_CTL_THREAD_SNAPSHOT_ENABLE (1 << 1) +#define GEN6_SF_CTL_SNAPSHOT_ENABLE (1 << 0) + +#define GEN6_SF_STRG_VAL 0x7b04 +#define GEN6_SF_RDATA 0x7b18 + +#define GEN6_WIZ_CTL 0x7c00 +#define GEN6_WIZ_CTL_SNAPSHOT_COMPLETE (1 << 31) +#define GEN6_WIZ_CTL_SUBSPAN_INSTANCE_SHIFT 16 +#define GEN6_WIZ_CTL_SNAPSHOT_MUX_WIZ_KERNEL_POINTER (0 << 8) +#define GEN6_WIZ_CTL_SNAPSHOT_MUX_SUBSPAN_INSTANCE (1 << 8) +#define GEN6_WIZ_CTL_SNAPSHOT_MUX_PRIMITIVE_SEQUENCE (2 << 8) +#define GEN6_WIZ_CTL_SINGLE_SUBSPAN_DISPATCH (1 << 6) +#define GEN6_WIZ_CTL_IGNORE_COLOR_SCOREBOARD_STALLS (1 << 5) +#define GEN6_WIZ_CTL_ENABLE_SUBSPAN_INSTANCE_COMPARE (1 << 4) +#define GEN6_WIZ_CTL_USE_UPSTREAM_SNAPSHOT_FLAG (1 << 3) +#define GEN6_WIZ_CTL_SNAPSHOT_ALL_THREADS (1 << 2) +#define GEN6_WIZ_CTL_THREAD_SNAPSHOT_ENABLE (1 << 1) +#define GEN6_WIZ_CTL_SNAPSHOT_ENABLE (1 << 0) + +#define GEN6_WIZ_STRG_VAL 0x7c04 +#define GEN6_WIZ_RDATA 0x7c18 + +#define GEN6_TS_CTL 0x7e00 +#define GEN6_TS_CTL_SNAPSHOT_COMPLETE (1 << 31) +#define GEN6_TS_CTL_SNAPSHOT_MESSAGE_ERROR (0 << 8) +#define GEN6_TS_CTL_SNAPSHOT_INTERFACE_DESCRIPTOR (3 << 8) +#define GEN6_TS_CTL_SNAPSHOT_ALL_CHILD_THREADS (1 << 2) +#define GEN6_TS_CTL_SNAPSHOT_ALL_ROOT_THREADS (1 << 1) +#define GEN6_TS_CTL_SNAPSHOT_ENABLE (1 << 0) + +#define GEN6_TS_STRG_VAL 0x7e04 +#define GEN6_TS_RDATA 0x7e08 + +#define GEN6_TD_CTL 0x8000 +#define GEN6_TD_CTL_MUX_SHIFT 8 +#define GEN6_TD_CTL_EXTERNAL_HALT_R0_DEBUG_MATCH (1 << 7) +#define GEN6_TD_CTL_FORCE_EXTERNAL_HALT (1 << 6) +#define GEN6_TD_CTL_EXCEPTION_MASK_OVERRIDE (1 << 5) +#define GEN6_TD_CTL_FORCE_THREAD_BREAKPOINT_ENABLE (1 << 4) +#define GEN6_TD_CTL_BREAKPOINT_ENABLE (1 << 2) +#define GEN6_TD_CTL2 0x8004 +#define GEN6_TD_CTL2_ILLEGAL_OPCODE_EXCEPTION_OVERRIDE (1 << 28) +#define GEN6_TD_CTL2_MASKSTACK_EXCEPTION_OVERRIDE (1 << 26) +#define GEN6_TD_CTL2_SOFTWARE_EXCEPTION_OVERRIDE (1 << 25) +#define GEN6_TD_CTL2_ACTIVE_THREAD_LIMIT_SHIFT 16 +#define GEN6_TD_CTL2_ACTIVE_THREAD_LIMIT_ENABLE (1 << 8) +#define GEN6_TD_CTL2_THREAD_SPAWNER_EXECUTION_MASK_ENABLE (1 << 7) +#define GEN6_TD_CTL2_WIZ_EXECUTION_MASK_ENABLE (1 << 6) +#define GEN6_TD_CTL2_SF_EXECUTION_MASK_ENABLE (1 << 5) +#define GEN6_TD_CTL2_CLIPPER_EXECUTION_MASK_ENABLE (1 << 4) +#define GEN6_TD_CTL2_GS_EXECUTION_MASK_ENABLE (1 << 3) +#define GEN6_TD_CTL2_VS_EXECUTION_MASK_ENABLE (1 << 0) +#define GEN6_TD_VF_VS_EMSK 0x8008 +#define GEN6_TD_GS_EMSK 0x800c +#define GEN6_TD_CLIP_EMSK 0x8010 +#define GEN6_TD_SF_EMSK 0x8014 +#define GEN6_TD_WIZ_EMSK 0x8018 +#define GEN6_TD_0_6_EHTRG_VAL 0x801c +#define GEN6_TD_0_7_EHTRG_VAL 0x8020 +#define GEN6_TD_0_6_EHTRG_MSK 0x8024 +#define GEN6_TD_0_7_EHTRG_MSK 0x8028 +#define GEN6_TD_RDATA 0x802c +#define GEN6_TD_TS_EMSK 0x8030 + +#define GEN6_EU_CTL 0x8800 +#define GEN6_EU_CTL_SELECT_SHIFT 16 +#define GEN6_EU_CTL_DATA_MUX_SHIFT 8 +#define GEN6_EU_ATT_0 0x8810 +#define GEN6_EU_ATT_1 0x8814 +#define GEN6_EU_ATT_DATA_0 0x8820 +#define GEN6_EU_ATT_DATA_1 0x8824 +#define GEN6_EU_ATT_CLR_0 0x8830 +#define GEN6_EU_ATT_CLR_1 0x8834 +#define GEN6_EU_RDATA 0x8840 + +#define GEN6_3D(Pipeline,Opcode,Subopcode) ((3 << 29) | \ + ((Pipeline) << 27) | \ + ((Opcode) << 24) | \ + ((Subopcode) << 16)) + +#define GEN6_STATE_BASE_ADDRESS GEN6_3D(0, 1, 1) +#define GEN6_STATE_SIP GEN6_3D(0, 1, 2) + +#define GEN6_PIPELINE_SELECT GEN6_3D(1, 1, 4) + +#define GEN6_MEDIA_STATE_POINTERS GEN6_3D(2, 0, 0) +#define GEN6_MEDIA_OBJECT GEN6_3D(2, 1, 0) + +#define GEN6_3DSTATE_BINDING_TABLE_POINTERS GEN6_3D(3, 0, 1) +# define GEN6_3DSTATE_BINDING_TABLE_MODIFY_PS (1 << 12)/* for GEN6 */ +# define GEN6_3DSTATE_BINDING_TABLE_MODIFY_GS (1 << 9) /* for GEN6 */ +# define GEN6_3DSTATE_BINDING_TABLE_MODIFY_VS (1 << 8) /* for GEN6 */ + +#define GEN6_3DSTATE_VERTEX_BUFFERS GEN6_3D(3, 0, 8) +#define GEN6_3DSTATE_VERTEX_ELEMENTS GEN6_3D(3, 0, 9) +#define GEN6_3DSTATE_INDEX_BUFFER GEN6_3D(3, 0, 0xa) +#define GEN6_3DSTATE_VF_STATISTICS GEN6_3D(3, 0, 0xb) + +#define GEN6_3DSTATE_DRAWING_RECTANGLE GEN6_3D(3, 1, 0) +#define GEN6_3DSTATE_CONSTANT_COLOR GEN6_3D(3, 1, 1) +#define GEN6_3DSTATE_SAMPLER_PALETTE_LOAD GEN6_3D(3, 1, 2) +#define GEN6_3DSTATE_CHROMA_KEY GEN6_3D(3, 1, 4) +#define GEN6_3DSTATE_DEPTH_BUFFER GEN6_3D(3, 1, 5) +# define GEN6_3DSTATE_DEPTH_BUFFER_TYPE_SHIFT 29 +# define GEN6_3DSTATE_DEPTH_BUFFER_FORMAT_SHIFT 18 + +#define GEN6_3DSTATE_POLY_STIPPLE_OFFSET GEN6_3D(3, 1, 6) +#define GEN6_3DSTATE_POLY_STIPPLE_PATTERN GEN6_3D(3, 1, 7) +#define GEN6_3DSTATE_LINE_STIPPLE GEN6_3D(3, 1, 8) +#define GEN6_3DSTATE_GLOBAL_DEPTH_OFFSET_CLAMP GEN6_3D(3, 1, 9) +/* These two are BLC and CTG only, not BW or CL */ +#define GEN6_3DSTATE_AA_LINE_PARAMS GEN6_3D(3, 1, 0xa) +#define GEN6_3DSTATE_GS_SVB_INDEX GEN6_3D(3, 1, 0xb) + +#define GEN6_PIPE_CONTROL GEN6_3D(3, 2, 0) + +#define GEN6_3DPRIMITIVE GEN6_3D(3, 3, 0) + +#define GEN6_3DSTATE_CLEAR_PARAMS GEN6_3D(3, 1, 0x10) +/* DW1 */ +# define GEN6_3DSTATE_DEPTH_CLEAR_VALID (1 << 15) + +/* for GEN6+ */ +#define GEN6_3DSTATE_SAMPLER_STATE_POINTERS GEN6_3D(3, 0, 0x02) +# define GEN6_3DSTATE_SAMPLER_STATE_MODIFY_PS (1 << 12) +# define GEN6_3DSTATE_SAMPLER_STATE_MODIFY_GS (1 << 9) +# define GEN6_3DSTATE_SAMPLER_STATE_MODIFY_VS (1 << 8) + +#define