From 3edfff16cde0033fbec3c2deb390212acff15d6d Mon Sep 17 00:00:00 2001 From: Jesse Barnes Date: Wed, 27 Feb 2013 14:51:32 +0000 Subject: [PATCH] rendercopy/bdw: Initial gen8 rendercopy Signed-off-by: Jesse Barnes Signed-off-by: Ben Widawsky --- lib/Makefile.am | 1 + lib/gen8_render.h | 302 +++++++++++++++++++ lib/rendercopy.h | 4 + lib/rendercopy_gen8.c | 786 ++++++++++++++++++++++++++++++++++++++++++++++++++ lib/rendercopy_i830.c | 2 + shaders/ps/blit.g8a | 66 +++++ 6 files changed, 1161 insertions(+) create mode 100644 lib/gen8_render.h create mode 100644 lib/rendercopy_gen8.c create mode 100644 shaders/ps/blit.g8a diff --git a/lib/Makefile.am b/lib/Makefile.am index d7f4275..114dc49 100644 --- a/lib/Makefile.am +++ b/lib/Makefile.am @@ -31,6 +31,7 @@ libintel_tools_la_SOURCES = \ gen7_render.h \ rendercopy_gen6.c \ rendercopy_gen7.c \ + rendercopy_gen8.c \ rendercopy.h \ rendercopy.c \ intel_reg_map.c \ diff --git a/lib/gen8_render.h b/lib/gen8_render.h new file mode 100644 index 0000000..a5fbf84 --- /dev/null +++ b/lib/gen8_render.h @@ -0,0 +1,302 @@ +#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 gen8_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 media_boundary_pixel_mode:2; + uint32_t render_cache_read_write:1; + uint32_t smapler_l2_bypass:1; + uint32_t vert_line_stride_ofs:1; + uint32_t vert_line_stride:1; + uint32_t tiled_mode:2; + uint32_t horizontal_alignment:2; + uint32_t vertical_alignment:2; + uint32_t surface_format:9; /**< BRW_SURFACEFORMAT_x */ + uint32_t pad0:1; + uint32_t is_array:1; + uint32_t surface_type:3; /**< BRW_SURFACE_1D/2D/3D/CUBE */ + } ss0; + + struct { + uint32_t qpitch:15; + uint32_t pad1:4; + uint32_t base_mip_level:5; + uint32_t memory_object_control:7; + uint32_t pad0:1; + } ss1; + + struct { + uint32_t width:14; + uint32_t pad1:2; + uint32_t height:14; + uint32_t pad0:2; + } ss2; + + struct { + uint32_t pitch:18; + uint32_t pad:3; + uint32_t depth:11; + } ss3; + + struct { + uint32_t minimum_array_element:27; + uint32_t pad0:5; + } ss4; + + struct { + uint32_t mip_count:4; + uint32_t min_lod:4; + uint32_t pad3:6; + uint32_t coherency_type:1; + uint32_t pad2:5; + uint32_t ewa_disable_for_cube:1; + uint32_t y_offset:3; + uint32_t pad0:1; + uint32_t x_offset:7; + } ss5; + + struct { + uint32_t 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; + } ss7; + + struct { + uint32_t base_addr; + } ss8; + + struct { + uint32_t base_addr_hi:16; + uint32_t pad0:16; + } ss9; + + struct { + uint32_t pad0:12; + uint32_t aux_base_addr:20; + } ss10; + + struct { + uint32_t aux_base_addr_hi:16; + uint32_t pad:16; + } ss11; + + struct { + uint32_t hiz_depth_clear_value; + } ss12; + + struct { + uint32_t reserved; + } ss13; + + struct { + uint32_t reserved; + } ss14; + + struct { + uint32_t reserved; + } ss15; +}; + +struct gen8_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 lod_preclamp:2; + uint32_t default_color_mode:1; + uint32_t pad0:1; + uint32_t disable:1; + } ss0; + + struct + { + uint32_t cube_control_mode:1; + uint32_t shadow_function:3; + uint32_t chromakey_mode:1; + uint32_t chromakey_index:2; + uint32_t chromakey_enable:1; + uint32_t max_lod:12; + uint32_t min_lod:12; + } ss1; + + struct + { + uint32_t lod_clamp_mag_mode:1; + uint32_t flexible_filter_valign:1; + uint32_t flexible_filter_halign:1; + uint32_t flexible_filter_coeff_size:1; + uint32_t flexible_filter_mode:1; + uint32_t pad1:1; + uint32_t indirect_state_ptr:18; + uint32_t pad0:2; + uint32_t sep_filter_height:2; + uint32_t sep_filter_width:2; + uint32_t sep_filter_coeff_table_size:2; + } 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 pad0:2; + uint32_t non_sep_filter_footprint_mask:8; + } ss3; +}; + +struct gen8_blend_state { + struct { + uint32_t pad0:19; + uint32_t y_dither_offset:2; + uint32_t x_dither_offset:2; + uint32_t dither_enable:1; + uint32_t alpha_test_func:3; + uint32_t alpha_test:1; + uint32_t alpha_to_coverage_dither:1; + uint32_t alpha_to_one:1; + uint32_t ia_blend:1; + uint32_t alpha_to_coverage:1; + } bs0; + + struct { + uint64_t write_disable_blue:1; + uint64_t