diff options
author | Chia-I Wu <olvaffe@gmail.com> | 2014-11-10 12:55:34 +0800 |
---|---|---|
committer | Chia-I Wu <olvaffe@gmail.com> | 2014-11-10 13:14:04 +0800 |
commit | c88c49baf420556a130041044a7f074d6368f080 (patch) | |
tree | 2decddeb850b7d44a980b6139f66524b8f7f5975 /src | |
parent | f3b709c0ac073cd0ec90a3a0d91d1ee94668e043 (diff) |
ilo: use top/bottom split for state functions
Follow the builder and split state functions into top (vertex processing) and
bottom (pixel processing).
Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Diffstat (limited to 'src')
-rw-r--r-- | src/gallium/drivers/ilo/Makefile.sources | 6 | ||||
-rw-r--r-- | src/gallium/drivers/ilo/ilo_blitter_rectlist.c | 2 | ||||
-rw-r--r-- | src/gallium/drivers/ilo/ilo_builder_3d_top.h | 2 | ||||
-rw-r--r-- | src/gallium/drivers/ilo/ilo_shader.c | 2 | ||||
-rw-r--r-- | src/gallium/drivers/ilo/ilo_state.c | 2 | ||||
-rw-r--r-- | src/gallium/drivers/ilo/ilo_state_3d.h (renamed from src/gallium/drivers/ilo/ilo_state_gen.h) | 8 | ||||
-rw-r--r-- | src/gallium/drivers/ilo/ilo_state_3d_bottom.c (renamed from src/gallium/drivers/ilo/ilo_state_gen6.c) | 1479 | ||||
-rw-r--r-- | src/gallium/drivers/ilo/ilo_state_3d_top.c | 1566 | ||||
-rw-r--r-- | src/gallium/drivers/ilo/ilo_state_gen7.c | 641 |
9 files changed, 1856 insertions, 1852 deletions
diff --git a/src/gallium/drivers/ilo/Makefile.sources b/src/gallium/drivers/ilo/Makefile.sources index 9e7977f7cee..8dd2f9b5b79 100644 --- a/src/gallium/drivers/ilo/Makefile.sources +++ b/src/gallium/drivers/ilo/Makefile.sources @@ -48,9 +48,9 @@ C_SOURCES := \ ilo_shader.h \ ilo_state.c \ ilo_state.h \ - ilo_state_gen.h \ - ilo_state_gen6.c \ - ilo_state_gen7.c \ + ilo_state_3d.h \ + ilo_state_3d_bottom.c \ + ilo_state_3d_top.c \ ilo_transfer.c \ ilo_transfer.h \ ilo_video.c \ diff --git a/src/gallium/drivers/ilo/ilo_blitter_rectlist.c b/src/gallium/drivers/ilo/ilo_blitter_rectlist.c index 5ce8b530eae..0efe82082cf 100644 --- a/src/gallium/drivers/ilo/ilo_blitter_rectlist.c +++ b/src/gallium/drivers/ilo/ilo_blitter_rectlist.c @@ -30,7 +30,7 @@ #include "ilo_draw.h" #include "ilo_state.h" -#include "ilo_state_gen.h" +#include "ilo_state_3d.h" #include "ilo_blit.h" #include "ilo_blitter.h" diff --git a/src/gallium/drivers/ilo/ilo_builder_3d_top.h b/src/gallium/drivers/ilo/ilo_builder_3d_top.h index 85f1b3b3e28..5c4fbf993ce 100644 --- a/src/gallium/drivers/ilo/ilo_builder_3d_top.h +++ b/src/gallium/drivers/ilo/ilo_builder_3d_top.h @@ -35,7 +35,7 @@ #include "ilo_resource.h" #include "ilo_shader.h" #include "ilo_state.h" -#include "ilo_state_gen.h" +#include "ilo_state_3d.h" #include "ilo_builder.h" static inline void diff --git a/src/gallium/drivers/ilo/ilo_shader.c b/src/gallium/drivers/ilo/ilo_shader.c index 626bc1b9c45..7177b8c50cc 100644 --- a/src/gallium/drivers/ilo/ilo_shader.c +++ b/src/gallium/drivers/ilo/ilo_shader.c @@ -32,7 +32,7 @@ #include "shader/ilo_shader_internal.h" #include "ilo_builder.h" #include "ilo_state.h" -#include "ilo_state_gen.h" +#include "ilo_state_3d.h" #include "ilo_shader.h" struct ilo_shader_cache { diff --git a/src/gallium/drivers/ilo/ilo_state.c b/src/gallium/drivers/ilo/ilo_state.c index c68062adb03..b852f9f6bc7 100644 --- a/src/gallium/drivers/ilo/ilo_state.c +++ b/src/gallium/drivers/ilo/ilo_state.c @@ -33,7 +33,7 @@ #include "ilo_resource.h" #include "ilo_shader.h" #include "ilo_state.h" -#include "ilo_state_gen.h" +#include "ilo_state_3d.h" static void finalize_shader_states(struct ilo_state_vector *vec) diff --git a/src/gallium/drivers/ilo/ilo_state_gen.h b/src/gallium/drivers/ilo/ilo_state_3d.h index 9e2f7c4a54b..c1e54c0c5b9 100644 --- a/src/gallium/drivers/ilo/ilo_state_gen.h +++ b/src/gallium/drivers/ilo/ilo_state_3d.h @@ -1,7 +1,7 @@ /* * Mesa 3-D graphics library * - * Copyright (C) 2012-2013 LunarG, Inc. + * Copyright (C) 2012-2014 LunarG, Inc. * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -25,8 +25,8 @@ * Chia-I Wu <olv@lunarg.com> */ -#ifndef ILO_STATE_GEN_H -#define ILO_STATE_GEN_H +#ifndef ILO_STATE_3D_H +#define ILO_STATE_3D_H #include "genhw/genhw.h" #include "intel_winsys.h" @@ -333,4 +333,4 @@ ilo_gpe_set_fb(const struct ilo_dev_info *dev, const struct pipe_framebuffer_state *state, struct ilo_fb_state *fb); -#endif /* ILO_STATE_GEN_H */ +#endif /* ILO_STATE_3D_H */ diff --git a/src/gallium/drivers/ilo/ilo_state_gen6.c b/src/gallium/drivers/ilo/ilo_state_3d_bottom.c index bc80aeb972d..d4b6643f9a8 100644 --- a/src/gallium/drivers/ilo/ilo_state_gen6.c +++ b/src/gallium/drivers/ilo/ilo_state_3d_bottom.c @@ -1,7 +1,7 @@ /* * Mesa 3-D graphics library * - * Copyright (C) 2012-2013 LunarG, Inc. + * Copyright (C) 2012-2014 LunarG, Inc. * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -35,531 +35,7 @@ #include "ilo_resource.h" #include "ilo_shader.h" #include "ilo_state.h" -#include "ilo_state_gen.h" - -/** - * Translate a pipe logicop to the matching hardware logicop. - */ -static int -gen6_translate_pipe_logicop(unsigned logicop) -{ - switch (logicop) { - case PIPE_LOGICOP_CLEAR: return GEN6_LOGICOP_CLEAR; - case PIPE_LOGICOP_NOR: return GEN6_LOGICOP_NOR; - case PIPE_LOGICOP_AND_INVERTED: return GEN6_LOGICOP_AND_INVERTED; - case PIPE_LOGICOP_COPY_INVERTED: return GEN6_LOGICOP_COPY_INVERTED; - case PIPE_LOGICOP_AND_REVERSE: return GEN6_LOGICOP_AND_REVERSE; - case PIPE_LOGICOP_INVERT: return GEN6_LOGICOP_INVERT; - case PIPE_LOGICOP_XOR: return GEN6_LOGICOP_XOR; - case PIPE_LOGICOP_NAND: return GEN6_LOGICOP_NAND; - case PIPE_LOGICOP_AND: return GEN6_LOGICOP_AND; - case PIPE_LOGICOP_EQUIV: return GEN6_LOGICOP_EQUIV; - case PIPE_LOGICOP_NOOP: return GEN6_LOGICOP_NOOP; - case PIPE_LOGICOP_OR_INVERTED: return GEN6_LOGICOP_OR_INVERTED; - case PIPE_LOGICOP_COPY: return GEN6_LOGICOP_COPY; - case PIPE_LOGICOP_OR_REVERSE: return GEN6_LOGICOP_OR_REVERSE; - case PIPE_LOGICOP_OR: return GEN6_LOGICOP_OR; - case PIPE_LOGICOP_SET: return GEN6_LOGICOP_SET; - default: - assert(!"unknown logicop function"); - return GEN6_LOGICOP_CLEAR; - } -} - -/** - * Translate a pipe blend function to the matching hardware blend function. - */ -static int -gen6_translate_pipe_blend(unsigned blend) -{ - switch (blend) { - case PIPE_BLEND_ADD: return GEN6_BLENDFUNCTION_ADD; - case PIPE_BLEND_SUBTRACT: return GEN6_BLENDFUNCTION_SUBTRACT; - case PIPE_BLEND_REVERSE_SUBTRACT: return GEN6_BLENDFUNCTION_REVERSE_SUBTRACT; - case PIPE_BLEND_MIN: return GEN6_BLENDFUNCTION_MIN; - case PIPE_BLEND_MAX: return GEN6_BLENDFUNCTION_MAX; - default: - assert(!"unknown blend function"); - return GEN6_BLENDFUNCTION_ADD; - }; -} - -/** - * Translate a pipe blend factor to the matching hardware blend factor. - */ -static int -gen6_translate_pipe_blendfactor(unsigned blendfactor) -{ - switch (blendfactor) { - case PIPE_BLENDFACTOR_ONE: return GEN6_BLENDFACTOR_ONE; - case PIPE_BLENDFACTOR_SRC_COLOR: return GEN6_BLENDFACTOR_SRC_COLOR; - case PIPE_BLENDFACTOR_SRC_ALPHA: return GEN6_BLENDFACTOR_SRC_ALPHA; - case PIPE_BLENDFACTOR_DST_ALPHA: return GEN6_BLENDFACTOR_DST_ALPHA; - case PIPE_BLENDFACTOR_DST_COLOR: return GEN6_BLENDFACTOR_DST_COLOR; - case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: return GEN6_BLENDFACTOR_SRC_ALPHA_SATURATE; - case PIPE_BLENDFACTOR_CONST_COLOR: return GEN6_BLENDFACTOR_CONST_COLOR; - case PIPE_BLENDFACTOR_CONST_ALPHA: return GEN6_BLENDFACTOR_CONST_ALPHA; - case PIPE_BLENDFACTOR_SRC1_COLOR: return GEN6_BLENDFACTOR_SRC1_COLOR; - case PIPE_BLENDFACTOR_SRC1_ALPHA: return GEN6_BLENDFACTOR_SRC1_ALPHA; - case PIPE_BLENDFACTOR_ZERO: return GEN6_BLENDFACTOR_ZERO; - case PIPE_BLENDFACTOR_INV_SRC_COLOR: return GEN6_BLENDFACTOR_INV_SRC_COLOR; - case PIPE_BLENDFACTOR_INV_SRC_ALPHA: return GEN6_BLENDFACTOR_INV_SRC_ALPHA; - case PIPE_BLENDFACTOR_INV_DST_ALPHA: return GEN6_BLENDFACTOR_INV_DST_ALPHA; - case PIPE_BLENDFACTOR_INV_DST_COLOR: return GEN6_BLENDFACTOR_INV_DST_COLOR; - case PIPE_BLENDFACTOR_INV_CONST_COLOR: return GEN6_BLENDFACTOR_INV_CONST_COLOR; - case PIPE_BLENDFACTOR_INV_CONST_ALPHA: return GEN6_BLENDFACTOR_INV_CONST_ALPHA; - case PIPE_BLENDFACTOR_INV_SRC1_COLOR: return GEN6_BLENDFACTOR_INV_SRC1_COLOR; - case PIPE_BLENDFACTOR_INV_SRC1_ALPHA: return GEN6_BLENDFACTOR_INV_SRC1_ALPHA; - default: - assert(!"unknown blend factor"); - return GEN6_BLENDFACTOR_ONE; - }; -} - -/** - * Translate a pipe stencil op to the matching hardware stencil op. - */ -static int -gen6_translate_pipe_stencil_op(unsigned stencil_op) -{ - switch (stencil_op) { - case PIPE_STENCIL_OP_KEEP: return GEN6_STENCILOP_KEEP; - case PIPE_STENCIL_OP_ZERO: return GEN6_STENCILOP_ZERO; - case PIPE_STENCIL_OP_REPLACE: return GEN6_STENCILOP_REPLACE; - case PIPE_STENCIL_OP_INCR: return GEN6_STENCILOP_INCRSAT; - case PIPE_STENCIL_OP_DECR: return GEN6_STENCILOP_DECRSAT; - case PIPE_STENCIL_OP_INCR_WRAP: return GEN6_STENCILOP_INCR; - case PIPE_STENCIL_OP_DECR_WRAP: return GEN6_STENCILOP_DECR; - case PIPE_STENCIL_OP_INVERT: return GEN6_STENCILOP_INVERT; - default: - assert(!"unknown stencil op"); - return GEN6_STENCILOP_KEEP; - } -} - -/** - * Translate a pipe texture mipfilter to the matching hardware mipfilter. - */ -static int -gen6_translate_tex_mipfilter(unsigned filter) -{ - switch (filter) { - case PIPE_TEX_MIPFILTER_NEAREST: return GEN6_MIPFILTER_NEAREST; - case PIPE_TEX_MIPFILTER_LINEAR: return GEN6_MIPFILTER_LINEAR; - case PIPE_TEX_MIPFILTER_NONE: return GEN6_MIPFILTER_NONE; - default: - assert(!"unknown mipfilter"); - return GEN6_MIPFILTER_NONE; - } -} - -/** - * Translate a pipe texture filter to the matching hardware mapfilter. - */ -static int -gen6_translate_tex_filter(unsigned filter) -{ - switch (filter) { - case PIPE_TEX_FILTER_NEAREST: return GEN6_MAPFILTER_NEAREST; - case PIPE_TEX_FILTER_LINEAR: return GEN6_MAPFILTER_LINEAR; - default: - assert(!"unknown