GEN6_3DSTATE_URB GEN6_3D(3, 0, 0x05) +/* DW1 */ +# define GEN6_3DSTATE_URB_VS_SIZE_SHIFT 16 +# define GEN6_3DSTATE_URB_VS_ENTRIES_SHIFT 0 +/* DW2 */ +# define GEN6_3DSTATE_URB_GS_ENTRIES_SHIFT 8 +# define GEN6_3DSTATE_URB_GS_SIZE_SHIFT 0 + +#define GEN6_3DSTATE_VIEWPORT_STATE_POINTERS GEN6_3D(3, 0, 0x0d) +# define GEN6_3DSTATE_VIEWPORT_STATE_MODIFY_CC (1 << 12) +# define GEN6_3DSTATE_VIEWPORT_STATE_MODIFY_SF (1 << 11) +# define GEN6_3DSTATE_VIEWPORT_STATE_MODIFY_CLIP (1 << 10) + +#define GEN6_3DSTATE_CC_STATE_POINTERS GEN6_3D(3, 0, 0x0e) + +#define GEN6_3DSTATE_VS GEN6_3D(3, 0, 0x10) + +#define GEN6_3DSTATE_GS GEN6_3D(3, 0, 0x11) +/* DW4 */ +# define GEN6_3DSTATE_GS_DISPATCH_START_GRF_SHIFT 0 + +#define GEN6_3DSTATE_CLIP GEN6_3D(3, 0, 0x12) + +#define GEN6_3DSTATE_SF GEN6_3D(3, 0, 0x13) +/* DW1 */ +# define GEN6_3DSTATE_SF_NUM_OUTPUTS_SHIFT 22 +# define GEN6_3DSTATE_SF_URB_ENTRY_READ_LENGTH_SHIFT 11 +# define GEN6_3DSTATE_SF_URB_ENTRY_READ_OFFSET_SHIFT 4 +/* DW2 */ +/* DW3 */ +# define GEN6_3DSTATE_SF_CULL_BOTH (0 << 29) +# define GEN6_3DSTATE_SF_CULL_NONE (1 << 29) +# define GEN6_3DSTATE_SF_CULL_FRONT (2 << 29) +# define GEN6_3DSTATE_SF_CULL_BACK (3 << 29) +/* DW4 */ +# define GEN6_3DSTATE_SF_TRI_PROVOKE_SHIFT 29 +# define GEN6_3DSTATE_SF_LINE_PROVOKE_SHIFT 27 +# define GEN6_3DSTATE_SF_TRIFAN_PROVOKE_SHIFT 25 + + +#define GEN6_3DSTATE_WM GEN6_3D(3, 0, 0x14) +/* DW2 */ +# define GEN6_3DSTATE_WM_SAMPLER_COUNT_SHITF 27 +# define GEN6_3DSTATE_WM_BINDING_TABLE_ENTRY_COUNT_SHIFT 18 +/* DW4 */ +# define GEN6_3DSTATE_WM_DISPATCH_START_GRF_0_SHIFT 16 +/* DW5 */ +# define GEN6_3DSTATE_WM_MAX_THREADS_SHIFT 25 +# define GEN6_3DSTATE_WM_DISPATCH_ENABLE (1 << 19) +# define GEN6_3DSTATE_WM_16_DISPATCH_ENABLE (1 << 1) +# define GEN6_3DSTATE_WM_8_DISPATCH_ENABLE (1 << 0) +/* DW6 */ +# define GEN6_3DSTATE_WM_NUM_SF_OUTPUTS_SHIFT 20 +# define GEN6_3DSTATE_WM_NONPERSPECTIVE_SAMPLE_BARYCENTRIC (1 << 15) +# define GEN6_3DSTATE_WM_NONPERSPECTIVE_CENTROID_BARYCENTRIC (1 << 14) +# define GEN6_3DSTATE_WM_NONPERSPECTIVE_PIXEL_BARYCENTRIC (1 << 13) +# define GEN6_3DSTATE_WM_PERSPECTIVE_SAMPLE_BARYCENTRIC (1 << 12) +# define GEN6_3DSTATE_WM_PERSPECTIVE_CENTROID_BARYCENTRIC (1 << 11) +# define GEN6_3DSTATE_WM_PERSPECTIVE_PIXEL_BARYCENTRIC (1 << 10) + + +#define GEN6_3DSTATE_CONSTANT_VS GEN6_3D(3, 0, 0x15) +#define GEN6_3DSTATE_CONSTANT_GS GEN6_3D(3, 0, 0x16) +#define GEN6_3DSTATE_CONSTANT_PS GEN6_3D(3, 0, 0x17) + +#define GEN6_3DSTATE_SAMPLE_MASK GEN6_3D(3, 0, 0x18) + +#define GEN6_3DSTATE_MULTISAMPLE GEN6_3D(3, 1, 0x0d) +/* DW1 */ +# define GEN6_3DSTATE_MULTISAMPLE_PIXEL_LOCATION_CENTER (0 << 4) +# define GEN6_3DSTATE_MULTISAMPLE_PIXEL_LOCATION_UPPER_LEFT (1 << 4) +# define GEN6_3DSTATE_MULTISAMPLE_NUMSAMPLES_1 (0 << 1) +# define GEN6_3DSTATE_MULTISAMPLE_NUMSAMPLES_4 (2 << 1) +# define GEN6_3DSTATE_MULTISAMPLE_NUMSAMPLES_8 (3 << 1) + +#define PIPELINE_SELECT_3D 0 +#define PIPELINE_SELECT_MEDIA 1 + +#define UF0_CS_REALLOC (1 << 13) +#define UF0_VFE_REALLOC (1 << 12) +#define UF0_SF_REALLOC (1 << 11) +#define UF0_CLIP_REALLOC (1 << 10) +#define UF0_GS_REALLOC (1 << 9) +#define UF0_VS_REALLOC (1 << 8) +#define UF1_CLIP_FENCE_SHIFT 20 +#define UF1_GS_FENCE_SHIFT 10 +#define UF1_VS_FENCE_SHIFT 0 +#define UF2_CS_FENCE_SHIFT 20 +#define UF2_VFE_FENCE_SHIFT 10 +#define UF2_SF_FENCE_SHIFT 0 + +/* for GEN6_STATE_BASE_ADDRESS */ +#define BASE_ADDRESS_MODIFY (1 << 0) + +/* for GEN6_3DSTATE_PIPELINED_POINTERS */ +#define GEN6_GS_DISABLE 0 +#define GEN6_GS_ENABLE 1 +#define GEN6_CLIP_DISABLE 0 +#define GEN6_CLIP_ENABLE 1 + +/* for GEN6_PIPE_CONTROL */ +#define GEN6_PIPE_CONTROL_NOWRITE (0 << 14) +#define GEN6_PIPE_CONTROL_WRITE_QWORD (1 << 14) +#define GEN6_PIPE_CONTROL_WRITE_DEPTH (2 << 14) +#define GEN6_PIPE_CONTROL_WRITE_TIME (3 << 14) +#define GEN6_PIPE_CONTROL_DEPTH_STALL (1 << 13) +#define GEN6_PIPE_CONTROL_WC_FLUSH (1 << 12) +#define GEN6_PIPE_CONTROL_IS_FLUSH (1 << 11) +#define GEN6_PIPE_CONTROL_TC_FLUSH (1 << 10) +#define GEN6_PIPE_CONTROL_NOTIFY_ENABLE (1 << 8) +#define GEN6_PIPE_CONTROL_GLOBAL_GTT (1 << 2) +#define GEN6_PIPE_CONTROL_LOCAL_PGTT (0 << 2) +#define GEN6_PIPE_CONTROL_DEPTH_CACHE_FLUSH (1 << 0) + +/* 3DPRIMITIVE bits */ +#define GEN6_3DPRIMITIVE_VERTEX_SEQUENTIAL (0 << 15) +#define GEN6_3DPRIMITIVE_VERTEX_RANDOM (1 << 15) +/* Primitive types are in gen6_defines.h */ +#define GEN6_3DPRIMITIVE_TOPOLOGY_SHIFT 10 + +#define GEN6_SVG_CTL 0x7400 + +#define GEN6_SVG_CTL_GS_BA (0 << 8) +#define GEN6_SVG_CTL_SS_BA (1 << 8) +#define GEN6_SVG_CTL_IO_BA (2 << 8) +#define GEN6_SVG_CTL_GS_AUB (3 << 8) +#define GEN6_SVG_CTL_IO_AUB (4 << 8) +#define GEN6_SVG_CTL_SIP (5 << 8) + +#define GEN6_SVG_RDATA 0x7404 +#define GEN6_SVG_WORK_CTL 0x7408 + +#define GEN6_VF_CTL 0x7500 + +#define GEN6_VF_CTL_SNAPSHOT_COMPLETE (1 << 31) +#define GEN6_VF_CTL_SNAPSHOT_MUX_SELECT_THREADID (0 << 8) +#define GEN6_VF_CTL_SNAPSHOT_MUX_SELECT_VF_DEBUG (1 << 8) +#define GEN6_VF_CTL_SNAPSHOT_TYPE_VERTEX_SEQUENCE (0 << 4) +#define GEN6_VF_CTL_SNAPSHOT_TYPE_VERTEX_INDEX (1 << 4) +#define GEN6_VF_CTL_SKIP_INITIAL_PRIMITIVES (1 << 3) +#define GEN6_VF_CTL_MAX_PRIMITIVES_LIMIT_ENABLE (1 << 2) +#define GEN6_VF_CTL_VERTEX_RANGE_LIMIT_ENABLE (1 << 