write_disable_green:1; + uint64_t write_disable_red:1; + uint64_t write_disable_alpha:1; + uint64_t pad1:1; + uint64_t alpha_blend_func:3; + uint64_t dest_alpha_blend_factor:5; + uint64_t source_alpha_blend_factor:5; + uint64_t color_blend_func:3; + uint64_t dest_blend_factor:5; + uint64_t source_blend_factor:5; + uint64_t color_buffer_blend:1; + uint64_t post_blend_color_clamp:1; + uint64_t pre_blend_color_clamp:1; + uint64_t color_clamp_range:2; + uint64_t pre_blend_source_only_clamp:1; + uint64_t pad0:22; + uint64_t logic_op_func:4; + uint64_t logic_op_enable:1; + } bs[16]; +}; + +struct gen7_sf_clip_viewport { + struct { + float m00; + float m11; + float m22; + float m30; + float m31; + float m32; + } viewport; + + uint32_t pad0[2]; + + struct { + float xmin; + float xmax; + float ymin; + float ymax; + } guardband; + + float pad1[4]; +}; + +struct gen6_scissor_rect +{ + uint32_t xmin:16; + uint32_t ymin:16; + uint32_t xmax:16; + uint32_t ymax:16; +}; + +#endif diff --git a/lib/rendercopy.h b/lib/rendercopy.h index ace4392..4605264 100644 --- a/lib/rendercopy.h +++ b/lib/rendercopy.h @@ -68,6 +68,10 @@ typedef void (*render_copyfunc_t)(struct intel_batchbuffer *batch, render_copyfunc_t get_render_copyfunc(int devid); +void gen8_render_copyfunc(struct intel_batchbuffer *batch, + struct scratch_buf *src, unsigned src_x, unsigned src_y, + unsigned width, unsigned height, + struct scratch_buf *dst, unsigned dst_x, unsigned dst_y); void gen7_render_copyfunc(struct intel_batchbuffer *batch, struct scratch_buf *src, unsigned src_x, unsigned src_y, unsigned width, unsigned height, diff --git a/lib/rendercopy_gen8.c b/lib/rendercopy_gen8.c new file mode 100644 index 0000000..cc371b3 --- /dev/null +++ b/lib/rendercopy_gen8.c @@ -0,0 +1,786 @@ +#include "rendercopy.h" +#include "gen8_render.h" + +#include + +#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; +} 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, 0x21403ae8, 0x3a0000c0, 0x008d0040 }, + { 0x0060005a, 0x21603ae8, 0x3a0000c0, 0x008d0080 }, + { 0x0060005a, 0x21803ae8, 0x3a0000d0, 0x008d0040 }, + { 0x0060005a, 0x21a03ae8, 0x3a0000d0, 0x008d0080 }, + { 0x02800031, 0x2e0022e8, 0x0e000140, 0x08840001 }, + { 0x05800031, 0x200022e0, 0x0e000e00, 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 +}; + +static uint32_t +batch_used(struct intel_batchbuffer *batch) +{ + return batch->ptr - batch->buffer; +} + +static uint32_t +batch_align(struct intel_batchbuffer *batch, uint32_t align) +{ + uint32_t offset = batch_used(batch); + offset = ALIGN(offset, align); + batch->ptr = batch->buffer + offset; + return offset; +} + +static void * +batch_alloc(struct intel_batchbuffer *batch, uint32_t size, uint32_t align) +{ + uint32_t offset = batch_align(batch, align); + batch->ptr += size; + return memset(batch->buffer + offset, 0, size); +} + +static uint32_t +batch_offset(struct intel_batchbuffer *batch, void *ptr) +{ + return (uint8_t *)ptr - batch->buffer; +} + +static uint32_t +batch_copy(struct intel_batchbuffer *batch, const void *ptr, uint32_t size, uint32_t align) +{ + return batch_offset(batch, memcpy(batch_alloc(batch, size, align), ptr, size)); +} + +static void +gen6_render_flush(struct intel_batchbuffer *batch, 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); +} + +/* Mostly copy+paste from gen6, except height, width, pitch moved */ +static uint32_t +gen8_bind_buf(struct intel_batchbuffer *batch, struct scratch_buf *buf, + uint32_t format, int is_dst) { + struct gen8_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(batch, sizeof(*ss), 32); + ss->ss0.surface_type = GEN6_SURFACE_2D; + ss->ss0.surface_format = format; + ss->ss0.render_cache_read_write = 1; + if (buf->tiling == I915_TILING_X) + ss->ss0.tiled_mode = 2; + else if (buf->tiling == I915_TILING_Y) + ss->ss0.tiled_mode = 3; + + ss->ss8.base_addr = buf->bo->offset; + + ret = drm_intel_bo_emit_reloc(batch->bo, + batch_offset(batch, 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->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 +gen8_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] = + gen8_bind_buf(batch, dst, GEN6_SURFACEFORMAT_B8G8R8A8_UNORM, 1); + binding_table[1] = + gen8_bind_buf(batch, src, GEN6_SURFACEFORMAT_B8G8R8A8_UNORM, 0); + + return batch_offset(batch, binding_table); +} + +/* Mostly copy+paste from gen6, except wrap modes moved */ +static uint32_t +gen8_create_sampler(struct intel_batchbuffer *batch) { + struct gen8_sampler_state *ss; + + ss = batch_alloc(batch, sizeof(*ss), 32); + + 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; + + /* 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; + + return batch_offset(batch, ss); +} + +/** + * 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); + 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); +} + +static uint32_t +gen6_create_cc_state(struct intel_batchbuffer *batch) +{ + struct gen6_color_calc_state *cc_state; + cc_state = batch_alloc(batch, sizeof(*cc_state), 64); + return batch_offset(batch, cc_state); +} + +static uint32_t +gen8_create_blend_state(struct intel_batchbuffer *batch) +{ + struct gen8_blend_state *blend; + int i; + + blend = batch_alloc(batch, sizeof(*blend), 64); + for (i = 0; i < 16; i++) { + blend->bs[i].pre_blend_color_clamp = 1; + blend->bs[i].color_buffer_blend = 0; + } + return batch_offset(batch, blend); +} + +static uint32_t +gen6_create_cc_viewport(struct intel_batchbuffer *batch) +{ + struct gen6_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 uint32_t +gen6_create_scissor_rect(struct intel_batchbuffer *batch) +{ + struct gen6_scissor_rect *scissor; + scissor = batch_alloc(batch, sizeof(*scissor), 64); + return batch_offset(batch, scissor); +} + + + + + +static void +gen6_emit_sip(struct intel_batchbuffer *batch) { + OUT_BATCH(GEN6_STATE_SIP | 0); + OUT_BATCH(0); +} + +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); +} + +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); + + 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); +} + +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); +} + +static void +gen8_emit_cc(struct intel_batchbuffer *batch) { + OUT_BATCH(GEN7_3DSTATE_BLEND_STATE_POINTERS); + OUT_BATCH(cc.blend_state | 1); + + OUT_BATCH(GEN6_3DSTATE_CC_STATE_POINTERS); + OUT_BATCH(cc.cc_state | 1); +} + +static void +gen7_emit_multisample(struct intel_batchbuffer *batch) { + OUT_BATCH(GEN6_3DSTATE_MULTISAMPLE | 2); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + + OUT_BATCH(GEN6_3DSTATE_SAMPLE_MASK); + OUT_BATCH(1); +} + +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); +} + +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); +} + +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); +} + +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); +} + +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); +} + +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); +} + +static void +gen7_emit_sf(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(0); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + 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); + OUT_BATCH(0); +} + +static void +gen8_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); + + OUT_BATCH(GEN6_3DSTATE_CONSTANT_PS | (7-2)); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + + OUT_BATCH(GEN7_3DSTATE_PS | (10-2)); + OUT_BATCH(kernel); + OUT_BATCH(0); /* kernel hi */ + OUT_BATCH(1 << GEN6_3DSTATE_WM_SAMPLER_COUNT_SHITF | + 2 << GEN6_3DSTATE_WM_BINDING_TABLE_ENTRY_COUNT_SHIFT); + OUT_BATCH(0); /* scratch space stuff */ + OUT_BATCH(0); /* scratch hi */ + OUT_BATCH((max_threads - 1) << GEN7_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 1 hi */ +} + +static void +gen8_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); + + OUT_BATCH(GEN7_3DSTATE_STENCIL_BUFFER | (3-2)); + OUT_BATCH(0); + 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 +} + +static void +gen6_emit_drawing_rectangle(struct intel_batchbuffer *batch, 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); +} + +/* 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. + * + */ + +#define BATCH_STATE_SPLIT 2048 +void gen8_render_copyfunc(struct