sampler filter"); - return GEN6_MAPFILTER_NEAREST; - } -} - -/** - * Translate a pipe texture coordinate wrapping mode to the matching hardware - * wrapping mode. - */ -static int -gen6_translate_tex_wrap(unsigned wrap, bool clamp_to_edge) -{ - /* clamp to edge or border? */ - if (wrap == PIPE_TEX_WRAP_CLAMP) { - wrap = (clamp_to_edge) ? - PIPE_TEX_WRAP_CLAMP_TO_EDGE : PIPE_TEX_WRAP_CLAMP_TO_BORDER; - } - - switch (wrap) { - case PIPE_TEX_WRAP_REPEAT: return GEN6_TEXCOORDMODE_WRAP; - case PIPE_TEX_WRAP_CLAMP_TO_EDGE: return GEN6_TEXCOORDMODE_CLAMP; - case PIPE_TEX_WRAP_CLAMP_TO_BORDER: return GEN6_TEXCOORDMODE_CLAMP_BORDER; - case PIPE_TEX_WRAP_MIRROR_REPEAT: return GEN6_TEXCOORDMODE_MIRROR; - case PIPE_TEX_WRAP_CLAMP: - case PIPE_TEX_WRAP_MIRROR_CLAMP: - case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE: - case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER: - default: - assert(!"unknown sampler wrap mode"); - return GEN6_TEXCOORDMODE_WRAP; - } -} - -/** - * Translate a pipe shadow compare function to the matching hardware shadow - * function. - */ -static int -gen6_translate_shadow_func(unsigned func) -{ - /* - * For PIPE_FUNC_x, the reference value is on the left-hand side of the - * comparison, and 1.0 is returned when the comparison is true. - * - * For GEN6_COMPAREFUNCTION_x, the reference value is on the right-hand side of - * the comparison, and 0.0 is returned when the comparison is true. - */ - switch (func) { - case PIPE_FUNC_NEVER: return GEN6_COMPAREFUNCTION_ALWAYS; - case PIPE_FUNC_LESS: return GEN6_COMPAREFUNCTION_LEQUAL; - case PIPE_FUNC_EQUAL: return GEN6_COMPAREFUNCTION_NOTEQUAL; - case PIPE_FUNC_LEQUAL: return GEN6_COMPAREFUNCTION_LESS; - case PIPE_FUNC_GREATER: return GEN6_COMPAREFUNCTION_GEQUAL; - case PIPE_FUNC_NOTEQUAL: return GEN6_COMPAREFUNCTION_EQUAL; - case PIPE_FUNC_GEQUAL: return GEN6_COMPAREFUNCTION_GREATER; - case PIPE_FUNC_ALWAYS: return GEN6_COMPAREFUNCTION_NEVER; - default: - assert(!"unknown shadow compare function"); - return GEN6_COMPAREFUNCTION_NEVER; - } -} - -/** - * Translate a pipe DSA test function to the matching hardware compare - * function. - */ -static int -gen6_translate_dsa_func(unsigned func) -{ - switch (func) { - case PIPE_FUNC_NEVER: return GEN6_COMPAREFUNCTION_NEVER; - case PIPE_FUNC_LESS: return GEN6_COMPAREFUNCTION_LESS; - case PIPE_FUNC_EQUAL: return GEN6_COMPAREFUNCTION_EQUAL; - case PIPE_FUNC_LEQUAL: return GEN6_COMPAREFUNCTION_LEQUAL; - case PIPE_FUNC_GREATER: return GEN6_COMPAREFUNCTION_GREATER; - case PIPE_FUNC_NOTEQUAL: return GEN6_COMPAREFUNCTION_NOTEQUAL; - case PIPE_FUNC_GEQUAL: return GEN6_COMPAREFUNCTION_GEQUAL; - case PIPE_FUNC_ALWAYS: return GEN6_COMPAREFUNCTION_ALWAYS; - default: - assert(!"unknown depth/stencil/alpha test function"); - return GEN6_COMPAREFUNCTION_NEVER; - } -} - -static void -ve_init_cso(const struct ilo_dev_info *dev, - const struct pipe_vertex_element *state, - unsigned vb_index, - struct ilo_ve_cso *cso) -{ - int comp[4] = { - GEN6_VFCOMP_STORE_SRC, - GEN6_VFCOMP_STORE_SRC, - GEN6_VFCOMP_STORE_SRC, - GEN6_VFCOMP_STORE_SRC, - }; - int format; - - ILO_DEV_ASSERT(dev, 6, 7.5); - - switch (util_format_get_nr_components(state->src_format)) { - case 1: comp[1] = GEN6_VFCOMP_STORE_0; - case 2: comp[2] = GEN6_VFCOMP_STORE_0; - case 3: comp[3] = (util_format_is_pure_integer(state->src_format)) ? - GEN6_VFCOMP_STORE_1_INT : - GEN6_VFCOMP_STORE_1_FP; - } - - format = ilo_translate_vertex_format(dev, state->src_format); - - STATIC_ASSERT(Elements(cso->payload) >= 2); - cso->payload[0] = - vb_index << GEN6_VE_STATE_DW0_VB_INDEX__SHIFT | - GEN6_VE_STATE_DW0_VALID | - format << GEN6_VE_STATE_DW0_FORMAT__SHIFT | - state->src_offset << GEN6_VE_STATE_DW0_VB_OFFSET__SHIFT; - - cso->payload[1] = - comp[0] << GEN6_VE_STATE_DW1_COMP0__SHIFT | - comp[1] << GEN6_VE_STATE_DW1_COMP1__SHIFT | - comp[2] << GEN6_VE_STATE_DW1_COMP2__SHIFT | - comp[3] << GEN6_VE_STATE_DW1_COMP3__SHIFT; -} - -void -ilo_gpe_init_ve(const struct ilo_dev_info *dev, - unsigned num_states, - const struct pipe_vertex_element *states, - struct ilo_ve_state *ve) -{ - unsigned i; - - ILO_DEV_ASSERT(dev, 6, 7.5); - - ve->count = num_states; - ve->vb_count = 0; - - for (i = 0; i < num_states; i++) { - const unsigned pipe_idx = states[i].vertex_buffer_index; - const unsigned instance_divisor = states[i].instance_divisor; - unsigned hw_idx; - - /* - * map the pipe vb to the hardware vb, which has a fixed instance - * divisor - */ - for (hw_idx = 0; hw_idx < ve->vb_count; hw_idx++) { - if (ve->vb_mapping[hw_idx] == pipe_idx && - ve->instance_divisors[hw_idx] == instance_divisor) - break; - } - - /* create one if there is no matching hardware vb */ - if (hw_idx >= ve->vb_count) { - hw_idx = ve->vb_count++; - - ve->vb_mapping[hw_idx] = pipe_idx; - ve->instance_divisors[hw_idx] = instance_divisor; - } - - ve_init_cso(dev, &states[i], hw_idx, &ve->cso[i]); - } -} - -void -ilo_gpe_set_ve_edgeflag(const struct ilo_dev_info *dev, - struct ilo_ve_cso *cso) -{ - int format; - - ILO_DEV_ASSERT(dev, 6, 7.5); - - /* - * From the Sandy Bridge PRM, volume 2 part 1, page 94: - * - * "- This bit (Edge Flag Enable) must only be ENABLED on the last - * valid VERTEX_ELEMENT structure. - * - * - When set, Component 0 Control must be set to VFCOMP_STORE_SRC, - * and Component 1-3 Control must be set to VFCOMP_NOSTORE. - * - * - The Source Element Format must be set to the UINT format. - * - * - [DevSNB]: Edge Flags are not supported for QUADLIST - * primitives. Software may elect to convert QUADLIST primitives - * to some set of corresponding edge-flag-supported primitive - * types (e.g., POLYGONs) prior to submission to the 3D pipeline." - */ - cso->payload[0] |= GEN6_VE_STATE_DW0_EDGE_FLAG_ENABLE; - - /* - * Edge flags have format GEN6_FORMAT_R8_USCALED when defined via - * glEdgeFlagPointer(), and format GEN6_FORMAT_R32_FLOAT when defined - * via glEdgeFlag(), as can be seen in vbo_attrib_tmp.h. - * - * Since all the hardware cares about is whether the flags are zero or not, - * we can treat them as the corresponding _UINT formats. - */ - format = GEN_EXTRACT(cso->payload[0], GEN6_VE_STATE_DW0_FORMAT); - cso->payload[0] &= ~GEN6_VE_STATE_DW0_FORMAT__MASK; - - switch (format) { - case GEN6_FORMAT_R32_FLOAT: - format = GEN6_FORMAT_R32_UINT; - break; - case GEN6_FORMAT_R8_USCALED: - format = GEN6_FORMAT_R8_UINT; - break; - default: - break; - } - - cso->payload[0] |= GEN_SHIFT32(format, GEN6_VE_STATE_DW0_FORMAT); - - cso->payload[1] = - GEN6_VFCOMP_STORE_SRC << GEN6_VE_STATE_DW1_COMP0__SHIFT | - GEN6_VFCOMP_NOSTORE << GEN6_VE_STATE_DW1_COMP1__SHIFT | - GEN6_VFCOMP_NOSTORE << GEN6_VE_STATE_DW1_COMP2__SHIFT | - GEN6_VFCOMP_NOSTORE << GEN6_VE_STATE_DW1_COMP3__SHIFT; -} - -void -ilo_gpe_init_ve_nosrc(const struct ilo_dev_info *dev, - int comp0, int comp1, int comp2, int comp3, - struct ilo_ve_cso *cso) -{ - ILO_DEV_ASSERT(dev, 6, 7.5); - - STATIC_ASSERT(Elements(cso->payload) >= 2); - - assert(comp0 != GEN6_VFCOMP_STORE_SRC && - comp1 != GEN6_VFCOMP_STORE_SRC && - comp2 != GEN6_VFCOMP_STORE_SRC && - comp3 != GEN6_VFCOMP_STORE_SRC); - - cso->payload[0] = GEN6_VE_STATE_DW0_VALID; - cso->payload[1] = - comp0 << GEN6_VE_STATE_DW1_COMP0__SHIFT | - comp1 << GEN6_VE_STATE_DW1_COMP1__SHIFT | - comp2 << GEN6_VE_STATE_DW1_COMP2__SHIFT | - comp3 << GEN6_VE_STATE_DW1_COMP3__SHIFT; -} - -void -ilo_gpe_init_vs_cso(const struct ilo_dev_info *dev, - const struct ilo_shader_state *vs, - struct ilo_shader_cso *cso) -{ - int start_grf, vue_read_len, sampler_count, max_threads; - uint32_t dw2, dw4, dw5; - - ILO_DEV_ASSERT(dev, 6, 7.5); - - start_grf = ilo_shader_get_kernel_param(vs, ILO_KERNEL_URB_DATA_START_REG); - vue_read_len = ilo_shader_get_kernel_param(vs, ILO_KERNEL_INPUT_COUNT); - sampler_count = ilo_shader_get_kernel_param(vs, ILO_KERNEL_SAMPLER_COUNT); - - /* - * From the Sandy Bridge PRM, volume 2 part 1, page 135: - * - * "(Vertex URB Entry Read Length) Specifies the number of pairs of - * 128-bit vertex elements to be passed into the payload for each - * vertex." - * - * "It is UNDEFINED to set this field to 0 indicating no Vertex URB - * data to be read and passed to the thread." - */ - vue_read_len = (vue_read_len + 1) / 2; - if (!vue_read_len) - vue_read_len = 1; - - max_threads = dev->thread_count; - if (ilo_dev_gen(dev) == ILO_GEN(7.5) && dev->gt == 2) - max_threads *= 2; - - dw2 = (true) ? 