1) +#define GEN6_VF_CTL_SNAPSHOT_ENABLE (1 << 0) + +#define GEN6_VF_STRG_VAL 0x7504 +#define GEN6_VF_STR_VL_OVR 0x7508 +#define GEN6_VF_VC_OVR 0x750c +#define GEN6_VF_STR_PSKIP 0x7510 +#define GEN6_VF_MAX_PRIM 0x7514 +#define GEN6_VF_RDATA 0x7518 + +#define GEN6_VS_CTL 0x7600 +#define GEN6_VS_CTL_SNAPSHOT_COMPLETE (1 << 31) +#define GEN6_VS_CTL_SNAPSHOT_MUX_VERTEX_0 (0 << 8) +#define GEN6_VS_CTL_SNAPSHOT_MUX_VERTEX_1 (1 << 8) +#define GEN6_VS_CTL_SNAPSHOT_MUX_VALID_COUNT (2 << 8) +#define GEN6_VS_CTL_SNAPSHOT_MUX_VS_KERNEL_POINTER (3 << 8) +#define GEN6_VS_CTL_SNAPSHOT_ALL_THREADS (1 << 2) +#define GEN6_VS_CTL_THREAD_SNAPSHOT_ENABLE (1 << 1) +#define GEN6_VS_CTL_SNAPSHOT_ENABLE (1 << 0) + +#define GEN6_VS_STRG_VAL 0x7604 +#define GEN6_VS_RDATA 0x7608 + +#define GEN6_SF_CTL 0x7b00 +#define GEN6_SF_CTL_SNAPSHOT_COMPLETE (1 << 31) +#define GEN6_SF_CTL_SNAPSHOT_MUX_VERTEX_0_FF_ID (0 << 8) +#define GEN6_SF_CTL_SNAPSHOT_MUX_VERTEX_0_REL_COUNT (1 << 8) +#define GEN6_SF_CTL_SNAPSHOT_MUX_VERTEX_1_FF_ID (2 << 8) +#define GEN6_SF_CTL_SNAPSHOT_MUX_VERTEX_1_REL_COUNT (3 << 8) +#define GEN6_SF_CTL_SNAPSHOT_MUX_VERTEX_2_FF_ID (4 << 8) +#define GEN6_SF_CTL_SNAPSHOT_MUX_VERTEX_2_REL_COUNT (5 << 8) +#define GEN6_SF_CTL_SNAPSHOT_MUX_VERTEX_COUNT (6 << 8) +#define GEN6_SF_CTL_SNAPSHOT_MUX_SF_KERNEL_POINTER (7 << 8) +#define GEN6_SF_CTL_MIN_MAX_PRIMITIVE_RANGE_ENABLE (1 << 4) +#define GEN6_SF_CTL_DEBUG_CLIP_RECTANGLE_ENABLE (1 << 3) +#define GEN6_SF_CTL_SNAPSHOT_ALL_THREADS (1 << 2) +#define GEN6_SF_CTL_THREAD_SNAPSHOT_ENABLE (1 << 1) +#define GEN6_SF_CTL_SNAPSHOT_ENABLE (1 << 0) + +#define GEN6_SF_STRG_VAL 0x7b04 +#define GEN6_SF_RDATA 0x7b18 + +#define GEN6_WIZ_CTL 0x7c00 +#define GEN6_WIZ_CTL_SNAPSHOT_COMPLETE (1 << 31) +#define GEN6_WIZ_CTL_SUBSPAN_INSTANCE_SHIFT 16 +#define GEN6_WIZ_CTL_SNAPSHOT_MUX_WIZ_KERNEL_POINTER (0 << 8) +#define GEN6_WIZ_CTL_SNAPSHOT_MUX_SUBSPAN_INSTANCE (1 << 8) +#define GEN6_WIZ_CTL_SNAPSHOT_MUX_PRIMITIVE_SEQUENCE (2 << 8) +#define GEN6_WIZ_CTL_SINGLE_SUBSPAN_DISPATCH (1 << 6) +#define GEN6_WIZ_CTL_IGNORE_COLOR_SCOREBOARD_STALLS (1 << 5) +#define GEN6_WIZ_CTL_ENABLE_SUBSPAN_INSTANCE_COMPARE (1 << 4) +#define GEN6_WIZ_CTL_USE_UPSTREAM_SNAPSHOT_FLAG (1 << 3) +#define GEN6_WIZ_CTL_SNAPSHOT_ALL_THREADS (1 << 2) +#define GEN6_WIZ_CTL_THREAD_SNAPSHOT_ENABLE (1 << 1) +#define GEN6_WIZ_CTL_SNAPSHOT_ENABLE (1 << 0) + +#define GEN6_WIZ_STRG_VAL 0x7c04 +#define GEN6_WIZ_RDATA 0x7c18 + +#define GEN6_TS_CTL 0x7e00 +#define GEN6_TS_CTL_SNAPSHOT_COMPLETE (1 << 31) +#define GEN6_TS_CTL_SNAPSHOT_MESSAGE_ERROR (0 << 8) +#define GEN6_TS_CTL_SNAPSHOT_INTERFACE_DESCRIPTOR (3 << 8) +#define GEN6_TS_CTL_SNAPSHOT_ALL_CHILD_THREADS (1 << 2) +#define GEN6_TS_CTL_SNAPSHOT_ALL_ROOT_THREADS (1 << 1) +#define GEN6_TS_CTL_SNAPSHOT_ENABLE (1 << 0) + +#define GEN6_TS_STRG_VAL 0x7e04 +#define GEN6_TS_RDATA 0x7e08 + +#define GEN6_TD_CTL 0x8000 +#define GEN6_TD_CTL_MUX_SHIFT 8 +#define GEN6_TD_CTL_EXTERNAL_HALT_R0_DEBUG_MATCH (1 << 7) +#define GEN6_TD_CTL_FORCE_EXTERNAL_HALT (1 << 6) +#define GEN6_TD_CTL_EXCEPTION_MASK_OVERRIDE (1 << 5) +#define GEN6_TD_CTL_FORCE_THREAD_BREAKPOINT_ENABLE (1 << 4) +#define GEN6_TD_CTL_BREAKPOINT_ENABLE (1 << 2) +#define GEN6_TD_CTL2 0x8004 +#define GEN6_TD_CTL2_ILLEGAL_OPCODE_EXCEPTION_OVERRIDE (1 << 28) +#define GEN6_TD_CTL2_MASKSTACK_EXCEPTION_OVERRIDE (1 << 26) +#define GEN6_TD_CTL2_SOFTWARE_EXCEPTION_OVERRIDE (1 << 25) +#define GEN6_TD_CTL2_ACTIVE_THREAD_LIMIT_SHIFT 16 +#define GEN6_TD_CTL2_ACTIVE_THREAD_LIMIT_ENABLE (1 << 8) +#define GEN6_TD_CTL2_THREAD_SPAWNER_EXECUTION_MASK_ENABLE (1 << 7) +#define GEN6_TD_CTL2_WIZ_EXECUTION_MASK_ENABLE (1 << 6) +#define GEN6_TD_CTL2_SF_EXECUTION_MASK_ENABLE (1 << 5) +#define GEN6_TD_CTL2_CLIPPER_EXECUTION_MASK_ENABLE (1 << 4) +#define GEN6_TD_CTL2_GS_EXECUTION_MASK_ENABLE (1 << 3) +#define GEN6_TD_CTL2_VS_EXECUTION_MASK_ENABLE (1 << 0) +#define GEN6_TD_VF_VS_EMSK 0x8008 +#define GEN6_TD_GS_EMSK 0x800c +#define GEN6_TD_CLIP_EMSK 0x8010 +#define GEN6_TD_SF_EMSK 0x8014 +#define GEN6_TD_WIZ_EMSK 0x8018 +#define GEN6_TD_0_6_EHTRG_VAL 0x801c +#define GEN6_TD_0_7_EHTRG_VAL 0x8020 +#define GEN6_TD_0_6_EHTRG_MSK 0x8024 +#define GEN6_TD_0_7_EHTRG_MSK 0x8028 +#define GEN6_TD_RDATA 0x802c +#define GEN6_TD_TS_EMSK 0x8030 + +#define GEN6_EU_CTL 0x8800 +#define GEN6_EU_CTL_SELECT_SHIFT 16 +#define GEN6_EU_CTL_DATA_MUX_SHIFT 8 +#define GEN6_EU_ATT_0 0x8810 +#define GEN6_EU_ATT_1 0x8814 +#define GEN6_EU_ATT_DATA_0 0x8820 +#define GEN6_EU_ATT_DATA_1 0x8824 +#define GEN6_EU_ATT_CLR_0 0x8830 +#define GEN6_EU_ATT_CLR_1 0x8834 +#define GEN6_EU_RDATA 0x8840 + +/* 3D state: + */ +#define _3DOP_3DSTATE_PIPELINED 0x0 +#define _3DOP_3DSTATE_NONPIPELINED 0x1 +#define _3DOP_3DCONTROL 0x2 +#define _3DOP_3DPRIMITIVE 0x3 + +#define _3DSTATE_PIPELINED_POINTERS 0x00 +#define _3DSTATE_BINDING_TABLE_POINTERS 0x01 +#define _3DSTATE_VERTEX_BUFFERS 0x08 +#define _3DSTATE_VERTEX_ELEMENTS 0x09 +#define _3DSTATE_INDEX_BUFFER 0x0A +#define _3DSTATE_VF_STATISTICS 0x0B +#define _3DSTATE_DRAWING_RECTANGLE 0x00 +#define _3DSTATE_CONSTANT_COLOR 0x01 +#define _3DSTATE_SAMPLER_PALETTE_LOAD 0x02 +#define _3DSTATE_CHROMA_KEY 0x04 +#define _3DSTATE_DEPTH_BUFFER 0x05 +#define _3DSTATE_POLY_STIPPLE_OFFSET 