intel_batchbuffer *batch, + struct scratch_buf *src, unsigned src_x, unsigned src_y, + 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 = gen8_bind_surfaces(batch, src, dst); + ps_sampler_state = gen8_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.blend_state = gen8_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; + + /* 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); + + 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); + + gen8_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_clip(batch); + + gen7_emit_sf(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); + + gen8_emit_ps(batch, ps_kernel_off); + + OUT_BATCH(GEN6_3DSTATE_SCISSOR_STATE_POINTERS); + OUT_BATCH(scissor_state); + + gen8_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); + + OUT_BATCH(MI_BATCH_BUFFER_END); + + batch_end = batch_align(batch, 8); + assert(batch_end < BATCH_STATE_SPLIT); + + dump_batch(batch); + + gen6_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 diff --git a/lib/rendercopy_i830.c b/lib/rendercopy_i830.c index 20f3dc6..0347597 100644 --- a/lib/rendercopy_i830.c +++ b/lib/rendercopy_i830.c @@ -240,6 +240,8 @@ render_copyfunc_t get_render_copyfunc(int devid) copy = gen6_render_copyfunc; else if (IS_GEN7(devid)) copy = gen7_render_copyfunc; + else if (IS_GEN8(devid)) + copy = gen8_render_copyfunc; return copy; } diff --git a/shaders/ps/blit.g8a b/shaders/ps/blit.g8a new file mode 100644 index 0000000..deeedcc --- /dev/null +++ b/shaders/ps/blit.g8a @@ -0,0 +1,66 @@ +/* Assemble with ".../intel-gen4asm/src/intel-gen4asm -g 7" */ + + +/* Move pixels into g10-g13. The pixel shaader does not load what you want. It + * loads the input data for a plane function to calculate what you want. The + * following is boiler plate code to move our normalized texture coordinates + * (u,v) into g10-g13. It does this 4 subspans (16 pixels) at a time. + * + * This should do the same thing, but it doesn't work for some reason. + * pln(16) g10 g6<0,1,0>F g2<8,8,1>F { align1 }; + * pln(16) g12 g6.16<1>F g2<8,8,1>F { align1 }; + */ +/* U */ +pln (8) g10<1>F g6.0<0,1,0>F g2.0<8,8,1>F { align1 }; /* pixel 0-7 */ +pln (8) g11<1>F g6.0<0,1,0>F g4.0<8,8,1>F { align1 }; /* pixel 8-15 */ +/* V */ +pln (8) g12<1>F g6.16<0,1,0> g2.0<8,8,1>F { align1 }; /* pixel 0-7 */ +pln (8) g13<1>F g6.16<0,1,0> g4.0<8,8,1>F { align1 }; /* pixel 8-15 */ + + +/* Next the we want the sampler to fetch the src texture (ie. src buffer). This + * is done with a pretty simple send message. The output goes to g112, which is + * exactly what we're supposed to use in our final send message. + * In intel-gen4asm, we should end up parsed by the following rule: + * predicate SEND execsize dst sendleadreg sndopr directsrcoperand instoptions + * + * Send message descriptor: + * 28:25 = message len = 4 // our 4 registers have 16 pixels + * 24:20 = response len = 8 // Each pixel is RGBA32, so we need 8 registers + * 19:19 = header present = 0 + * 18:17 = SIMD16 = 2 + * 16:12 = TYPE = 0 (regular sample) + * 11:08 = Sampler index = ignored/0 + * 7:0 = binding table index = src = 1 + * 0x8840001 + * + * Send message extra descriptor + * 5:5 = End of Thread = 0 + * 3:0 = Target Function ID = SFID_SAMPLER (2) + * 0x2 + */ + +send(16) g112 g10 0x2 0x8840001 { align1 }; + +/* g112-g119 now contains the sample source input, and all we must do is write + * it out to the destination render target. This is done with the send message + * as well. The only extra bits are to terminate the pixel shader. + * + * Send message descriptor: + * 28:25 = message len = 8 // 16 pixels RGBA32 + * 24:20 = response len = 0 + * 19:19 = header present = 0 + * 17:14 = message type = Render Target Write (12) + * 12:12 = Last Render Target Select = 1 + * 10:08 = Message Type = SIMD16 (0) + * 07:00 = Binding Table Index = dest = 0 + * 0x10031000 + * + * Send message extra descriptor + * 5:5 = End of Thread = 1 + * 3:0 = Target Function ID = SFID_DP_RC (5) + * 0x25 + */ +send(16) null g112 0x25 0x10031000 { align1, EOT }; + +/* vim: set ft=c ts=4 sw=2 tw=80 et: */ -- 2.7.4