0 : GEN6_THREADDISP_FP_MODE_ALT; - dw2 |= ((sampler_count + 3) / 4) << GEN6_THREADDISP_SAMPLER_COUNT__SHIFT; - - dw4 = start_grf << GEN6_VS_DW4_URB_GRF_START__SHIFT | - vue_read_len << GEN6_VS_DW4_URB_READ_LEN__SHIFT | - 0 << GEN6_VS_DW4_URB_READ_OFFSET__SHIFT; - - dw5 = GEN6_VS_DW5_STATISTICS | - GEN6_VS_DW5_VS_ENABLE; - - if (ilo_dev_gen(dev) >= ILO_GEN(7.5)) - dw5 |= (max_threads - 1) << GEN75_VS_DW5_MAX_THREADS__SHIFT; - else - dw5 |= (max_threads - 1) << GEN6_VS_DW5_MAX_THREADS__SHIFT; - - STATIC_ASSERT(Elements(cso->payload) >= 3); - cso->payload[0] = dw2; - cso->payload[1] = dw4; - cso->payload[2] = dw5; -} - -void -ilo_gpe_init_gs_cso_gen6(const struct ilo_dev_info *dev, - const struct ilo_shader_state *gs, - struct ilo_shader_cso *cso) -{ - int start_grf, vue_read_len, max_threads; - uint32_t dw2, dw4, dw5, dw6; - - ILO_DEV_ASSERT(dev, 6, 6); - - if (ilo_shader_get_type(gs) == PIPE_SHADER_GEOMETRY) { - start_grf = ilo_shader_get_kernel_param(gs, - ILO_KERNEL_URB_DATA_START_REG); - - vue_read_len = ilo_shader_get_kernel_param(gs, ILO_KERNEL_INPUT_COUNT); - } - else { - start_grf = ilo_shader_get_kernel_param(gs, - ILO_KERNEL_VS_GEN6_SO_START_REG); - - vue_read_len = ilo_shader_get_kernel_param(gs, ILO_KERNEL_OUTPUT_COUNT); - } - - /* - * From the Sandy Bridge PRM, volume 2 part 1, page 153: - * - * "Specifies the amount of URB data read and passed in the thread - * payload for each Vertex URB entry, in 256-bit register increments. - * - * It is UNDEFINED to set this field (Vertex URB Entry Read Length) to - * 0 indicating no Vertex URB data to be read and passed to the - * thread." - */ - vue_read_len = (vue_read_len + 1) / 2; - if (!vue_read_len) - vue_read_len = 1; - - /* - * From the Sandy Bridge PRM, volume 2 part 1, page 154: - * - * "Maximum Number of Threads valid range is [0,27] when Rendering - * Enabled bit is set." - * - * From the Sandy Bridge PRM, volume 2 part 1, page 173: - * - * "Programming Note: If the GS stage is enabled, software must always - * allocate at least one GS URB Entry. This is true even if the GS - * thread never needs to output vertices to the pipeline, e.g., when - * only performing stream output. This is an artifact of the need to - * pass the GS thread an initial destination URB handle." - * - * As such, we always enable rendering, and limit the number of threads. - */ - if (dev->gt == 2) { - /* maximum is 60, but limited to 28 */ - max_threads = 28; - } - else { - /* maximum is 24, but limited to 21 (see brwCreateContext()) */ - max_threads = 21; - } - - dw2 = GEN6_THREADDISP_SPF; - - dw4 = vue_read_len << GEN6_GS_DW4_URB_READ_LEN__SHIFT | - 0 << GEN6_GS_DW4_URB_READ_OFFSET__SHIFT | - start_grf << GEN6_GS_DW4_URB_GRF_START__SHIFT; - - dw5 = (max_threads - 1) << GEN6_GS_DW5_MAX_THREADS__SHIFT | - GEN6_GS_DW5_STATISTICS | - GEN6_GS_DW5_SO_STATISTICS | - GEN6_GS_DW5_RENDER_ENABLE; - - /* - * we cannot make use of GEN6_GS_REORDER because it will reorder - * triangle strips according to D3D rules (triangle 2N+1 uses vertices - * (2N+1, 2N+3, 2N+2)), instead of GL rules (triangle 2N+1 uses vertices - * (2N+2, 2N+1, 2N+3)). - */ - dw6 = GEN6_GS_DW6_GS_ENABLE; - - if (ilo_shader_get_kernel_param(gs, ILO_KERNEL_GS_DISCARD_ADJACENCY)) - dw6 |= GEN6_GS_DW6_DISCARD_ADJACENCY; - - if (ilo_shader_get_kernel_param(gs, ILO_KERNEL_VS_GEN6_SO)) { - const uint32_t svbi_post_inc = - ilo_shader_get_kernel_param(gs, ILO_KERNEL_GS_GEN6_SVBI_POST_INC); - - dw6 |= GEN6_GS_DW6_SVBI_PAYLOAD_ENABLE; - if (svbi_post_inc) { - dw6 |= GEN6_GS_DW6_SVBI_POST_INC_ENABLE | - svbi_post_inc << GEN6_GS_DW6_SVBI_POST_INC_VAL__SHIFT; - } - } - - STATIC_ASSERT(Elements(cso->payload) >= 4); - cso->payload[0] = dw2; - cso->payload[1] = dw4; - cso->payload[2] = dw5; - cso->payload[3] = dw6; -} +#include "ilo_state_3d.h" void ilo_gpe_init_rasterizer_clip(const struct ilo_dev_info *dev, @@ -905,6 +381,53 @@ ilo_gpe_init_rasterizer_wm_gen6(const struct ilo_dev_info *dev, } void +ilo_gpe_init_rasterizer_wm_gen7(const struct ilo_dev_info *dev, + const struct pipe_rasterizer_state *state, + struct ilo_rasterizer_wm *wm) +{ + uint32_t dw1, dw2; + + ILO_DEV_ASSERT(dev, 7, 7.5); + + dw1 = GEN7_WM_DW1_ZW_INTERP_PIXEL | + GEN7_WM_DW1_AA_LINE_WIDTH_2_0 | + GEN7_WM_DW1_MSRASTMODE_OFF_PIXEL; + + /* same value as in 3DSTATE_SF */ + if (state->line_smooth) + dw1 |= GEN7_WM_DW1_AA_LINE_CAP_1_0; + + if (state->poly_stipple_enable) + dw1 |= GEN7_WM_DW1_POLY_STIPPLE_ENABLE; + if (state->line_stipple_enable) + dw1 |= GEN7_WM_DW1_LINE_STIPPLE_ENABLE; + + if (state->bottom_edge_rule) + dw1 |= GEN7_WM_DW1_POINT_RASTRULE_UPPER_RIGHT; + + dw2 = GEN7_WM_DW2_MSDISPMODE_PERSAMPLE; + + /* + * assertion that makes sure + * + * dw1 |= wm->dw_msaa_rast; + * dw2 |= wm->dw_msaa_disp; + * + * is valid + */ + STATIC_ASSERT(GEN7_WM_DW1_MSRASTMODE_OFF_PIXEL == 0 && + GEN7_WM_DW2_MSDISPMODE_PERSAMPLE == 0); + + wm->dw_msaa_rast = + (state->multisample) ? GEN7_WM_DW1_MSRASTMODE_ON_PATTERN : 0; + wm->dw_msaa_disp = GEN7_WM_DW2_MSDISPMODE_PERPIXEL; + + STATIC_ASSERT(Elements(wm->payload) >= 2); + wm->payload[0] = dw1; + wm->payload[1] = dw2; +} + +void ilo_gpe_init_fs_cso_gen6(const struct ilo_dev_info *dev, const struct ilo_shader_state *fs, struct ilo_shader_cso *cso) @@ -1001,6 +524,114 @@ ilo_gpe_init_fs_cso_gen6(const struct ilo_dev_info *dev, cso->payload[3] = dw6; } +void +ilo_gpe_init_fs_cso_gen7(const struct ilo_dev_info *dev, + const struct ilo_shader_state *fs, + struct ilo_shader_cso *cso) +{ + int start_grf, sampler_count, max_threads; + uint32_t dw2, dw4, dw5; + uint32_t wm_interps, wm_dw1; + + ILO_DEV_ASSERT(dev, 7, 7.5); + + start_grf = ilo_shader_get_kernel_param(fs, ILO_KERNEL_URB_DATA_START_REG); + sampler_count = ilo_shader_get_kernel_param(fs, ILO_KERNEL_SAMPLER_COUNT); + + dw2 = (true) ? 0 : GEN6_THREADDISP_FP_MODE_ALT; + dw2 |= ((sampler_count + 3) / 4) << GEN6_THREADDISP_SAMPLER_COUNT__SHIFT; + + dw4 = GEN7_PS_DW4_POSOFFSET_NONE; + + /* see brwCreateContext() */ + switch (ilo_dev_gen(dev)) { + case ILO_GEN(7.5): + max_threads = (dev->gt == 3) ? 408 : (dev->gt == 2) ? 204 : 102; + dw4 |= (max_threads - 1) << GEN75_PS_DW4_MAX_THREADS__SHIFT; + dw4 |= 1 << GEN75_PS_DW4_SAMPLE_MASK__SHIFT; + break; + case ILO_GEN(7): + default: + max_threads = (dev->gt == 2) ? 172 : 48; + dw4 |= (max_threads - 1) << GEN7_PS_DW4_MAX_THREADS__SHIFT; + break; + } + + if (ilo_shader_get_kernel_param(fs, ILO_KERNEL_PCB_CBUF0_SIZE)) + dw4 |= GEN7_PS_DW4_PUSH_CONSTANT_ENABLE; + + if (ilo_shader_get_kernel_param(fs, ILO_KERNEL_INPUT_COUNT)) + dw4 |= GEN7_PS_DW4_ATTR_ENABLE; + + assert(!ilo_shader_get_kernel_param(fs, ILO_KERNEL_FS_DISPATCH_16_OFFSET)); + dw4 |= GEN7_PS_DW4_8_PIXEL_DISPATCH; + + dw5 = start_grf << GEN7_PS_DW5_URB_GRF_START0__SHIFT | + 0 << GEN7_PS_DW5_URB_GRF_START1__SHIFT | + 0 << GEN7_PS_DW5_URB_GRF_START2__SHIFT; + + /* FS affects 3DSTATE_WM too */ + wm_dw1 = 0; + + /* + * TODO set this bit only when + * + * a) fs writes colors and color is not masked, or + * b) fs writes depth, or + * c) fs or cc kills + */ + wm_dw1 |= GEN7_WM_DW1_PS_ENABLE; + + /* + * From the Ivy Bridge PRM, volume 2 part 1, page 278: + * + * "This bit (Pixel Shader Kill Pixel), if ENABLED, indicates that + * the PS kernel or color calculator has the ability to kill + * (discard) pixels or samples, other than due to depth or stencil + * testing. This bit is required to be ENABLED in the following + * situations: + * + * - The API pixel shader program contains "killpix" or "discard" + * instructions, or other code in the pixel shader kernel that + * can cause the final pixel mask to differ from the pixel mask + * received on dispatch. + * + * - A sampler with chroma key enabled with kill pixel mode is used + * by the pixel shader. + * + * - Any render target has Alpha Test Enable or AlphaToCoverage + * Enable enabled. + * + * - The pixel shader kernel generates and outputs oMask. + * + * Note: As ClipDistance clipping is fully supported in hardware + * and therefore not via PS instructions, there should be no need + * to ENABLE this bit due to ClipDistance clipping." + */ + if (ilo_shader_get_kernel_param(fs, ILO_KERNEL_FS_USE_KILL)) + wm_dw1 |= GEN7_WM_DW1_PS_KILL; + + if (ilo_shader_get_kernel_param(fs, ILO_KERNEL_FS_OUTPUT_Z)) + wm_dw1 |= GEN7_WM_DW1_PSCDEPTH_ON; + + if (ilo_shader_get_kernel_param(fs, ILO_KERNEL_FS_INPUT_Z)) + wm_dw1 |= GEN7_WM_DW1_PS_USE_DEPTH; + + if (ilo_shader_get_kernel_param(fs, ILO_KERNEL_FS_INPUT_W)) + wm_dw1 |= GEN7_WM_DW1_PS_USE_W; + + wm_interps = ilo_shader_get_kernel_param(fs, + ILO_KERNEL_FS_BARYCENTRIC_INTERPOLATIONS); + + wm_dw1 |= wm_interps << GEN7_WM_DW1_BARYCENTRIC_INTERP__SHIFT; + + STATIC_ASSERT(Elements(cso->payload) >= 4); + cso->payload[0] = dw2; + cso->payload[1] = dw4; + cso->payload[2] = dw5; + cso->payload[3] = wm_dw1; +} + struct ilo_zs_surface_info { int surface_type; int format; @@ -1474,6 +1105,106 @@ ilo_gpe_set_viewport_cso(const struct ilo_dev_info *dev, vp->max_z = scale_z * 1.0f + state->translate[2]; } +/** + * Translate a pipe logicop to the matching hardware logicop. + */ +static int +gen6_translate_pipe_logicop(unsigned logicop) +{ + switch (logicop) { + case PIPE_LOGICOP_CLEAR: return GEN6_LOGICOP_CLEAR; + case PIPE_LOGICOP_NOR: return GEN6_LOGICOP_NOR; + case PIPE_LOGICOP_AND_INVERTED: return GEN6_LOGICOP_AND_INVERTED; + case PIPE_LOGICOP_COPY_INVERTED: return GEN6_LOGICOP_COPY_INVERTED; + case PIPE_LOGICOP_AND_REVERSE: return GEN6_LOGICOP_AND_REVERSE; + case PIPE_LOGICOP_INVERT: return GEN6_LOGICOP_INVERT; + case PIPE_LOGICOP_XOR: return GEN6_LOGICOP_XOR; + case PIPE_LOGICOP_NAND: return GEN6_LOGICOP_NAND; + case PIPE_LOGICOP_AND: return GEN6_LOGICOP_AND; + case PIPE_LOGICOP_EQUIV: return GEN6_LOGICOP_EQUIV; + case PIPE_LOGICOP_NOOP: return GEN6_LOGICOP_NOOP; + case PIPE_LOGICOP_OR_INVERTED: return GEN6_LOGICOP_OR_INVERTED; + case PIPE_LOGICOP_COPY: return GEN6_LOGICOP_COPY; + case PIPE_LOGICOP_OR_REVERSE: return GEN6_LOGICOP_OR_REVERSE; + case PIPE_LOGICOP_OR: return GEN6_LOGICOP_OR; + case PIPE_LOGICOP_SET: return GEN6_LOGICOP_SET; + default: + assert(!"unknown logicop function"); + return GEN6_LOGICOP_CLEAR; + } +} + +/** + * Translate a pipe blend function to the matching hardware blend function. + */ +static int +gen6_translate_pipe_blend(unsigned blend) +{ + switch (blend) { + case PIPE_BLEND_ADD: return GEN6_BLENDFUNCTION_ADD; + case PIPE_BLEND_SUBTRACT: return GEN6_BLENDFUNCTION_SUBTRACT; + case PIPE_BLEND_REVERSE_SUBTRACT: return GEN6_BLENDFUNCTION_REVERSE_SUBTRACT; + case PIPE_BLEND_MIN: return GEN6_BLENDFUNCTION_MIN; + case PIPE_BLEND_MAX: return GEN6_BLENDFUNCTION_MAX; + default: + assert(!"unknown blend function"); + return GEN6_BLENDFUNCTION_ADD; + }; +} + +/** + * Translate a pipe blend factor to the matching hardware blend factor. + */ +static int +gen6_translate_pipe_blendfactor(unsigned blendfactor) +{ + switch (blendfactor) { + case PIPE_BLENDFACTOR_ONE: return GEN6_BLENDFACTOR_ONE; + case PIPE_BLENDFACTOR_SRC_COLOR: return GEN6_BLENDFACTOR_SRC_COLOR; + case PIPE_BLENDFACTOR_SRC_ALPHA: return GEN6_BLENDFACTOR_SRC_ALPHA; + case PIPE_BLENDFACTOR_DST_ALPHA: return GEN6_BLENDFACTOR_DST_ALPHA; + case