0x06 +#define _3DSTATE_POLY_STIPPLE_PATTERN 0x07 +#define _3DSTATE_LINE_STIPPLE 0x08 +#define _3DSTATE_GLOBAL_DEPTH_OFFSET_CLAMP 0x09 +#define _3DCONTROL 0x00 +#define _3DPRIMITIVE 0x00 + +#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 GEN6_ANISORATIO_2 0 +#define GEN6_ANISORATIO_4 1 +#define GEN6_ANISORATIO_6 2 +#define GEN6_ANISORATIO_8 3 +#define GEN6_ANISORATIO_10 4 +#define GEN6_ANISORATIO_12 5 +#define GEN6_ANISORATIO_14 6 +#define GEN6_ANISORATIO_16 7 + +#define GEN6_BLENDFACTOR_ONE 0x1 +#define GEN6_BLENDFACTOR_SRC_COLOR 0x2 +#define GEN6_BLENDFACTOR_SRC_ALPHA 0x3 +#define GEN6_BLENDFACTOR_DST_ALPHA 0x4 +#define GEN6_BLENDFACTOR_DST_COLOR 0x5 +#define GEN6_BLENDFACTOR_SRC_ALPHA_SATURATE 0x6 +#define GEN6_BLENDFACTOR_CONST_COLOR 0x7 +#define GEN6_BLENDFACTOR_CONST_ALPHA 0x8 +#define GEN6_BLENDFACTOR_SRC1_COLOR 0x9 +#define GEN6_BLENDFACTOR_SRC1_ALPHA 0x0A +#define GEN6_BLENDFACTOR_ZERO 0x11 +#define GEN6_BLENDFACTOR_INV_SRC_COLOR 0x12 +#define GEN6_BLENDFACTOR_INV_SRC_ALPHA 0x13 +#define GEN6_BLENDFACTOR_INV_DST_ALPHA 0x14 +#define GEN6_BLENDFACTOR_INV_DST_COLOR 0x15 +#define GEN6_BLENDFACTOR_INV_CONST_COLOR 0x17 +#define GEN6_BLENDFACTOR_INV_CONST_ALPHA 0x18 +#define GEN6_BLENDFACTOR_INV_SRC1_COLOR 0x19 +#define GEN6_BLENDFACTOR_INV_SRC1_ALPHA 0x1A + +#define GEN6_BLENDFUNCTION_ADD 0 +#define GEN6_BLENDFUNCTION_SUBTRACT 1 +#define GEN6_BLENDFUNCTION_REVERSE_SUBTRACT 2 +#define GEN6_BLENDFUNCTION_MIN 3 +#define GEN6_BLENDFUNCTION_MAX 4 + +#define GEN6_ALPHATEST_FORMAT_UNORM8 0 +#define GEN6_ALPHATEST_FORMAT_FLOAT32 1 + +#define GEN6_CHROMAKEY_KILL_ON_ANY_MATCH 0 +#define GEN6_CHROMAKEY_REPLACE_BLACK 1 + +#define GEN6_CLIP_API_OGL 0 +#define GEN6_CLIP_API_DX 1 + +#define GEN6_CLIPMODE_NORMAL 0 +#define GEN6_CLIPMODE_CLIP_ALL 1 +#define GEN6_CLIPMODE_CLIP_NON_REJECTED 2 +#define GEN6_CLIPMODE_REJECT_ALL 3 +#define GEN6_CLIPMODE_ACCEPT_ALL 4 + +#define GEN6_CLIP_NDCSPACE 0 +#define GEN6_CLIP_SCREENSPACE 1 + +#define GEN6_COMPAREFUNCTION_ALWAYS 0 +#define GEN6_COMPAREFUNCTION_NEVER 1 +#define GEN6_COMPAREFUNCTION_LESS 2 +#define GEN6_COMPAREFUNCTION_EQUAL 3 +#define GEN6_COMPAREFUNCTION_LEQUAL 4 +#define GEN6_COMPAREFUNCTION_GREATER 5 +#define GEN6_COMPAREFUNCTION_NOTEQUAL 6 +#define GEN6_COMPAREFUNCTION_GEQUAL 7 + +#define GEN6_COVERAGE_PIXELS_HALF 0 +#define GEN6_COVERAGE_PIXELS_1 1 +#define GEN6_COVERAGE_PIXELS_2 2 +#define GEN6_COVERAGE_PIXELS_4 3 + +#define GEN6_CULLMODE_BOTH 0 +#define GEN6_CULLMODE_NONE 1 +#define GEN6_CULLMODE_FRONT 2 +#define GEN6_CULLMODE_BACK 3 + +#define GEN6_DEFAULTCOLOR_R8G8B8A8_UNORM 0 +#define GEN6_DEFAULTCOLOR_R32G32B32A32_FLOAT 1 + +#define GEN6_DEPTHFORMAT_D32_FLOAT_S8X24_UINT 0 +#define GEN6_DEPTHFORMAT_D32_FLOAT 1 +#define GEN6_DEPTHFORMAT_D24_UNORM_S8_UINT 2 +#define GEN6_DEPTHFORMAT_D16_UNORM 5 + +#define GEN6_FLOATING_POINT_IEEE_754 0 +#define GEN6_FLOATING_POINT_NON_IEEE_754 1 + +#define GEN6_FRONTWINDING_CW 0 +#define GEN6_FRONTWINDING_CCW 1 + +#define GEN6_INDEX_BYTE 0 +#define GEN6_INDEX_WORD 1 +#define GEN6_INDEX_DWORD 2 + +#define GEN6_LOGICOPFUNCTION_CLEAR 0 +#define GEN6_LOGICOPFUNCTION_NOR 1 +#define GEN6_LOGICOPFUNCTION_AND_INVERTED 2 +#define GEN6_LOGICOPFUNCTION_COPY_INVERTED 3 +#define GEN6_LOGICOPFUNCTION_AND_REVERSE 4 +#define GEN6_LOGICOPFUNCTION_INVERT 5 +#define GEN6_LOGICOPFUNCTION_XOR 6 +#define GEN6_LOGICOPFUNCTION_NAND 7 +#define GEN6_LOGICOPFUNCTION_AND 8 +#define GEN6_LOGICOPFUNCTION_EQUIV 9 +#define GEN6_LOGICOPFUNCTION_NOOP 10 +#define GEN6_LOGICOPFUNCTION_OR_INVERTED 11 +#define GEN6_LOGICOPFUNCTION_COPY 12 +#define GEN6_LOGICOPFUNCTION_OR_REVERSE 13 +#define GEN6_LOGICOPFUNCTION_OR 14 +#define GEN6_LOGICOPFUNCTION_SET 15 + +#define GEN6_MAPFILTER_NEAREST 0x0 +#define GEN6_MAPFILTER_LINEAR 0x1 +#define GEN6_MAPFILTER_ANISOTROPIC 0x2 + +#define GEN6_MIPFILTER_NONE 0 +#define GEN6_MIPFILTER_NEAREST 1 +#define GEN6_MIPFILTER_LINEAR 3 + +#define GEN6_POLYGON_FRONT_FACING 0 +#define GEN6_POLYGON_BACK_FACING 1 + +#define GEN6_PREFILTER_ALWAYS 0x0 +#define GEN6_PREFILTER_NEVER 0x1 +#define GEN6_PREFILTER_LESS 0x2 +#define GEN6_PREFILTER_EQUAL 0x3 +#define GEN6_PREFILTER_LEQUAL 0x4 +#define GEN6_PREFILTER_GREATER 0x5 +#define GEN6_PREFILTER_NOTEQUAL 0x6 +#define GEN6_PREFILTER_GEQUAL 0x7 + +#define GEN6_PROVOKING_VERTEX_0 0 +#define GEN6_PROVOKING_VERTEX_1 1 +#define GEN6_PROVOKING_VERTEX_2 2 + +#define GEN6_RASTRULE_UPPER_LEFT 0 +#define GEN6_RASTRULE_UPPER_RIGHT 1 + +#define GEN6_RENDERTARGET_CLAMPRANGE_UNORM 0 +#define GEN6_RENDERTARGET_CLAMPRANGE_SNORM 1 +#define GEN6_RENDERTARGET_CLAMPRANGE_FORMAT 2 + +#define GEN6_STENCILOP_KEEP 0 +#define GEN6_STENCILOP_ZERO 1 +#define GEN6_STENCILOP_REPLACE 2 +#define GEN6_STENCILOP_INCRSAT 3 +#define GEN6_STENCILOP_DECRSAT 4 +#define GEN6_STENCILOP_INCR 5 +#define GEN6_STENCILOP_DECR 6 +#define GEN6_STENCILOP_INVERT 7 + +#define GEN6_SURFACE_MIPMAPLAYOUT_BELOW 0 +#define GEN6_SURFACE_MIPMAPLAYOUT_RIGHT 1 + +#define GEN6_SURFACEFORMAT_R32G32B32A32_FLOAT 0x000 +#define GEN6_SURFACEFORMAT_R32G32B32A32_SINT 0x001 +#define GEN6_SURFACEFORMAT_R32G32B32A32_UINT 0x002 +#define GEN6_SURFACEFORMAT_R32G32B32A32_UNORM 0x003 +#define GEN6_SURFACEFORMAT_R32G32B32A32_SNORM 0x004 +#define GEN6_SURFACEFORMAT_R64G64_FLOAT 0x005 +#define GEN6_SURFACEFORMAT_R32G32B32X32_FLOAT 0x006 +#define GEN6_SURFACEFORMAT_R32G32B32A32_SSCALED 0x007 +#define GEN6_SURFACEFORMAT_R32G32B32A32_USCALED 0x008 +#define GEN6_SURFACEFORMAT_R32G32B32_FLOAT 0x040 +#define GEN6_SURFACEFORMAT_R32G32B32_SINT 0x041 +#define GEN6_SURFACEFORMAT_R32G32B32_UINT 0x042 +#define GEN6_SURFACEFORMAT_R32G32B32_UNORM 0x043 +#define GEN6_SURFACEFORMAT_R32G32B32_SNORM 0x044 +#define GEN6_SURFACEFORMAT_R32G32B32_SSCALED 0x045 +#define GEN6_SURFACEFORMAT_R32G32B32_USCALED 0x046 +#define GEN6_SURFACEFORMAT_R16G16B16A16_UNORM 0x080 +#define GEN6_SURFACEFORMAT_R16G16B16A16_SNORM 0x081 +#define GEN6_SURFACEFORMAT_R16G16B16A16_SINT 0x082 +#define GEN6_SURFACEFORMAT_R16G16B16A16_UINT 0x083 +#define GEN6_SURFACEFORMAT_R16G16B16A16_FLOAT 0x084 +#define GEN6_SURFACEFORMAT_R32G32_FLOAT 0x085 +#define GEN6_SURFACEFORMAT_R32G32_SINT 0x086 +#define GEN6_SURFACEFORMAT_R32G32_UINT 0x087 +#define GEN6_SURFACEFORMAT_R32_FLOAT_X8X24_TYPELESS 0x088 +#define GEN6_SURFACEFORMAT_X32_TYPELESS_G8X24_UINT 0x089 +#define GEN6_SURFACEFORMAT_L32A32_FLOAT 0x08A +#define GEN6_SURFACEFORMAT_R32G32_UNORM 0x08B +#define GEN6_SURFACEFORMAT_R32G32_SNORM 0x08C +#define GEN6_SURFACEFORMAT_R64_FLOAT 0x08D +#define GEN6_SURFACEFORMAT_R16G16B16X16_UNORM 0x08E +#define GEN6_SURFACEFORMAT_R16G16B16X16_FLOAT 0x08F +#define GEN6_SURFACEFORMAT_A32X32_FLOAT 0x090 +#define GEN6_SURFACEFORMAT_L32X32_FLOAT 0x091 +#define GEN6_SURFACEFORMAT_I32X32_FLOAT 0x092 +#define GEN6_SURFACEFORMAT_R16G16B16A16_SSCALED 0x093 +#define GEN6_SURFACEFORMAT_R16G16B16A16_USCALED 0x094 +#define GEN6_SURFACEFORMAT_R32G32_SSCALED 0x095 +#define GEN6_SURFACEFORMAT_R32G32_USCALED 0x096 +#define GEN6_SURFACEFORMAT_B8G8R8A8_UNORM 0x0C0 +#define GEN6_SURFACEFORMAT_B8G8R8A8_UNORM_SRGB 0x0C1 +#define GEN6_SURFACEFORMAT_R10G10B10A2_UNORM 0x0C2 +#define GEN6_SURFACEFORMAT_R10G10B10A2_UNORM_SRGB 0x0C3 +#define GEN6_SURFACEFORMAT_R10G10B10A2_UINT 0x0C4 +#define GEN6_SURFACEFORMAT_R10G10B10_SNORM_A2_UNORM 0x0C5 +#define GEN6_SURFACEFORMAT_R8G8B8A8_UNORM 0x0C7 +#define GEN6_SURFACEFORMAT_R8G8B8A8_UNORM_SRGB 0x0C8 +#define GEN6_SURFACEFORMAT_R8G8B8A8_SNORM 0x0C9 +#define GEN6_SURFACEFORMAT_R8G8B8A8_SINT 0x0CA +#define GEN6_SURFACEFORMAT_R8G8B8A8_UINT 0x0CB +#define GEN6_SURFACEFORMAT_R16G16_UNORM 0x0CC +#define GEN6_SURFACEFORMAT_R16G16_SNORM 0x0CD +#define GEN6_SURFACEFORMAT_R16G16_SINT 0x0CE +#define GEN6_SURFACEFORMAT_R16G16_UINT 0x0CF +#define GEN6_SURFACEFORMAT_R16G16_FLOAT 0x0D0 +#define GEN6_SURFACEFORMAT_B10G10R10A2_UNORM 0x0D1 +#define GEN6_SURFACEFORMAT_B10G10R10A2_UNORM_SRGB 0x0D2 +#define GEN6_SURFACEFORMAT_R11G11B10_FLOAT 0x0D3 +#define GEN6_SURFACEFORMAT_R32_SINT 0x0D6 +#define GEN6_SURFACEFORMAT_R32_UINT 0x0D7 +#define GEN6_SURFACEFORMAT_R32_FLOAT 0x0D8 +#define GEN6_SURFACEFORMAT_R24_UNORM_X8_TYPELESS 0x0D9 +#define GEN6_SURFACEFORMAT_X24_TYPELESS_G8_UINT 0x0DA +#define GEN6_SURFACEFORMAT_L16A16_UNORM 0x0DF +#define GEN6_SURFACEFORMAT_I24X8_UNORM 0x0E0 +#define GEN6_SURFACEFORMAT_L24X8_UNORM 0x0E1 +#define GEN6_SURFACEFORMAT_A24X8_UNORM 0x0E2 +#define GEN6_SURFACEFORMAT_I32_FLOAT 0x0E3 +#define GEN6_SURFACEFORMAT_L32_FLOAT 0x0E4 +#define GEN6_SURFACEFORMAT_A32_FLOAT 0x0E5 +#define GEN6_SURFACEFORMAT_B8G8R8X8_UNORM 0x0E9 +#define GEN6_SURFACEFORMAT_B8G8R8X8_UNORM_SRGB 0x0EA +#define GEN6_SURFACEFORMAT_R8G8B8X8_UNORM 0x0EB +#define GEN6_SURFACEFORMAT_R8G8B8X8_UNORM_SRGB 0x0EC +#define GEN6_SURFACEFORMAT_R9G9B9E5_SHAREDEXP 0x0ED +#define GEN6_SURFACEFORMAT_B10G10R10X2_UNORM 0x0EE +#define GEN6_SURFACEFORMAT_L16A16_FLOAT 0x0F0 +#define GEN6_SURFACEFORMAT_R32_UNORM 0x0F1 +#define GEN6_SURFACEFORMAT_R32_SNORM 0x0F2 +#define GEN6_SURFACEFORMAT_R10G10B10X2_USCALED 0x0F3 +#define GEN6_SURFACEFORMAT_R8G8B8A8_SSCALED 0x0F4 +#define GEN6_SURFACEFORMAT_R8G8B8A8_USCALED 0x0F5 +#define GEN6_SURFACEFORMAT_R16G16_SSCALED 0x0F6 +#define GEN6_SURFACEFORMAT_R16G16_USCALED 0x0F7 +#define GEN6_SURFACEFORMAT_R32_SSCALED 0x0F8 +#define GEN6_SURFACEFORMAT_R32_USCALED 0x0F9 +#define GEN6_SURFACEFORMAT_B5G6R5_UNORM 0x100 +#define GEN6_SURFACEFORMAT_B5G6R5_UNORM_SRGB 0x101 +#define GEN6_SURFACEFORMAT_B5G5R5A1_UNORM 0x102 +#define GEN6_SURFACEFORMAT_B5G5R5A1_UNORM_SRGB 0x103 +#define GEN6_SURFACEFORMAT_B4G4R4A4_UNORM 0x104 +#define GEN6_SURFACEFORMAT_B4G4R4A4_UNORM_SRGB 0x105 +#define GEN6_SURFACEFORMAT_R8G8_UNORM 0x106 +#define GEN6_SURFACEFORMAT_R8G8_SNORM 0x107 +#define GEN6_SURFACEFORMAT_R8G8_SINT 0x108 +#define GEN6_SURFACEFORMAT_R8G8_UINT 0x109 +#define GEN6_SURFACEFORMAT_R16_UNORM 0x10A +#define GEN6_SURFACEFORMAT_R16_SNORM 0x10B +#define GEN6_SURFACEFORMAT_R16_SINT 0x10C +#define GEN6_SURFACEFORMAT_R16_UINT 0x10D +#define GEN6_SURFACEFORMAT_R16_FLOAT 0x10E +#define GEN6_SURFACEFORMAT_I16_UNORM 0x111 +#define GEN6_SURFACEFORMAT_L16_UNORM 0x112 +#define GEN6_SURFACEFORMAT_A16_UNORM 0x113 +#define GEN6_SURFACEFORMAT_L8A8_UNORM 0x114 +#define GEN6_SURFACEFORMAT_I16_FLOAT 0x115 +#define GEN6_SURFACEFORMAT_L16_FLOAT 0x116 +#define GEN6_SURFACEFORMAT_A16_FLOAT 0x117 +#define GEN6_SURFACEFORMAT_R5G5_SNORM_B6_UNORM 0x119 +#define GEN6_SURFACEFORMAT_B5G5R5X1_UNORM 0x11A +#define GEN6_SURFACEFORMAT_B5G5R5X1_UNORM_SRGB 