PIPE_BLENDFACTOR_DST_COLOR: return GEN6_BLENDFACTOR_DST_COLOR; + case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: return GEN6_BLENDFACTOR_SRC_ALPHA_SATURATE; + case PIPE_BLENDFACTOR_CONST_COLOR: return GEN6_BLENDFACTOR_CONST_COLOR; + case PIPE_BLENDFACTOR_CONST_ALPHA: return GEN6_BLENDFACTOR_CONST_ALPHA; + case PIPE_BLENDFACTOR_SRC1_COLOR: return GEN6_BLENDFACTOR_SRC1_COLOR; + case PIPE_BLENDFACTOR_SRC1_ALPHA: return GEN6_BLENDFACTOR_SRC1_ALPHA; + case PIPE_BLENDFACTOR_ZERO: return GEN6_BLENDFACTOR_ZERO; + case PIPE_BLENDFACTOR_INV_SRC_COLOR: return GEN6_BLENDFACTOR_INV_SRC_COLOR; + case PIPE_BLENDFACTOR_INV_SRC_ALPHA: return GEN6_BLENDFACTOR_INV_SRC_ALPHA; + case PIPE_BLENDFACTOR_INV_DST_ALPHA: return GEN6_BLENDFACTOR_INV_DST_ALPHA; + case PIPE_BLENDFACTOR_INV_DST_COLOR: return GEN6_BLENDFACTOR_INV_DST_COLOR; + case PIPE_BLENDFACTOR_INV_CONST_COLOR: return GEN6_BLENDFACTOR_INV_CONST_COLOR; + case PIPE_BLENDFACTOR_INV_CONST_ALPHA: return GEN6_BLENDFACTOR_INV_CONST_ALPHA; + case PIPE_BLENDFACTOR_INV_SRC1_COLOR: return GEN6_BLENDFACTOR_INV_SRC1_COLOR; + case PIPE_BLENDFACTOR_INV_SRC1_ALPHA: return GEN6_BLENDFACTOR_INV_SRC1_ALPHA; + default: + assert(!"unknown blend factor"); + return GEN6_BLENDFACTOR_ONE; + }; +} + +/** + * Translate a pipe stencil op to the matching hardware stencil op. + */ +static int +gen6_translate_pipe_stencil_op(unsigned stencil_op) +{ + switch (stencil_op) { + case PIPE_STENCIL_OP_KEEP: return GEN6_STENCILOP_KEEP; + case PIPE_STENCIL_OP_ZERO: return GEN6_STENCILOP_ZERO; + case PIPE_STENCIL_OP_REPLACE: return GEN6_STENCILOP_REPLACE; + case PIPE_STENCIL_OP_INCR: return GEN6_STENCILOP_INCRSAT; + case PIPE_STENCIL_OP_DECR: return GEN6_STENCILOP_DECRSAT; + case PIPE_STENCIL_OP_INCR_WRAP: return GEN6_STENCILOP_INCR; + case PIPE_STENCIL_OP_DECR_WRAP: return GEN6_STENCILOP_DECR; + case PIPE_STENCIL_OP_INVERT: return GEN6_STENCILOP_INVERT; + default: + assert(!"unknown stencil op"); + return GEN6_STENCILOP_KEEP; + } +} + static int gen6_blend_factor_dst_alpha_forced_one(int factor) { @@ -1620,6 +1351,28 @@ ilo_gpe_init_blend(const struct ilo_dev_info *dev, } } +/** + * Translate a pipe DSA test function to the matching hardware compare + * function. + */ +static int +gen6_translate_dsa_func(unsigned func) +{ + switch (func) { + case PIPE_FUNC_NEVER: return GEN6_COMPAREFUNCTION_NEVER; + case PIPE_FUNC_LESS: return GEN6_COMPAREFUNCTION_LESS; + case PIPE_FUNC_EQUAL: return GEN6_COMPAREFUNCTION_EQUAL; + case PIPE_FUNC_LEQUAL: return GEN6_COMPAREFUNCTION_LEQUAL; + case PIPE_FUNC_GREATER: return GEN6_COMPAREFUNCTION_GREATER; + case PIPE_FUNC_NOTEQUAL: return GEN6_COMPAREFUNCTION_NOTEQUAL; + case PIPE_FUNC_GEQUAL: return GEN6_COMPAREFUNCTION_GEQUAL; + case PIPE_FUNC_ALWAYS: return GEN6_COMPAREFUNCTION_ALWAYS; + default: + assert(!"unknown depth/stencil/alpha test function"); + return GEN6_COMPAREFUNCTION_NEVER; + } +} + void ilo_gpe_init_dsa(const struct ilo_dev_info *dev, const struct pipe_depth_stencil_alpha_state *state, @@ -1763,680 +1516,6 @@ ilo_gpe_set_scissor_null(const struct ilo_dev_info *dev, } void -ilo_gpe_init_view_surface_null_gen6(const struct ilo_dev_info *dev, - unsigned width, unsigned height, - unsigned depth, unsigned level, - struct ilo_view_surface *surf) -{ - uint32_t *dw; - - ILO_DEV_ASSERT(dev, 6, 6); - - assert(width >= 1 && height >= 1 && depth >= 1); - - /* - * From the Sandy Bridge PRM, volume 4 part 1, page 71: - * - * "A null surface will be used in instances where an actual surface is - * not bound. When a write message is generated to a null surface, no - * actual surface is written to. When a read message (including any - * sampling engine message) is generated to a null surface, the result - * is all zeros. Note that a null surface type is allowed to be used - * with all messages, even if it is not specificially indicated as - * supported. All of the remaining fields in surface state are ignored - * for null surfaces, with the following exceptions: - * - * * [DevSNB+]: Width, Height, Depth, and LOD fields must match the - * depth buffer's corresponding state for all render target - * surfaces, including null. - * * Surface Format must be R8G8B8A8_UNORM." - * - * From the Sandy Bridge PRM, volume 4 part 1, page 82: - * - * "If Surface Type is SURFTYPE_NULL, this field (Tiled Surface) must be - * true" - */ - - STATIC_ASSERT(Elements(surf->payload) >= 6); - dw = surf->payload; - - dw[0] = GEN6_SURFTYPE_NULL << GEN6_SURFACE_DW0_TYPE__SHIFT | - GEN6_FORMAT_B8G8R8A8_UNORM << GEN6_SURFACE_DW0_FORMAT__SHIFT; - - dw[1] = 0; - - dw[2] = (height - 1) << GEN6_SURFACE_DW2_HEIGHT__SHIFT | - (width - 1) << GEN6_SURFACE_DW2_WIDTH__SHIFT | - level << GEN6_SURFACE_DW2_MIP_COUNT_LOD__SHIFT; - - dw[3] = (depth - 1) << GEN6_SURFACE_DW3_DEPTH__SHIFT | - GEN6_TILING_X; - - dw[4] = 0; - dw[5] = 0; - - surf->bo = NULL; -} - -void -ilo_gpe_init_view_surface_for_buffer_gen6(const struct ilo_dev_info *dev, - const struct ilo_buffer *buf, - unsigned offset, unsigned size, - unsigned struct_size, - enum pipe_format elem_format, - bool is_rt, bool render_cache_rw, - struct ilo_view_surface *surf) -{ - const int elem_size = util_format_get_blocksize(elem_format); - int width, height, depth, pitch; - int surface_format, num_entries; - uint32_t *dw; - - ILO_DEV_ASSERT(dev, 6, 6); - - /* - * For SURFTYPE_BUFFER, a SURFACE_STATE specifies an element of a - * structure in a buffer. - */ - - surface_format = ilo_translate_color_format(dev, elem_format); - - num_entries = size / struct_size; - /* see if there is enough space to fit another element */ - if (size % struct_size >= elem_size) - num_entries++; - - /* - * From the Sandy Bridge PRM, volume 4 part 1, page 76: - * - * "For SURFTYPE_BUFFER render targets, this field (Surface Base - * Address) specifies the base address of first element of the - * surface. The surface is interpreted as a simple array of that - * single element type. The address must be naturally-aligned to the - * element size (e.g., a buffer containing R32G32B32A32_FLOAT elements - * must be 16-byte aligned). - * - * For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies - * the base address of the first element of the surface, computed in - * software by adding the surface base address to the byte offset of - * the element in the buffer." - */ - if (is_rt) - assert(offset % elem_size == 0); - - /* - * From the Sandy Bridge PRM, volume 4 part 1, page 77: - * - * "For buffer surfaces, the number of entries in the buffer ranges - * from 1 to 2^27." - */ - assert(num_entries >= 1 && num_entries <= 1 << 27); - - /* - * From the Sandy Bridge PRM, volume 4 part 1, page 81: - * - * "For surfaces of type SURFTYPE_BUFFER, this field (Surface Pitch) - * indicates the size of the structure." - */ - pitch = struct_size; - - pitch--; - num_entries--; - /* bits [6:0] */ - width = (num_entries & 0x0000007f); - /* bits [19:7] */ - height = (num_entries & 0x000fff80) >> 7; - /* bits [26:20] */ - depth = (num_entries & 0x07f00000) >> 20; - - STATIC_ASSERT(Elements(surf->payload) >= 6); - dw = surf->payload; - - dw[0] = GEN6_SURFTYPE_BUFFER << GEN6_SURFACE_DW0_TYPE__SHIFT | - surface_format << GEN6_SURFACE_DW0_FORMAT__SHIFT; - if (render_cache_rw) - dw[0] |= GEN6_SURFACE_DW0_RENDER_CACHE_RW; - - dw[1] = offset; - - dw[2] = height << GEN6_SURFACE_DW2_HEIGHT__SHIFT | - width << GEN6_SURFACE_DW2_WIDTH__SHIFT; - - dw[3] = depth << GEN6_SURFACE_DW3_DEPTH__SHIFT | - pitch << GEN6_SURFACE_DW3_PITCH__SHIFT; - - dw[4] = 0; - dw[5] = 0; - - /* do not increment reference count */ - surf->bo = buf->bo; -} - -void -ilo_gpe_init_view_surface_for_texture_gen6(const struct ilo_dev_info *dev, - const struct ilo_texture *tex, - enum pipe_format format, - unsigned first_level, - unsigned num_levels, - unsigned first_layer, - unsigned num_layers, - bool is_rt, - struct ilo_view_surface *surf) -{ - int surface_type, surface_format; - int width, height, depth, pitch, lod; - uint32_t *dw; - - ILO_DEV_ASSERT(dev, 6, 6); - - surface_type = ilo_gpe_gen6_translate_texture(tex->base.target); - assert(surface_type != GEN6_SURFTYPE_BUFFER); - - if (format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT && tex->separate_s8) - format = PIPE_FORMAT_Z32_FLOAT; - - if (is_rt) - surface_format = ilo_translate_render_format(dev, format); - else - surface_format = ilo_translate_texture_format(dev, format); - assert(surface_format >= 0); - - width = tex->layout.width0; - height = tex->layout.height0; - depth = (tex->base.target == PIPE_TEXTURE_3D) ? - tex->base.depth0 : num_layers; - pitch = tex->layout.bo_stride; - - if (surface_type == GEN6_SURFTYPE_CUBE) { - /* - * From the Sandy Bridge PRM, volume 4 part 1, page 81: - * - * "For SURFTYPE_CUBE: [DevSNB+]: for Sampling Engine Surfaces, the - * range of this field (Depth) is [0,84], indicating the number of - * cube array elements (equal to the number of underlying 2D array - * elements divided by 6). For other surfaces, this field must be - * zero." - * - * When is_rt is true, we treat the texture as a 2D one to avoid the - * restriction. - */ - if (is_rt) { - surface_type = GEN6_SURFTYPE_2D; - } - else { - assert(num_layers % 6 == 0); - depth = num_layers / 6; - } - } - - /* sanity check the size */ - assert(width >= 1 && height >= 1 && depth >= 1 && pitch >= 1); - switch (surface_type) { - case GEN6_SURFTYPE_1D: - assert(width <= 8192 && height == 1 && depth <= 512); - assert(first_layer < 512 && num_layers <= 512); - break; - case GEN6_SURFTYPE_2D: - assert(width <= 8192 && height <= 8192 && depth <= 512); - assert(first_layer < 512 && num_layers <= 512); - break; - case GEN6_SURFTYPE_3D: - assert(width <= 2048 && height <= 2048 && depth <= 2048); - assert(first_layer < 2048 && num_layers <= 512); - if (!is_rt) - assert(first_layer == 0); - break; - case GEN6_SURFTYPE_CUBE: - assert(width <= 8192 && height <= 8192 && depth <= 85); - assert(width == height); - assert(first_layer < 512 && num_layers <= 512); - if (is_rt) - assert(first_layer == 0); - break; - default: - assert(!"unexpected surface type"); - break; - } - - /* non-full array spacing is supported only on GEN7+ */ - assert(tex->layout.walk != ILO_LAYOUT_WALK_LOD); - /* non-interleaved samples are supported only on GEN7+ */ - if (tex->base.nr_samples > 1) - assert(tex->layout.interleaved_samples); - - if (is_rt) { - assert(num_levels == 1); - lod = first_level; - } - else { - lod = num_levels - 1; - } - - /* - * From the Sandy Bridge PRM, volume 4 part 1, page 76: - * - * "Linear render target surface base addresses must be element-size - * aligned, for non-YUV surface formats, or a multiple of 2 - * element-sizes for YUV surface formats. Other linear surfaces have - * no alignment requirements (byte alignment is sufficient.)" - * - * From the Sandy Bridge PRM, volume 4 part 1, page 81: - * - * "For linear render target surfaces, the pitch must be a multiple - * of the element size for non-YUV surface formats. Pitch must be a - * multiple of 2 * element size for YUV surface formats." - * - * From the Sandy Bridge PRM, volume 4 part 1, page 86: - * - * "For linear surfaces, this field (X Offset) must be zero" - */ - if (tex->layout.tiling == INTEL_TILING_NONE) { - if (is_rt) { - const int elem_size = util_format_get_blocksize(format); - assert(pitch % elem_size == 0); - } - } - - STATIC_ASSERT(Elements(surf->payload) >= 6); - dw = surf->payload; - - dw[0] = surface_type << GEN6_SURFACE_DW0_TYPE__SHIFT | - surface_format << GEN6_SURFACE_DW0_FORMAT__SHIFT | - GEN6_SURFACE_DW0_MIPLAYOUT_BELOW; - - if (surface_type == GEN6_SURFTYPE_CUBE && !is_rt) { - dw[0] |= 1 << 9 | - GEN6_SURFACE_DW0_CUBE_FACE_ENABLES__MASK; - } - - if (is_rt) - dw[0] |= GEN6_SURFACE_DW0_RENDER_CACHE_RW; - - dw[1] = 0; - - dw[2] = (height - 1) << GEN6_SURFACE_DW2_HEIGHT__SHIFT | - (width - 1) << GEN6_SURFACE_DW2_WIDTH__SHIFT | - lod << GEN6_SURFACE_DW2_MIP_COUNT_LOD__SHIFT; - - dw[3] = (depth - 1) << GEN6_SURFACE_DW3_DEPTH__SHIFT | - (pitch - 1) << GEN6_SURFACE_DW3_PITCH__SHIFT | - ilo_gpe_gen6_translate_winsys_tiling(tex->layout.tiling); - - dw[4] = first_level << GEN6_SURFACE_DW4_MIN_LOD__SHIFT | - first_layer << 17 | - (num_layers - 1) << 8 | - ((tex->base.nr_samples > 1) ? GEN6_SURFACE_DW4_MULTISAMPLECOUNT_4 : - GEN6_SURFACE_DW4_MULTISAMPLECOUNT_1); - - dw[5] = 0; - - assert(tex->layout.align_j == 2 || tex->layout.align_j == 4); - if (tex->layout.align_j == 4) - dw[5] |= GEN6_SURFACE_DW5_VALIGN_4; - - /* do not increment reference count */ - surf->bo = tex->bo; -} - -static void -sampler_init_border_color_gen6(const struct ilo_dev_info *dev, - const union pipe_color_union *color, - uint32_t *dw, int num_dwords) -{ - float rgba[4] = { - color->f[0], color->f[1], color->f[2], color->f[3], - }; - - ILO_DEV_ASSERT(dev, 6, 6); - - assert(num_dwords >= 12); - - /* - * This state is not documented in the Sandy Bridge PRM, but in the - * Ironlake PRM. SNORM8 seems to be in DW11 instead of DW1. - */ - - /* IEEE_FP */ - dw[1] = fui(rgba[0]); - dw[2] = fui(rgba[1]); - dw[3] = fui(rgba[2]); - dw[4] = fui(rgba[3]); - - /* FLOAT_16 */ - dw[5] = util_float_to_half(rgba[0]) | - util_float_to_half(rgba[1]) << 16; - dw[6] = util_float_to_half(rgba[2]) | - util_float_to_half(rgba[3]) << 16; - - /* clamp to [-1.0f, 1.0f] */ - rgba[0] = CLAMP(rgba[0], -1.0f, 1.0f); - rgba[1] = CLAMP(rgba[1], -1.0f, 1.0f); - rgba[2] = CLAMP(rgba[2], -1.0f, 1.0f); - rgba[3] = CLAMP(rgba[3], -1.0f, 1.0f); - - /* SNORM16 */ - dw[9] = (int16_t) util_iround(rgba[0] * 32767.0f) | - (int16_t) util_iround(rgba[1] * 32767.0f) << 16; - dw[10] = (int16_t) util_iround(rgba[2] * 32767.0f) | - (int16_t) util_iround(rgba[3] * 32767.0f) << 16; - - /* SNORM8 */ - dw[11] = (int8_t) util_iround(rgba[0] * 127.0f) | - (int8_t) util_iround(rgba[1] * 127.0f) << 8 | - (int8_t) util_iround(rgba[2] * 127.0f) << 16 | - (int8_t) util_iround(rgba[3] * 127.0f) << 24; - - /* clamp to [0.0f, 1.0f] */ - rgba[0] = CLAMP(rgba[0], 0.0f, 1.0f); - rgba[1] = CLAMP(rgba[1], 0.0f, 1.0f); - rgba[2] = CLAMP(rgba[2], 0.0f, 1.0f); - rgba[3] = CLAMP(rgba[3], 0.0f, 1.0f); - - /* UNORM8 */ - dw[0] = (uint8_t) util_iround(rgba[0] * 255.0f) | - (uint8_t) util_iround(rgba[1] * 255.0f) << 8 | - (uint8_t) util_iround(rgba[2] * 255.0f) << 16 | - (uint8_t) util_iround(rgba[3] * 255.0f) << 24; - - /* UNORM16 */ - dw[7] = (uint16_t) util_iround(rgba[0] * 65535.0f) | - (uint16_t) util_iround(rgba[1] * 65535.0f) << 16; - dw[8] = (uint16_t) util_iround(rgba[2] * 65535.0f) | - (uint16_t) util_iround(rgba[3] * 65535.0f) << 16; -} - -void -ilo_gpe_init_sampler_cso(const struct ilo_dev_info *dev, - const struct pipe_sampler_state *state, - struct ilo_sampler_cso *sampler) -{ - int mip_filter, min_filter, mag_filter, max_aniso; - int lod_bias, max_lod, min_lod; - int wrap_s, wrap_t, wrap_r, wrap_cube; - bool clamp_is_to_edge; - uint32_t dw0, dw1, dw3; - - ILO_DEV_ASSERT(dev, 6, 7.5); - - memset(sampler, 0, sizeof(*sampler)); - - mip_filter = gen6_translate_tex_mipfilter(state->min_mip_filter); - min_filter = gen6_translate_tex_filter(state->min_img_filter); - mag_filter = gen6_translate_tex_filter(state->mag_img_filter); - - sampler->anisotropic = state->max_anisotropy; - - if (state->max_anisotropy >= 2 && state->max_anisotropy <= 16) - max_aniso = state->max_anisotropy / 2 - 1; - else if (state->max_anisotropy > 16) - max_aniso = GEN6_ANISORATIO_16; - else - max_aniso = GEN6_ANISORATIO_2; - - /* - * - * Here is how the hardware calculate per-pixel LOD, from my reading of the - * PRMs: - * - * 1) LOD is set to log2(ratio of texels to pixels) if not specified in - * other ways. The number of texels is measured using level - * SurfMinLod. - * 2) Bias is added to LOD. - * 3) LOD is clamped to [MinLod, MaxLod], and the clamped value is - * compared with Base to determine whether magnification or - * minification is needed. (if preclamp is disabled, LOD is compared - * with Base before clamping) - * 4) If magnification is needed, or no mipmapping is requested, LOD is - * set to floor(MinLod). - * 5) LOD is clamped to [0, MIPCnt], and SurfMinLod is added to LOD. - * - * With Gallium interface, Base is always zero and - * pipe_sampler_view::u.tex.first_level specifies SurfMinLod. - */ - if (ilo_dev_gen(dev) >= ILO_GEN(7)) { - const float scale = 256.0f; - - /* [-16.0, 16.0) in S4.8 */ - lod_bias = (int) - (CLAMP(state->lod_bias, -16.0f, 15.9f) * scale); - lod_bias &= 0x1fff; - - /* [0.0, 14.0] in U4.8 */ - max_lod = (int) (CLAMP(state->max_lod, 0.0f, 14.0f) * scale); - min_lod = (int) (CLAMP(state->min_lod, 0.0f, 14.0f) * scale); - } - else { - const float scale = 64.0f; - - /* [-16.0, 16.0) in S4.6 */ - lod_bias = (int) - (CLAMP(state->lod_bias, -16.0f, 15.9f) * scale); - lod_bias &= 0x7ff; - - /* [0.0, 13.0] in U4.6 */ - max_lod = (int) (CLAMP(state->max_lod, 0.0f, 13.0f) * scale); - min_lod = (int) (CLAMP(state->min_lod, 0.0f, 13.0f) * scale); - } - - /* - * We want LOD to be clamped to determine magnification/minification, and - * get set to zero when it is magnification or when mipmapping is disabled. - * The hardware would set LOD to floor(MinLod) and that is a problem when - * MinLod is greater than or equal to 1.0f. - * - * With Base being zero, it is always minification when MinLod is non-zero. - * To achieve our goal, we just need to set MinLod to zero and set - * MagFilter to MinFilter when mipmapping is disabled. - */ - if (state->min_mip_filter == PIPE_TEX_MIPFILTER_NONE && min_lod) { - min_lod = 0; - mag_filter = min_filter; - } - - /* - * For nearest filtering, PIPE_TEX_WRAP_CLAMP means - * PIPE_TEX_WRAP_CLAMP_TO_EDGE; for linear filtering, PIPE_TEX_WRAP_CLAMP - * means PIPE_TEX_WRAP_CLAMP_TO_BORDER while additionally clamping the - * texture coordinates to [0.0, 1.0]. - * - * The clamping will be taken care of in the shaders. There are two - * filters here, but let the minification one has a say. - */ - clamp_is_to_edge = (state->min_img_filter == PIPE_TEX_FILTER_NEAREST); - if (!clamp_is_to_edge) { - sampler->saturate_s = (state->wrap_s == PIPE_TEX_WRAP_CLAMP); - sampler->saturate_t = (state->wrap_t == PIPE_TEX_WRAP_CLAMP); - sampler->saturate_r = (state->wrap_r == PIPE_TEX_WRAP_CLAMP); - } - - /* determine wrap s/t/r */ - wrap_s = gen6_translate_tex_wrap(state->wrap_s, clamp_is_to_edge); - wrap_t = gen6_translate_tex_wrap(state->wrap_t, clamp_is_to_edge); - wrap_r = gen6_translate_tex_wrap(state->wrap_r, clamp_is_to_edge); - - /* - * From the Sandy Bridge PRM, volume 4 part 1, page 107: - * - * "When using cube map texture coordinates, only TEXCOORDMODE_CLAMP - * and TEXCOORDMODE_CUBE settings are valid, and each TC component - * must have the same Address Control mode." - * - * From the Ivy Bridge PRM, volume 4 part 1, page 96: - * - * "This field (Cube Surface Control Mode) must be set to - * CUBECTRLMODE_PROGRAMMED" - * - * Therefore, we cannot use "Cube Surface Control Mode" for semless cube - * map filtering. - */ - if (state->seamless_cube_map && - (state->min_img_filter != PIPE_TEX_FILTER_NEAREST || - state->mag_img_filter != PIPE_TEX_FILTER_NEAREST)) { - wrap_cube = GEN6_TEXCOORDMODE_CUBE; - } - else { - wrap_cube = GEN6_TEXCOORDMODE_CLAMP; - } - - if (!state->normalized_coords) { - /* - * From the Ivy Bridge PRM, volume 4 part 1, page 98: - * - * "The following state must be set as indicated if this field - * (Non-normalized Coordinate Enable) is enabled: - * - * - TCX/Y/Z Address Control Mode must be TEXCOORDMODE_CLAMP, - * TEXCOORDMODE_HALF_BORDER, or TEXCOORDMODE_CLAMP_BORDER. - * - Surface Type must be SURFTYPE_2D or