0x11B +#define GEN6_SURFACEFORMAT_R8G8_SSCALED 0x11C +#define GEN6_SURFACEFORMAT_R8G8_USCALED 0x11D +#define GEN6_SURFACEFORMAT_R16_SSCALED 0x11E +#define GEN6_SURFACEFORMAT_R16_USCALED 0x11F +#define GEN6_SURFACEFORMAT_R8_UNORM 0x140 +#define GEN6_SURFACEFORMAT_R8_SNORM 0x141 +#define GEN6_SURFACEFORMAT_R8_SINT 0x142 +#define GEN6_SURFACEFORMAT_R8_UINT 0x143 +#define GEN6_SURFACEFORMAT_A8_UNORM 0x144 +#define GEN6_SURFACEFORMAT_I8_UNORM 0x145 +#define GEN6_SURFACEFORMAT_L8_UNORM 0x146 +#define GEN6_SURFACEFORMAT_P4A4_UNORM 0x147 +#define GEN6_SURFACEFORMAT_A4P4_UNORM 0x148 +#define GEN6_SURFACEFORMAT_R8_SSCALED 0x149 +#define GEN6_SURFACEFORMAT_R8_USCALED 0x14A +#define GEN6_SURFACEFORMAT_R1_UINT 0x181 +#define GEN6_SURFACEFORMAT_YCRCB_NORMAL 0x182 +#define GEN6_SURFACEFORMAT_YCRCB_SWAPUVY 0x183 +#define GEN6_SURFACEFORMAT_BC1_UNORM 0x186 +#define GEN6_SURFACEFORMAT_BC2_UNORM 0x187 +#define GEN6_SURFACEFORMAT_BC3_UNORM 0x188 +#define GEN6_SURFACEFORMAT_BC4_UNORM 0x189 +#define GEN6_SURFACEFORMAT_BC5_UNORM 0x18A +#define GEN6_SURFACEFORMAT_BC1_UNORM_SRGB 0x18B +#define GEN6_SURFACEFORMAT_BC2_UNORM_SRGB 0x18C +#define GEN6_SURFACEFORMAT_BC3_UNORM_SRGB 0x18D +#define GEN6_SURFACEFORMAT_MONO8 0x18E +#define GEN6_SURFACEFORMAT_YCRCB_SWAPUV 0x18F +#define GEN6_SURFACEFORMAT_YCRCB_SWAPY 0x190 +#define GEN6_SURFACEFORMAT_DXT1_RGB 0x191 +#define GEN6_SURFACEFORMAT_FXT1 0x192 +#define GEN6_SURFACEFORMAT_R8G8B8_UNORM 0x193 +#define GEN6_SURFACEFORMAT_R8G8B8_SNORM 0x194 +#define GEN6_SURFACEFORMAT_R8G8B8_SSCALED 0x195 +#define GEN6_SURFACEFORMAT_R8G8B8_USCALED 0x196 +#define GEN6_SURFACEFORMAT_R64G64B64A64_FLOAT 0x197 +#define GEN6_SURFACEFORMAT_R64G64B64_FLOAT 0x198 +#define GEN6_SURFACEFORMAT_BC4_SNORM 0x199 +#define GEN6_SURFACEFORMAT_BC5_SNORM 0x19A +#define GEN6_SURFACEFORMAT_R16G16B16_UNORM 0x19C +#define GEN6_SURFACEFORMAT_R16G16B16_SNORM 0x19D +#define GEN6_SURFACEFORMAT_R16G16B16_SSCALED 0x19E +#define GEN6_SURFACEFORMAT_R16G16B16_USCALED 0x19F + +#define GEN6_SURFACERETURNFORMAT_FLOAT32 0 +#define GEN6_SURFACERETURNFORMAT_S1 1 + +#define GEN6_SURFACE_1D 0 +#define GEN6_SURFACE_2D 1 +#define GEN6_SURFACE_3D 2 +#define GEN6_SURFACE_CUBE 3 +#define GEN6_SURFACE_BUFFER 4 +#define GEN6_SURFACE_NULL 7 + +#define GEN6_BORDER_COLOR_MODE_DEFAULT 0 +#define GEN6_BORDER_COLOR_MODE_LEGACY 1 + +#define GEN6_TEXCOORDMODE_WRAP 0 +#define GEN6_TEXCOORDMODE_MIRROR 1 +#define GEN6_TEXCOORDMODE_CLAMP 2 +#define GEN6_TEXCOORDMODE_CUBE 3 +#define GEN6_TEXCOORDMODE_CLAMP_BORDER 4 +#define GEN6_TEXCOORDMODE_MIRROR_ONCE 5 + +#define GEN6_THREAD_PRIORITY_NORMAL 0 +#define GEN6_THREAD_PRIORITY_HIGH 1 + +#define GEN6_TILEWALK_XMAJOR 0 +#define GEN6_TILEWALK_YMAJOR 1 + +#define GEN6_VERTEX_SUBPIXEL_PRECISION_8BITS 0 +#define GEN6_VERTEX_SUBPIXEL_PRECISION_4BITS 1 + +#define GEN6_VERTEXBUFFER_ACCESS_VERTEXDATA 0 +#define GEN6_VERTEXBUFFER_ACCESS_INSTANCEDATA 1 + +#define GEN6_VFCOMPONENT_NOSTORE 0 +#define GEN6_VFCOMPONENT_STORE_SRC 1 +#define GEN6_VFCOMPONENT_STORE_0 2 +#define GEN6_VFCOMPONENT_STORE_1_FLT 3 +#define GEN6_VFCOMPONENT_STORE_1_INT 4 +#define GEN6_VFCOMPONENT_STORE_VID 5 +#define GEN6_VFCOMPONENT_STORE_IID 6 +#define GEN6_VFCOMPONENT_STORE_PID 7 + + + +/* Execution Unit (EU) defines + */ + +#define GEN6_ALIGN_1 0 +#define GEN6_ALIGN_16 1 + +#define GEN6_ADDRESS_DIRECT 0 +#define GEN6_ADDRESS_REGISTER_INDIRECT_REGISTER 1 + +#define GEN6_CHANNEL_X 0 +#define GEN6_CHANNEL_Y 1 +#define GEN6_CHANNEL_Z 2 +#define GEN6_CHANNEL_W 3 + +#define GEN6_COMPRESSION_NONE 0 +#define GEN6_COMPRESSION_2NDHALF 1 +#define GEN6_COMPRESSION_COMPRESSED 2 + +#define GEN6_CONDITIONAL_NONE 0 +#define GEN6_CONDITIONAL_Z 1 +#define GEN6_CONDITIONAL_NZ 2 +#define GEN6_CONDITIONAL_EQ 1 /* Z */ +#define GEN6_CONDITIONAL_NEQ 2 /* NZ */ +#define GEN6_CONDITIONAL_G 3 +#define GEN6_CONDITIONAL_GE 4 +#define GEN6_CONDITIONAL_L 5 +#define GEN6_CONDITIONAL_LE 6 +#define GEN6_CONDITIONAL_C 7 +#define GEN6_CONDITIONAL_O 8 + +#define GEN6_DEBUG_NONE 0 +#define GEN6_DEBUG_BREAKPOINT 1 + +#define GEN6_DEPENDENCY_NORMAL 0 +#define GEN6_DEPENDENCY_NOTCLEARED 1 +#define GEN6_DEPENDENCY_NOTCHECKED 2 +#define GEN6_DEPENDENCY_DISABLE 3 + +#define GEN6_EXECUTE_1 0 +#define GEN6_EXECUTE_2 1 +#define GEN6_EXECUTE_4 2 +#define GEN6_EXECUTE_8 3 +#define GEN6_EXECUTE_16 4 +#define GEN6_EXECUTE_32 5 + +#define GEN6_HORIZONTAL_STRIDE_0 0 +#define GEN6_HORIZONTAL_STRIDE_1 1 +#define GEN6_HORIZONTAL_STRIDE_2 2 +#define GEN6_HORIZONTAL_STRIDE_4 3 + +#define GEN6_INSTRUCTION_NORMAL 0 +#define GEN6_INSTRUCTION_SATURATE 1 + +#define GEN6_MASK_ENABLE 0 +#define GEN6_MASK_DISABLE 1 + +#define GEN6_OPCODE_MOV 1 +#define GEN6_OPCODE_SEL 2 +#define GEN6_OPCODE_NOT 4 +#define GEN6_OPCODE_AND 5 +#define GEN6_OPCODE_OR 6 +#define GEN6_OPCODE_XOR 7 +#define GEN6_OPCODE_SHR 8 +#define GEN6_OPCODE_SHL 9 +#define GEN6_OPCODE_RSR 10 +#define GEN6_OPCODE_RSL 11 +#define GEN6_OPCODE_ASR 12 +#define GEN6_OPCODE_CMP 16 +#define GEN6_OPCODE_JMPI 32 +#define GEN6_OPCODE_IF 34 +#define GEN6_OPCODE_IFF 35 +#define GEN6_OPCODE_ELSE 36 +#define GEN6_OPCODE_ENDIF 37 +#define GEN6_OPCODE_DO 38 +#define GEN6_OPCODE_WHILE 39 +#define GEN6_OPCODE_BREAK 40 +#define