SURFTYPE_3D. - * - Mag Mode Filter must be MAPFILTER_NEAREST or - * MAPFILTER_LINEAR. - * - Min Mode Filter must be MAPFILTER_NEAREST or - * MAPFILTER_LINEAR. - * - Mip Mode Filter must be MIPFILTER_NONE. - * - Min LOD must be 0. - * - Max LOD must be 0. - * - MIP Count must be 0. - * - Surface Min LOD must be 0. - * - Texture LOD Bias must be 0." - */ - assert(wrap_s == GEN6_TEXCOORDMODE_CLAMP || - wrap_s == GEN6_TEXCOORDMODE_CLAMP_BORDER); - assert(wrap_t == GEN6_TEXCOORDMODE_CLAMP || - wrap_t == GEN6_TEXCOORDMODE_CLAMP_BORDER); - assert(wrap_r == GEN6_TEXCOORDMODE_CLAMP || - wrap_r == GEN6_TEXCOORDMODE_CLAMP_BORDER); - - assert(mag_filter == GEN6_MAPFILTER_NEAREST || - mag_filter == GEN6_MAPFILTER_LINEAR); - assert(min_filter == GEN6_MAPFILTER_NEAREST || - min_filter == GEN6_MAPFILTER_LINEAR); - - /* work around a bug in util_blitter */ - mip_filter = GEN6_MIPFILTER_NONE; - - assert(mip_filter == GEN6_MIPFILTER_NONE); - } - - if (ilo_dev_gen(dev) >= ILO_GEN(7)) { - dw0 = 1 << 28 | - mip_filter << 20 | - lod_bias << 1; - - sampler->dw_filter = mag_filter << 17 | - min_filter << 14; - - sampler->dw_filter_aniso = GEN6_MAPFILTER_ANISOTROPIC << 17 | - GEN6_MAPFILTER_ANISOTROPIC << 14 | - 1; - - dw1 = min_lod << 20 | - max_lod << 8; - - if (state->compare_mode != PIPE_TEX_COMPARE_NONE) - dw1 |= gen6_translate_shadow_func(state->compare_func) << 1; - - dw3 = max_aniso << 19; - - /* round the coordinates for linear filtering */ - if (min_filter != GEN6_MAPFILTER_NEAREST) { - dw3 |= (GEN6_SAMPLER_DW3_U_MIN_ROUND | - GEN6_SAMPLER_DW3_V_MIN_ROUND | - GEN6_SAMPLER_DW3_R_MIN_ROUND); - } - if (mag_filter != GEN6_MAPFILTER_NEAREST) { - dw3 |= (GEN6_SAMPLER_DW3_U_MAG_ROUND | - GEN6_SAMPLER_DW3_V_MAG_ROUND | - GEN6_SAMPLER_DW3_R_MAG_ROUND); - } - - if (!state->normalized_coords) - dw3 |= 1 << 10; - - sampler->dw_wrap = wrap_s << 6 | - wrap_t << 3 | - wrap_r; - - /* - * As noted in the classic i965 driver, the HW may still reference - * wrap_t and wrap_r for 1D textures. We need to set them to a safe - * mode - */ - sampler->dw_wrap_1d = wrap_s << 6 | - GEN6_TEXCOORDMODE_WRAP << 3 | - GEN6_TEXCOORDMODE_WRAP; - - sampler->dw_wrap_cube = wrap_cube << 6 | - wrap_cube << 3 | - wrap_cube; - - STATIC_ASSERT(Elements(sampler->payload) >= 7); - - sampler->payload[0] = dw0; - sampler->payload[1] = dw1; - sampler->payload[2] = dw3; - - memcpy(&sampler->payload[3], - state->border_color.ui, sizeof(state->border_color.ui)); - } - else { - dw0 = 1 << 28 | - mip_filter << 20 | - lod_bias << 3; - - if (state->compare_mode != PIPE_TEX_COMPARE_NONE) - dw0 |= gen6_translate_shadow_func(state->compare_func); - - sampler->dw_filter = (min_filter != mag_filter) << 27 | - mag_filter << 17 | - min_filter << 14; - - sampler->dw_filter_aniso = GEN6_MAPFILTER_ANISOTROPIC << 17 | - GEN6_MAPFILTER_ANISOTROPIC << 14; - - dw1 = min_lod << 22 | - max_lod << 12; - - sampler->dw_wrap = wrap_s << 6 | - wrap_t << 3 | - wrap_r; - - sampler->dw_wrap_1d = wrap_s << 6 | - GEN6_TEXCOORDMODE_WRAP << 3 | - GEN6_TEXCOORDMODE_WRAP; - - sampler->dw_wrap_cube = wrap_cube << 6 | - wrap_cube << 3 | - wrap_cube; - - dw3 = max_aniso << 19; - - /* round the coordinates for linear filtering */ - if (min_filter != GEN6_MAPFILTER_NEAREST) { - dw3 |= (GEN6_SAMPLER_DW3_U_MIN_ROUND | - GEN6_SAMPLER_DW3_V_MIN_ROUND | - GEN6_SAMPLER_DW3_R_MIN_ROUND); - } - if (mag_filter != GEN6_MAPFILTER_NEAREST) { - dw3 |= (GEN6_SAMPLER_DW3_U_MAG_ROUND | - GEN6_SAMPLER_DW3_V_MAG_ROUND | - GEN6_SAMPLER_DW3_R_MAG_ROUND); - } - - if (!state->normalized_coords) - dw3 |= 1; - - STATIC_ASSERT(Elements(sampler->payload) >= 15); - - sampler->payload[0] = dw0; - sampler->payload[1] = dw1; - sampler->payload[2] = dw3; - - sampler_init_border_color_gen6(dev, - &state->border_color, &sampler->payload[3], 12); - } -} - -void ilo_gpe_set_fb(const struct ilo_dev_info *dev, const struct pipe_framebuffer_state *state, struct ilo_fb_state *fb) diff --git a/src/gallium/drivers/ilo/ilo_state_3d_top.c b/src/gallium/drivers/ilo/ilo_state_3d_top.c new file mode 100644 index 00000000000..3bd70bddbf8 --- /dev/null +++ b/src/gallium/drivers/ilo/ilo_state_3d_top.c @@ -0,0 +1,1566 @@ +/* + * Mesa 3-D graphics library + * + * Copyright (C) 2012-2014 LunarG, Inc. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included + * in all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + * DEALINGS IN THE SOFTWARE. + * + * Authors: + * Chia-I Wu <olv@lunarg.com> + */ + +#include "genhw/genhw.h" +#include "util/u_dual_blend.h" +#include "util/u_framebuffer.h" +#include "util/u_half.h" +#include "util/u_resource.h" + +#include "ilo_context.h" +#include "ilo_format.h" +#include "ilo_resource.h" +#include "ilo_shader.h" +#include "ilo_state.h" +#include "ilo_state_3d.h" + +static void +ve_init_cso(const struct ilo_dev_info *dev, + const struct pipe_vertex_element *state, + unsigned vb_index, + struct ilo_ve_cso *cso) +{ + int comp[4] = { + GEN6_VFCOMP_STORE_SRC, + GEN6_VFCOMP_STORE_SRC, + GEN6_VFCOMP_STORE_SRC, + GEN6_VFCOMP_STORE_SRC, + }; + int format; + + ILO_DEV_ASSERT(dev, 6, 7.5); + + switch (util_format_get_nr_components(state->src_format)) { + case 1: comp[1] = GEN6_VFCOMP_STORE_0; + case 2: comp[2] = GEN6_VFCOMP_STORE_0; + case 3: comp[3] = (util_format_is_pure_integer(state->src_format)) ? + GEN6_VFCOMP_STORE_1_INT : + GEN6_VFCOMP_STORE_1_FP; + } + + format = ilo_translate_vertex_format(dev, state->src_format); + + STATIC_ASSERT(Elements(cso->payload) >= 2); + cso->payload[0] = + vb_index << GEN6_VE_STATE_DW0_VB_INDEX__SHIFT | + GEN6_VE_STATE_DW0_VALID | + format << GEN6_VE_STATE_DW0_FORMAT__SHIFT | + state->src_offset << GEN6_VE_STATE_DW0_VB_OFFSET__SHIFT; + + cso->payload[1] = + comp[0] << GEN6_VE_STATE_DW1_COMP0__SHIFT | + comp[1] << GEN6_VE_STATE_DW1_COMP1__SHIFT | + comp[2] << GEN6_VE_STATE_DW1_COMP2__SHIFT | + comp[3] << GEN6_VE_STATE_DW1_COMP3__SHIFT; +} + +void +ilo_gpe_init_ve(const struct ilo_dev_info *dev, + unsigned num_states, + const struct pipe_vertex_element *states, + struct ilo_ve_state *ve) +{ + unsigned i; + + ILO_DEV_ASSERT(dev, 6, 7.5); + + ve->count = num_states; + ve->vb_count = 0; + + for (i = 0; i < num_states; i++) { + const unsigned pipe_idx = states[i].vertex_buffer_index; + const unsigned instance_divisor = states[i].instance_divisor; + unsigned hw_idx; + + /* + * map the pipe vb to the hardware vb, which has a fixed instance + * divisor + */ + for (hw_idx = 0; hw_idx < ve->vb_count; hw_idx++) { + if (ve->vb_mapping[hw_idx] == pipe_idx && + ve->instance_divisors[hw_idx] == instance_divisor) + break; + } + + /* create one if there is no matching hardware vb */ + if (hw_idx >= ve->vb_count) { + hw_idx = ve->vb_count++; + + ve->vb_mapping[hw_idx] = pipe_idx; + ve->instance_divisors[hw_idx] = instance_divisor; + } + + ve_init_cso(dev, &states[i], hw_idx, &ve->cso[i]); + } +} + +void +ilo_gpe_set_ve_edgeflag(const struct ilo_dev_info *dev, + struct ilo_ve_cso *cso) +{ + int format; + + ILO_DEV_ASSERT(dev, 6, 7.5); + + /* + * From the Sandy Bridge PRM, volume 2 part 1, page 94: + * + * "- This bit (Edge Flag Enable) must only be ENABLED on the last + * valid VERTEX_ELEMENT structure. + * + * - When set, Component 0 Control must be set to VFCOMP_STORE_SRC, + * and Component 1-3 Control must be set to VFCOMP_NOSTORE. + * + * - The Source Element Format must be set to the UINT format. + * + * - [DevSNB]: Edge Flags are not supported for QUADLIST + * primitives. Software may elect to convert QUADLIST primitives + * to some set of corresponding edge-flag-supported primitive + * types (e.g., POLYGONs) prior to submission to the 3D pipeline." + */ + cso->payload[0] |= GEN6_VE_STATE_DW0_EDGE_FLAG_ENABLE; + + /* + * Edge flags have format GEN6_FORMAT_R8_USCALED when defined via + * glEdgeFlagPointer(), and format GEN6_FORMAT_R32_FLOAT when defined + * via glEdgeFlag(), as can be seen in vbo_attrib_tmp.h. + * + * Since all the hardware cares about is whether the flags are zero or not, + * we can treat them as the corresponding _UINT formats. + */ + format = GEN_EXTRACT(cso->payload[0], GEN6_VE_STATE_DW0_FORMAT); + cso->payload[0] &= ~GEN6_VE_STATE_DW0_FORMAT__MASK; + + switch (format) { + case GEN6_FORMAT_R32_FLOAT: + format = GEN6_FORMAT_R32_UINT; + break; + case GEN6_FORMAT_R8_USCALED: + format = GEN6_FORMAT_R8_UINT; + break; + default: + break; + } + + cso->payload[0] |= GEN_SHIFT32(format, GEN6_VE_STATE_DW0_FORMAT); + + cso->payload[1] = + GEN6_VFCOMP_STORE_SRC << GEN6_VE_STATE_DW1_COMP0__SHIFT | + GEN6_VFCOMP_NOSTORE << GEN6_VE_STATE_DW1_COMP1__SHIFT | + GEN6_VFCOMP_NOSTORE << GEN6_VE_STATE_DW1_COMP2__SHIFT | + GEN6_VFCOMP_NOSTORE << GEN6_VE_STATE_DW1_COMP3__SHIFT; +} + +void +ilo_gpe_init_ve_nosrc(const struct ilo_dev_info *dev, + int comp0, int comp1, int comp2, int comp3, + struct ilo_ve_cso *cso) +{ + ILO_DEV_ASSERT(dev, 6, 7.5); + + STATIC_ASSERT(Elements(cso->payload) >= 2); + + assert(comp0 != GEN6_VFCOMP_STORE_SRC && + comp1 != GEN6_VFCOMP_STORE_SRC && + comp2 != GEN6_VFCOMP_STORE_SRC && + comp3 != GEN6_VFCOMP_STORE_SRC); + + cso->payload[0] = GEN6_VE_STATE_DW0_VALID; + cso->payload[1] = + comp0 << GEN6_VE_STATE_DW1_COMP0__SHIFT | + comp1 << GEN6_VE_STATE_DW1_COMP1__SHIFT | + comp2 << GEN6_VE_STATE_DW1_COMP2__SHIFT | + comp3 << GEN6_VE_STATE_DW1_COMP3__SHIFT; +} + +void +ilo_gpe_init_vs_cso(const struct ilo_dev_info *dev, + const struct ilo_shader_state *vs, + struct ilo_shader_cso *cso) +{ + int start_grf, vue_read_len, sampler_count, max_threads; + uint32_t dw2, dw4, dw5; + + ILO_DEV_ASSERT(dev, 6, 7.5); + + start_grf = ilo_shader_get_kernel_param(vs, ILO_KERNEL_URB_DATA_START_REG); + vue_read_len = ilo_shader_get_kernel_param(vs, ILO_KERNEL_INPUT_COUNT); + sampler_count = ilo_shader_get_kernel_param(vs, ILO_KERNEL_SAMPLER_COUNT); + + /* + * From the Sandy Bridge PRM, volume 2 part 1, page 135: + * + * "(Vertex URB Entry Read Length) Specifies the number of pairs of + * 128-bit vertex elements to be passed into the payload for each + * vertex." + * + * "It is UNDEFINED to set this field to 0 indicating no Vertex URB + * data to be read and passed to the thread." + */ + vue_read_len = (vue_read_len + 1) / 2; + if (!vue_read_len) + vue_read_len = 1; + + max_threads = dev->thread_count; + if (ilo_dev_gen(dev) == ILO_GEN(7.5) && dev->gt == 2) + max_threads *= 2; + + dw2 = (true) ? 