GEN6_OPCODE_CONTINUE 41 +#define GEN6_OPCODE_HALT 42 +#define GEN6_OPCODE_MSAVE 44 +#define GEN6_OPCODE_MRESTORE 45 +#define GEN6_OPCODE_PUSH 46 +#define GEN6_OPCODE_POP 47 +#define GEN6_OPCODE_WAIT 48 +#define GEN6_OPCODE_SEND 49 +#define GEN6_OPCODE_ADD 64 +#define GEN6_OPCODE_MUL 65 +#define GEN6_OPCODE_AVG 66 +#define GEN6_OPCODE_FRC 67 +#define GEN6_OPCODE_RNDU 68 +#define GEN6_OPCODE_RNDD 69 +#define GEN6_OPCODE_RNDE 70 +#define GEN6_OPCODE_RNDZ 71 +#define GEN6_OPCODE_MAC 72 +#define GEN6_OPCODE_MACH 73 +#define GEN6_OPCODE_LZD 74 +#define GEN6_OPCODE_SAD2 80 +#define GEN6_OPCODE_SADA2 81 +#define GEN6_OPCODE_DP4 84 +#define GEN6_OPCODE_DPH 85 +#define GEN6_OPCODE_DP3 86 +#define GEN6_OPCODE_DP2 87 +#define GEN6_OPCODE_DPA2 88 +#define GEN6_OPCODE_LINE 89 +#define GEN6_OPCODE_NOP 126 + +#define GEN6_PREDICATE_NONE 0 +#define GEN6_PREDICATE_NORMAL 1 +#define GEN6_PREDICATE_ALIGN1_ANYV 2 +#define GEN6_PREDICATE_ALIGN1_ALLV 3 +#define GEN6_PREDICATE_ALIGN1_ANY2H 4 +#define GEN6_PREDICATE_ALIGN1_ALL2H 5 +#define GEN6_PREDICATE_ALIGN1_ANY4H 6 +#define GEN6_PREDICATE_ALIGN1_ALL4H 7 +#define GEN6_PREDICATE_ALIGN1_ANY8H 8 +#define GEN6_PREDICATE_ALIGN1_ALL8H 9 +#define GEN6_PREDICATE_ALIGN1_ANY16H 10 +#define GEN6_PREDICATE_ALIGN1_ALL16H 11 +#define GEN6_PREDICATE_ALIGN16_REPLICATE_X 2 +#define GEN6_PREDICATE_ALIGN16_REPLICATE_Y 3 +#define GEN6_PREDICATE_ALIGN16_REPLICATE_Z 4 +#define GEN6_PREDICATE_ALIGN16_REPLICATE_W 5 +#define GEN6_PREDICATE_ALIGN16_ANY4H 6 +#define GEN6_PREDICATE_ALIGN16_ALL4H 7 + +#define GEN6_ARCHITECTURE_REGISTER_FILE 0 +#define GEN6_GENERAL_REGISTER_FILE 1 +#define GEN6_MESSAGE_REGISTER_FILE 2 +#define GEN6_IMMEDIATE_VALUE 3 + +#define GEN6_REGISTER_TYPE_UD 0 +#define GEN6_REGISTER_TYPE_D 1 +#define GEN6_REGISTER_TYPE_UW 2 +#define GEN6_REGISTER_TYPE_W 3 +#define GEN6_REGISTER_TYPE_UB 4 +#define GEN6_REGISTER_TYPE_B 5 +#define GEN6_REGISTER_TYPE_VF 5 /* packed float vector, immediates only? */ +#define GEN6_REGISTER_TYPE_HF 6 +#define GEN6_REGISTER_TYPE_V 6 /* packed int vector, immediates only, uword dest only */ +#define GEN6_REGISTER_TYPE_F 7 + +#define GEN6_ARF_NULL 0x00 +#define GEN6_ARF_ADDRESS 0x10 +#define GEN6_ARF_ACCUMULATOR 0x20 +#define GEN6_ARF_FLAG 0x30 +#define GEN6_ARF_MASK 0x40 +#define GEN6_ARF_MASK_STACK 0x50 +#define GEN6_ARF_MASK_STACK_DEPTH 0x60 +#define GEN6_ARF_STATE 0x70 +#define GEN6_ARF_CONTROL 0x80 +#define GEN6_ARF_NOTIFICATION_COUNT 0x90 +#define GEN6_ARF_IP 0xA0 + +#define GEN6_AMASK 0 +#define GEN6_IMASK 1 +#define GEN6_LMASK 2 +#define GEN6_CMASK 3 + + + +#define GEN6_THREAD_NORMAL 0 +#define GEN6_THREAD_ATOMIC 1 +#define GEN6_THREAD_SWITCH 2 + +#define GEN6_VERTICAL_STRIDE_0 0 +#define GEN6_VERTICAL_STRIDE_1 1 +#define GEN6_VERTICAL_STRIDE_2 2 +#define GEN6_VERTICAL_STRIDE_4 3 +#define GEN6_VERTICAL_STRIDE_8 4 +#define GEN6_VERTICAL_STRIDE_16 5 +#define GEN6_VERTICAL_STRIDE_32 6 +#define GEN6_VERTICAL_STRIDE_64 7 +#define GEN6_VERTICAL_STRIDE_128 8 +#define GEN6_VERTICAL_STRIDE_256 9 +#define GEN6_VERTICAL_STRIDE_ONE_DIMENSIONAL 0xF + +#define GEN6_WIDTH_1 0 +#define GEN6_WIDTH_2 1 +#define GEN6_WIDTH_4 2 +#define GEN6_WIDTH_8 3 +#define GEN6_WIDTH_16 4 + +#define GEN6_STATELESS_BUFFER_BOUNDARY_1K 0 +#define GEN6_STATELESS_BUFFER_BOUNDARY_2K 1 +#define GEN6_STATELESS_BUFFER_BOUNDARY_4K 2 +#define GEN6_STATELESS_BUFFER_BOUNDARY_8K 3 +#define GEN6_STATELESS_BUFFER_BOUNDARY_16K 4 +#define GEN6_STATELESS_BUFFER_BOUNDARY_32K 5 +#define GEN6_STATELESS_BUFFER_BOUNDARY_64K 6 +#define GEN6_STATELESS_BUFFER_BOUNDARY_128K 7 +#define GEN6_STATELESS_BUFFER_BOUNDARY_256K 8 +#define GEN6_STATELESS_BUFFER_BOUNDARY_512K 9 +#define GEN6_STATELESS_BUFFER_BOUNDARY_1M 10 +#define GEN6_STATELESS_BUFFER_BOUNDARY_2M 11 + +#define GEN6_POLYGON_FACING_FRONT 0 +#define GEN6_POLYGON_FACING_BACK 1 + +#define GEN6_MESSAGE_TARGET_NULL 0 +#define GEN6_MESSAGE_TARGET_MATH 1 +#define GEN6_MESSAGE_TARGET_SAMPLER 2 +#define GEN6_MESSAGE_TARGET_GATEWAY 3 +#define GEN6_MESSAGE_TARGET_DATAPORT_READ 4 +#define GEN6_MESSAGE_TARGET_DATAPORT_WRITE 5 +#define GEN6_MESSAGE_TARGET_URB 6 +#define GEN6_MESSAGE_TARGET_THREAD_SPAWNER 7 + +#define GEN6_SAMPLER_RETURN_FORMAT_FLOAT32 0 +#define GEN6_SAMPLER_RETURN_FORMAT_UINT32 2 +#define GEN6_SAMPLER_RETURN_FORMAT_SINT32 3 + +#define GEN6_SAMPLER_MESSAGE_SIMD8_SAMPLE 0 +#define GEN6_SAMPLER_MESSAGE_SIMD16_SAMPLE 0 +#define GEN6_SAMPLER_MESSAGE_SIMD16_SAMPLE_BIAS 0 +#define GEN6_SAMPLER_MESSAGE_SIMD8_KILLPIX 1 +#define GEN6_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_LOD 1 +#define GEN6_SAMPLER_MESSAGE_SIMD16_SAMPLE_LOD 1 +#define GEN6_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_GRADIENTS 2 +#define GEN6_SAMPLER_MESSAGE_SIMD8_SAMPLE_GRADIENTS 2 +#define GEN6_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_COMPARE 0 +#define GEN6_SAMPLER_MESSAGE_SIMD16_SAMPLE_COMPARE 2 +#define GEN6_SAMPLER_MESSAGE_SIMD4X2_RESINFO 2 +#define GEN6_SAMPLER_MESSAGE_SIMD8_RESINFO 2 +#define GEN6_SAMPLER_MESSAGE_SIMD16_RESINFO 2 +#define GEN6_SAMPLER_MESSAGE_SIMD4X2_LD 3 +#define GEN6_SAMPLER_MESSAGE_SIMD8_LD 3 +#define GEN6_SAMPLER_MESSAGE_SIMD16_LD 3 + +#define GEN6_DATAPORT_OWORD_BLOCK_1_OWORDLOW 0 +#define GEN6_DATAPORT_OWORD_BLOCK_1_OWORDHIGH 1 +#define GEN6_DATAPORT_OWORD_BLOCK_2_OWORDS 2 +#define GEN6_DATAPORT_OWORD_BLOCK_4_OWORDS 3 +#define GEN6_DATAPORT_OWORD_BLOCK_8_OWORDS 4 + +#define GEN6_DATAPORT_OWORD_DUAL_BLOCK_1OWORD 0 +#define GEN6_DATAPORT_OWORD_DUAL_BLOCK_4OWORDS 2 + +#define GEN6_DATAPORT_DWORD_SCATTERED_BLOCK_8DWORDS 2 +#define GEN6_DATAPORT_DWORD_SCATTERED_BLOCK_16DWORDS 3 + +#define GEN6_DATAPORT_READ_MESSAGE_OWORD_BLOCK_READ 0 +#define GEN6_DATAPORT_READ_MESSAGE_OWORD_DUAL_BLOCK_READ 1 +#define GEN6_DATAPORT_READ_MESSAGE_DWORD_BLOCK_READ 2 +#define GEN6_DATAPORT_READ_MESSAGE_DWORD_SCATTERED_READ 3 + +#define GEN6_DATAPORT_READ_TARGET_DATA_CACHE 0 +#define GEN6_DATAPORT_READ_TARGET_RENDER_CACHE 1 +#define GEN6_DATAPORT_READ_TARGET_SAMPLER_CACHE 2 + +#define GEN6_DATAPORT_RENDER_TARGET_WRITE_SIMD16_SINGLE_SOURCE 0 +#define GEN6_DATAPORT_RENDER_TARGET_WRITE_SIMD16_SINGLE_SOURCE_REPLICATED 1 +#define GEN6_DATAPORT_RENDER_TARGET_WRITE_SIMD8_DUAL_SOURCE_SUBSPAN01 2 +#define GEN6_DATAPORT_RENDER_TARGET_WRITE_SIMD8_DUAL_SOURCE_SUBSPAN23 3 +#define GEN6_DATAPORT_RENDER_TARGET_WRITE_SIMD8_SINGLE_SOURCE_SUBSPAN01 4 + +#define GEN6_DATAPORT_WRITE_MESSAGE_OWORD_BLOCK_WRITE 0 +#define GEN6_DATAPORT_WRITE_MESSAGE_OWORD_DUAL_BLOCK_WRITE 1 +#define GEN6_DATAPORT_WRITE_MESSAGE_DWORD_BLOCK_WRITE 2 +#define GEN6_DATAPORT_WRITE_MESSAGE_DWORD_SCATTERED_WRITE 3 +#define GEN6_DATAPORT_WRITE_MESSAGE_RENDER_TARGET_WRITE 4 +#define GEN6_DATAPORT_WRITE_MESSAGE_STREAMED_VERTEX_BUFFER_WRITE 5 +#define GEN6_DATAPORT_WRITE_MESSAGE_FLUSH_RENDER_CACHE 7 + +#define GEN6_MATH_FUNCTION_INV 1 +#define GEN6_MATH_FUNCTION_LOG 2 +#define GEN6_MATH_FUNCTION_EXP 3 +#define GEN6_MATH_FUNCTION_SQRT 4 +#define GEN6_MATH_FUNCTION_RSQ 5 +#define GEN6_MATH_FUNCTION_SIN 6 /* was 7 */ +#define GEN6_MATH_FUNCTION_COS 7 /* was 8 */ +#define GEN6_MATH_FUNCTION_SINCOS 8 /* was 6 */ +#define GEN6_MATH_FUNCTION_TAN 9 +#define GEN6_MATH_FUNCTION_POW 10 +#define GEN6_MATH_FUNCTION_INT_DIV_QUOTIENT_AND_REMAINDER 11 +#define GEN6_MATH_FUNCTION_INT_DIV_QUOTIENT 12 +#define GEN6_MATH_FUNCTION_INT_DIV_REMAINDER 13 + +#define GEN6_MATH_INTEGER_UNSIGNED 0 +#define GEN6_MATH_INTEGER_SIGNED 1 + +#define GEN6_MATH_PRECISION_FULL 0 +#define GEN6_MATH_PRECISION_PARTIAL 1 + +#define GEN6_MATH_SATURATE_NONE 0 +#define GEN6_MATH_SATURATE_SATURATE 1 + +#define GEN6_MATH_DATA_VECTOR 0 +#define GEN6_MATH_DATA_SCALAR 1 + +#define GEN6_URB_OPCODE_WRITE 0 + +#define GEN6_URB_SWIZZLE_NONE 0 +#define GEN6_URB_SWIZZLE_INTERLEAVE 1 +#define GEN6_URB_SWIZZLE_TRANSPOSE 2 + +#define GEN6_SCRATCH_SPACE_SIZE_1K 0 +#define GEN6_SCRATCH_SPACE_SIZE_2K 1 +#define GEN6_SCRATCH_SPACE_SIZE_4K 2 +#define GEN6_SCRATCH_SPACE_SIZE_8K 3 +#define GEN6_SCRATCH_SPACE_SIZE_16K 4 +#define GEN6_SCRATCH_SPACE_SIZE_32K 5 +#define GEN6_SCRATCH_SPACE_SIZE_64K 6 +#define GEN6_SCRATCH_SPACE_SIZE_128K 7 +#define GEN6_SCRATCH_SPACE_SIZE_256K 8 +#define GEN6_SCRATCH_SPACE_SIZE_512K 9 +#define GEN6_SCRATCH_SPACE_SIZE_1M 10 +#define GEN6_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 gen6_sampler_default_border_color { + float color[4]; +}; + +struct gen6_sampler_legacy_border_color { + uint8_t color[4]; +}; + +struct gen6_sampler_state { + struct { + uint32_t shadow_function:3; + uint32_t lod_bias:11; + uint32_t min_filter:3; + uint32_t mag_filter:3; + uint32_t mip_filter:2; + uint32_t base_level:5; + uint32_t pad:1; + uint32_t lod_preclamp:1; + uint32_t border_color_mode:1; + uint32_t pad0:1; + uint32_t disable:1; + } ss0; + + struct { + uint32_t r_wrap_mode:3; + uint32_t t_wrap_mode:3; + uint32_t s_wrap_mode:3; + uint32_t pad:3; + uint32_t max_lod:10; + uint32_t min_lod:10; + } ss1; + + struct { + uint32_t border_color; + } ss2; + + struct { + uint32_t pad:19; + 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 monochrome_filter_width:3; + uint32_t monochrome_filter_height:3; + } ss3; +}; + +struct gen6_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 ia_blend_func:3; + uint32_t pad0:1; + 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 gen6_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 gen6_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 gen6_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 pad:3; + uint32_t render_cache_read_mode:1; + uint32_t mipmap_layout_mode:1; + uint32_t vert_line_stride_ofs:1; + uint32_t vert_line_stride:1; + uint32_t color_blend:1; + uint32_t writedisable_blue:1; + uint32_t writedisable_green:1; + uint32_t writedisable_red:1; + uint32_t writedisable_alpha:1; + uint32_t surface_format:9; + uint32_t data_return_format:1; + uint32_t pad0:1; + uint32_t surface_type:3; + } ss0; + + struct { + uint32_t base_addr; + } ss1; + + struct { + uint32_t render_target_rotation:2; + uint32_t mip_count:4; + uint32_t width:13; + uint32_t height:13; + } ss2; + + struct { + uint32_t tile_walk:1; + uint32_t tiled_surface:1; + uint32_t pad:1; + uint32_t pitch:18; + uint32_t depth:11; + } ss3; + + struct { + uint32_t pad:19; + uint32_t min_array_elt:9; + uint32_t min_lod:4; + } ss4; + + struct { + uint32_t pad:20; + uint32_t y_offset:4; + uint32_t pad2:1; + uint32_t x_offset:7; + } ss5; +}; + +struct gen6_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 |