0 : GEN6_THREADDISP_FP_MODE_ALT; + dw2 |= ((sampler_count + 3) / 4) << GEN6_THREADDISP_SAMPLER_COUNT__SHIFT; + + dw4 = start_grf << GEN6_VS_DW4_URB_GRF_START__SHIFT | + vue_read_len << GEN6_VS_DW4_URB_READ_LEN__SHIFT | + 0 << GEN6_VS_DW4_URB_READ_OFFSET__SHIFT; + + dw5 = GEN6_VS_DW5_STATISTICS | + GEN6_VS_DW5_VS_ENABLE; + + if (ilo_dev_gen(dev) >= ILO_GEN(7.5)) + dw5 |= (max_threads - 1) << GEN75_VS_DW5_MAX_THREADS__SHIFT; + else + dw5 |= (max_threads - 1) << GEN6_VS_DW5_MAX_THREADS__SHIFT; + + STATIC_ASSERT(Elements(cso->payload) >= 3); + cso->payload[0] = dw2; + cso->payload[1] = dw4; + cso->payload[2] = dw5; +} + +void +ilo_gpe_init_gs_cso_gen6(const struct ilo_dev_info *dev, + const struct ilo_shader_state *gs, + struct ilo_shader_cso *cso) +{ + int start_grf, vue_read_len, max_threads; + uint32_t dw2, dw4, dw5, dw6; + + ILO_DEV_ASSERT(dev, 6, 6); + + if (ilo_shader_get_type(gs) == PIPE_SHADER_GEOMETRY) { + start_grf = ilo_shader_get_kernel_param(gs, + ILO_KERNEL_URB_DATA_START_REG); + + vue_read_len = ilo_shader_get_kernel_param(gs, ILO_KERNEL_INPUT_COUNT); + } + else { + start_grf = ilo_shader_get_kernel_param(gs, + ILO_KERNEL_VS_GEN6_SO_START_REG); + + vue_read_len = ilo_shader_get_kernel_param(gs, ILO_KERNEL_OUTPUT_COUNT); + } + + /* + * From the Sandy Bridge PRM, volume 2 part 1, page 153: + * + * "Specifies the amount of URB data read and passed in the thread + * payload for each Vertex URB entry, in 256-bit register increments. + * + * It is UNDEFINED to set this field (Vertex URB Entry Read Length) to + * 0 indicating no Vertex URB data to be read and passed to the + * thread." + */ + vue_read_len = (vue_read_len + 1) / 2; + if (!vue_read_len) + vue_read_len = 1; + + /* + * From the Sandy Bridge PRM, volume 2 part 1, page 154: + * + * "Maximum Number of Threads valid range is [0,27] when Rendering + * Enabled bit is set." + * + * From the Sandy Bridge PRM, volume 2 part 1, page 173: + * + * "Programming Note: If the GS stage is enabled, software must always + * allocate at least one GS URB Entry. This is true even if the GS + * thread never needs to output vertices to the pipeline, e.g., when + * only performing stream output. This is an artifact of the need to + * pass the GS thread an initial destination URB handle." + * + * As such, we always enable rendering, and limit the number of threads. + */ + if (dev->gt == 2) { + /* maximum is 60, but limited to 28 */ + max_threads = 28; + } + else { + /* maximum is 24, but limited to 21 (see brwCreateContext()) */ + max_threads = 21; + } + + dw2 = GEN6_THREADDISP_SPF; + + dw4 = vue_read_len << GEN6_GS_DW4_URB_READ_LEN__SHIFT | + 0 << GEN6_GS_DW4_URB_READ_OFFSET__SHIFT | + start_grf << GEN6_GS_DW4_URB_GRF_START__SHIFT; + + dw5 = (max_threads - 1) << GEN6_GS_DW5_MAX_THREADS__SHIFT | + GEN6_GS_DW5_STATISTICS | + GEN6_GS_DW5_SO_STATISTICS | + GEN6_GS_DW5_RENDER_ENABLE; + + /* + * we cannot make use of GEN6_GS_REORDER because it will reorder + * triangle strips according to D3D rules (triangle 2N+1 uses vertices + * (2N+1, 2N+3, 2N+2)), instead of GL rules (triangle 2N+1 uses vertices + * (2N+2, 2N+1, 2N+3)). + */ + dw6 = GEN6_GS_DW6_GS_ENABLE; + + if (ilo_shader_get_kernel_param(gs, ILO_KERNEL_GS_DISCARD_ADJACENCY)) + dw6 |= GEN6_GS_DW6_DISCARD_ADJACENCY; + + if (ilo_shader_get_kernel_param(gs, ILO_KERNEL_VS_GEN6_SO)) { + const uint32_t svbi_post_inc = + ilo_shader_get_kernel_param(gs, ILO_KERNEL_GS_GEN6_SVBI_POST_INC); + + dw6 |= GEN6_GS_DW6_SVBI_PAYLOAD_ENABLE; + if (svbi_post_inc) { + dw6 |= GEN6_GS_DW6_SVBI_POST_INC_ENABLE | + svbi_post_inc << GEN6_GS_DW6_SVBI_POST_INC_VAL__SHIFT; + } + } + + STATIC_ASSERT(Elements(cso->payload) >= 4); + cso->payload[0] = dw2; + cso->payload[1] = dw4; + cso->payload[2] = dw5; + cso->payload[3] = dw6; +} + +void +ilo_gpe_init_gs_cso_gen7(const struct ilo_dev_info *dev, + const struct ilo_shader_state *gs, + struct ilo_shader_cso *cso) +{ + int start_grf, vue_read_len, sampler_count, max_threads; + uint32_t dw2, dw4, dw5; + + ILO_DEV_ASSERT(dev, 7, 7.5); + + start_grf = ilo_shader_get_kernel_param(gs, ILO_KERNEL_URB_DATA_START_REG); + vue_read_len = ilo_shader_get_kernel_param(gs, ILO_KERNEL_INPUT_COUNT); + sampler_count = ilo_shader_get_kernel_param(gs, ILO_KERNEL_SAMPLER_COUNT); + + /* in pairs */ + vue_read_len = (vue_read_len + 1) / 2; + + switch (ilo_dev_gen(dev)) { + case ILO_GEN(7.5): + max_threads = (dev->gt >= 2) ? 256 : 70; + break; + case ILO_GEN(7): + max_threads = (dev->gt == 2) ? 128 : 36; + break; + default: + max_threads = 1; + break; + } + + dw2 = (true) ? 0 : GEN6_THREADDISP_FP_MODE_ALT; + dw2 |= ((sampler_count + 3) / 4) << GEN6_THREADDISP_SAMPLER_COUNT__SHIFT; + + dw4 = vue_read_len << GEN7_GS_DW4_URB_READ_LEN__SHIFT | + GEN7_GS_DW4_INCLUDE_VERTEX_HANDLES | + 0 << GEN7_GS_DW4_URB_READ_OFFSET__SHIFT | + start_grf << GEN7_GS_DW4_URB_GRF_START__SHIFT; + + dw5 = (max_threads - 1) << GEN7_GS_DW5_MAX_THREADS__SHIFT | + GEN7_GS_DW5_STATISTICS | + GEN7_GS_DW5_GS_ENABLE; + + STATIC_ASSERT(Elements(cso->payload) >= 3); + cso->payload[0] = dw2; + cso->payload[1] = dw4; + cso->payload[2] = dw5; +} + +void +ilo_gpe_init_view_surface_null_gen6(const struct ilo_dev_info *dev, + unsigned width, unsigned height, + unsigned depth, unsigned level, + struct ilo_view_surface *surf) +{ + uint32_t *dw; + + ILO_DEV_ASSERT(dev, 6, 6); + + assert(width >= 1 && height >= 1 && depth >= 1); + + /* + * From the Sandy Bridge PRM, volume 4 part 1, page 71: + * + * "A null surface will be used in instances where an actual surface is + * not bound. When a write message is generated to a null surface, no + * actual surface is written to. When a read message (including any + * sampling engine message) is generated to a null surface, the result + * is all zeros. Note that a null surface type is allowed to be used + * with all messages, even if it is not specificially indicated as + * supported. All of the remaining fields in surface state are ignored + * for null surfaces, with the following exceptions: + * + * * [DevSNB+]: Width, Height, Depth, and LOD fields must match the + * depth buffer's corresponding state for all render target + * surfaces, including null. + * * Surface Format must be R8G8B8A8_UNORM." + * + * From the Sandy Bridge PRM, volume 4 part 1, page 82: + * + * "If Surface Type is SURFTYPE_NULL, this field (Tiled Surface) must be + * true" + */ + + STATIC_ASSERT(Elements(surf->payload) >= 6); + dw = surf->payload; + + dw[0] = GEN6_SURFTYPE_NULL << GEN6_SURFACE_DW0_TYPE__SHIFT | + GEN6_FORMAT_B8G8R8A8_UNORM << GEN6_SURFACE_DW0_FORMAT__SHIFT; + + dw[1] = 0; + + dw[2] = (height - 1) << GEN6_SURFACE_DW2_HEIGHT__SHIFT | + (width - 1) << GEN6_SURFACE_DW2_WIDTH__SHIFT | + level << GEN6_SURFACE_DW2_MIP_COUNT_LOD__SHIFT; + + dw[3] = (depth - 1) << GEN6_SURFACE_DW3_DEPTH__SHIFT | + GEN6_TILING_X; + + dw[4] = 0; + dw[5] = 0; + + surf->bo = NULL; +} + +void +ilo_gpe_init_view_surface_for_buffer_gen6(const struct ilo_dev_info *dev, + const struct ilo_buffer *buf, + unsigned offset, unsigned size, + unsigned struct_size, + enum pipe_format elem_format, + bool is_rt, bool render_cache_rw, + struct ilo_view_surface *surf) +{ + const int elem_size = util_format_get_blocksize(elem_format); + int width, height, depth, pitch; + int surface_format, num_entries; + uint32_t *dw; + + ILO_DEV_ASSERT(dev, 6, 6); + + /* + * For SURFTYPE_BUFFER, a SURFACE_STATE specifies an element of a + * structure in a buffer. + */ + + surface_format = ilo_translate_color_format(dev, elem_format); + + num_entries = size / struct_size; + /* see if there is enough space to fit another element */ + if (size % struct_size >= elem_size) + num_entries++; + + /* + * From the Sandy Bridge PRM, volume 4 part 1, page 76: + * + * "For SURFTYPE_BUFFER render targets, this field (Surface Base + * Address) specifies the base address of first element of the + * surface. The surface is interpreted as a simple array of that + * single element type. The address must be naturally-aligned to the + * element size (e.g., a buffer containing R32G32B32A32_FLOAT elements + * must be 16-byte aligned). + * + * For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies + * the base address of the first element of the surface, computed in + * software by adding the surface base address to the byte offset of + * the element in the buffer." + */ + if (is_rt) + assert(offset % elem_size == 0); + + /* + * From the Sandy Bridge PRM, volume 4 part 1, page 77: + * + * "For buffer surfaces, the number of entries in the buffer ranges + * from 1 to 2^27." + */ + assert(num_entries >= 1 && num_entries <= 1 << 27); + + /* + * From the Sandy Bridge PRM, volume 4 part 1, page 81: + * + * "For surfaces of type SURFTYPE_BUFFER, this field (Surface Pitch) + * indicates the size of the structure." + */ + pitch = struct_size; + + pitch--; + num_entries--; + /* bits [6:0] */ + width = (num_entries & 0x0000007f); + /* bits [19:7] */ + height = (num_entries & 0x000fff80) >> 7; + /* bits [26:20] */ + depth = (num_entries & 0x07f00000) >> 20; + + STATIC_ASSERT(Elements(surf->payload) >= 6); + dw = surf->payload; + + dw[0] = GEN6_SURFTYPE_BUFFER << GEN6_SURFACE_DW0_TYPE__SHIFT | + surface_format << GEN6_SURFACE_DW0_FORMAT__SHIFT; + if (render_cache_rw) + dw[0] |= GEN6_SURFACE_DW0_RENDER_CACHE_RW; + + dw[1] = offset; + + dw[2] = height << GEN6_SURFACE_DW2_HEIGHT__SHIFT | + width << GEN6_SURFACE_DW2_WIDTH__SHIFT; + + dw[3] = depth << GEN6_SURFACE_DW3_DEPTH__SHIFT | + pitch << GEN6_SURFACE_DW3_PITCH__SHIFT; + + dw[4] = 0; + dw[5] = 0; + + /* do not increment reference count */ + surf->bo = buf->bo; +} + +void +ilo_gpe_init_view_surface_for_texture_gen6(const struct ilo_dev_info *dev, + const struct ilo_texture *tex, + enum pipe_format format, + unsigned first_level, + unsigned num_levels, + unsigned first_layer, + unsigned num_layers, + bool is_rt, + struct ilo_view_surface *surf) +{ + int surface_type, surface_format; + int width, height, depth, pitch, lod; + uint32_t *dw; + + ILO_DEV_ASSERT(dev, 6, 6); + + surface_type = ilo_gpe_gen6_translate_texture(tex->base.target); + assert(surface_type != GEN6_SURFTYPE_BUFFER); + + if (format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT && tex->separate_s8) + format = PIPE_FORMAT_Z32_FLOAT; + + if (is_rt) + surface_format = ilo_translate_render_format(dev, format); + else + surface_format = ilo_translate_texture_format(dev, format); + assert(surface_format >= 0); + + width = tex->layout.width0; + height = tex->layout.height0; + depth = (tex->base.target == PIPE_TEXTURE_3D) ? + tex->base.depth0 : num_layers; + pitch = tex->layout.bo_stride; + + if (surface_type == GEN6_SURFTYPE_CUBE) { + /* + * From the Sandy Bridge PRM, volume 4 part 1, page 81: + * + * "For SURFTYPE_CUBE: [DevSNB+]: for Sampling Engine Surfaces, the + * range of this field (Depth) is [0,84], indicating the number of + * cube array elements (equal to the number of underlying 2D array + * elements divided by 6). For other surfaces, this field must be + * zero." + * + * When is_rt is true, we treat the texture as a 2D one to avoid the + * restriction. + */ + if (is_rt) { + surface_type = GEN6_SURFTYPE_2D; + } + else { + assert(num_layers % 6 == 0); + depth = num_layers / 6; + } + } + + /* sanity check the size */ + assert(width >= 1 && height >= 1 && depth >= 1 && pitch >= 1); + switch (surface_type) { + case GEN6_SURFTYPE_1D: + assert(width <= 8192 && height == 1 && depth <= 512); + assert(first_layer < 512 && num_layers <= 512); + break; + case GEN6_SURFTYPE_2D: + assert(width <= 8192 && height <= 8192 && depth <= 512); + assert(first_layer < 512 && num_layers <= 512); + break; + case GEN6_SURFTYPE_3D: + assert(width <= 2048 && height <= 2048 && depth <= 2048); + assert(first_layer < 2048 && num_layers <= 512); + if (!is_rt) + assert(first_layer == 0); + break; + case GEN6_SURFTYPE_CUBE: + assert(width <= 8192 && height <= 8192 && depth <= 85); + assert(width == height); + assert(first_layer < 512 && num_layers <= 512); + if (is_rt) + assert(first_layer == 0); + break; + default: + assert(!"unexpected surface type"); + break; + } + + /* non-full array spacing is supported only on GEN7+ */ + assert(tex->layout.walk != ILO_LAYOUT_WALK_LOD); + /* non-interleaved samples are supported only on GEN7+ */ + if (tex->base.nr_samples > 1) + assert(tex->layout.interleaved_samples); + + if (is_rt) { + assert(num_levels == 1); + lod = first_level; + } + else { + lod = num_levels - 1; + } + + /* + * From the Sandy Bridge PRM, volume 4 part 1, page 76: + * + * "Linear render target surface base addresses must be element-size + * aligned, for non-YUV surface formats, or a multiple of 2 + * element-sizes for YUV surface formats. Other linear surfaces have + * no alignment requirements (byte alignment is sufficient.)" + * + * From the Sandy Bridge PRM, volume 4 part 1, page 81: + * + * "For linear render target surfaces, the pitch must be a multiple + * of the element size for non-YUV surface formats. Pitch must be a + * multiple of 2 * element size for YUV surface formats." + * + * From the Sandy Bridge PRM, volume 4 part 1, page 86: + * + * "For linear surfaces, this field (X Offset) must be zero" + */ + if (tex->layout.tiling == INTEL_TILING_NONE) { + if (is_rt) { + const int elem_size = util_format_get_blocksize(format); + assert(pitch % elem_size == 0); + } + } + + STATIC_ASSERT(Elements(surf->payload) >= 6); + dw = surf->payload; + + dw[0] = surface_type << GEN6_SURFACE_DW0_TYPE__SHIFT | + surface_format << GEN6_SURFACE_DW0_FORMAT__SHIFT | + GEN6_SURFACE_DW0_MIPLAYOUT_BELOW; + + if (surface_type == GEN6_SURFTYPE_CUBE && !is_rt) { + dw[0] |= 1 << 9 | + GEN6_SURFACE_DW0_CUBE_FACE_ENABLES__MASK; + } + + if (is_rt) + dw[0] |= GEN6_SURFACE_DW0_RENDER_CACHE_RW; + + dw[1] = 0; + + dw[2] = (height - 1) << GEN6_SURFACE_DW2_HEIGHT__SHIFT | + (width - 1) << GEN6_SURFACE_DW2_WIDTH__SHIFT | + lod << GEN6_SURFACE_DW2_MIP_COUNT_LOD__SHIFT; + + dw[3] = (depth - 1) << GEN6_SURFACE_DW3_DEPTH__SHIFT | + (pitch - 1) << GEN6_SURFACE_DW3_PITCH__SHIFT | + ilo_gpe_gen6_translate_winsys_tiling(tex->layout.tiling); + + dw[4] = first_level << GEN6_SURFACE_DW4_MIN_LOD__SHIFT | + first_layer << 17 | + (num_layers - 1) << 8 | + ((tex->base.nr_samples > 1) ? GEN6_SURFACE_DW4_MULTISAMPLECOUNT_4 : + GEN6_SURFACE_DW4_MULTISAMPLECOUNT_1); + + dw[5] = 0; + + assert(tex->layout.align_j == 2 || tex->layout.align_j == 4); + if (tex->layout.align_j == 4) + dw[5] |= GEN6_SURFACE_DW5_VALIGN_4; + + /* do not increment reference count */ + surf->bo = tex->bo; +} + +void +ilo_gpe_init_view_surface_null_gen7(const struct ilo_dev_info *dev, + unsigned width, unsigned height, + unsigned depth, unsigned level, + struct ilo_view_surface *surf) +{ + uint32_t *dw; + + ILO_DEV_ASSERT(dev, 7, 7.5); + + assert(width >= 1 && height >= 1 && depth >= 1); + + /* + * From the Ivy Bridge PRM, volume 4 part 1, page 62: + * + * "A null surface is used in instances where an actual surface is not + * bound. When a write message is generated to a null surface, no + * actual surface is written to. When a read message (including any + * sampling engine message) is generated to a null surface, the result + * is all zeros. Note that a null surface type is allowed to be used + * with all messages, even if it is not specificially indicated as + * supported. All of the remaining fields in surface state are ignored + * for null surfaces, with the following exceptions: + * + * * Width, Height, Depth, LOD, and Render Target View Extent fields + * must match the depth buffer's corresponding state for all render + * target surfaces, including null. + * * All sampling engine and data port messages support null surfaces + * with the above behavior, even if not mentioned as specifically + * supported, except for the following: + * * Data Port Media Block Read/Write messages. + * * The Surface Type of a surface used as a render target (accessed + * via the Data Port's Render Target Write message) must be the same + * as the Surface Type of all other render targets and of the depth + * buffer (defined in 3DSTATE_DEPTH_BUFFER), unless either the depth + * buffer or render targets are SURFTYPE_NULL." + * + * From the Ivy Bridge PRM, volume 4 part 1, page 65: + * + * "If Surface Type is SURFTYPE_NULL, this field (Tiled Surface) must be + * true" + */ + + STATIC_ASSERT(Elements(surf->payload) >= 8); + dw = surf->payload; + + dw[0] = GEN6_SURFTYPE_NULL << GEN7_SURFACE_DW0_TYPE__SHIFT | + GEN6_FORMAT_B8G8R8A8_UNORM << GEN7_SURFACE_DW0_FORMAT__SHIFT | + GEN6_TILING_X << 13; + + dw[1] = 0; + + dw[2] = GEN_SHIFT32(height - 1, GEN7_SURFACE_DW2_HEIGHT) | + GEN_SHIFT32(width - 1, GEN7_SURFACE_DW2_WIDTH); + + dw[3] = GEN_SHIFT32(depth - 1, GEN7_SURFACE_DW3_DEPTH); + + dw[4] = 0; + dw[5] = level; + + dw[6] = 0; + dw[7] = 0; + + surf->bo = NULL; +} + +void +ilo_gpe_init_view_surface_for_buffer_gen7(const struct ilo_dev_info *dev, + const struct ilo_buffer *buf, + unsigned offset, unsigned size, + unsigned struct_size, + enum pipe_format elem_format, + bool is_rt, bool render_cache_rw, + struct ilo_view_surface *surf) +{ + const bool typed = (elem_format != PIPE_FORMAT_NONE); + const bool structured = (!typed && struct_size > 1); + const int elem_size = (typed) ? + util_format_get_blocksize(elem_format) : 1; + int width, height, depth, pitch; + int surface_type, surface_format, num_entries; + uint32_t *dw; + + ILO_DEV_ASSERT(dev, 7, 7.5); + + surface_type = (structured) ? GEN7_SURFTYPE_STRBUF : GEN6_SURFTYPE_BUFFER; + + surface_format = (typed) ? + ilo_translate_color_format(dev, elem_format) : GEN6_FORMAT_RAW; + + num_entries = size / struct_size; + /* see if there is enough space to fit another element */ + if (size % struct_size >= elem_size && !structured) + num_entries++; + + /* + * From the Ivy Bridge PRM, volume 4 part 1, page 67: + * + * "For SURFTYPE_BUFFER render targets, this field (Surface Base + * Address) specifies the base address of first element of the + * surface. The surface is interpreted as a simple array of that + * single element type. The address must be naturally-aligned to the + * element size (e.g., a buffer containing R32G32B32A32_FLOAT elements + * must be 16-byte aligned) + * + * For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies + * the base address of the first element of the surface, computed in + * software by adding the surface base address to the byte offset of + * the element in the buffer." + */ + if (is_rt) + assert(offset % elem_size == 0); + + /* + * From the Ivy Bridge PRM, volume 4 part 1, page 68: + * + * "For typed buffer and structured buffer surfaces, the number of + * entries in the buffer ranges from 1 to 2^27. For raw buffer + * surfaces, the number of entries in the buffer is the number of + * bytes which can range from 1 to 2^30." + */ + assert(num_entries >= 1 && + num_entries <= 1 << ((typed || structured) ? 27 : 30)); + + /* + * From the Ivy Bridge PRM, volume 4 part 1, page 69: + * + * "For SURFTYPE_BUFFER: The low two bits of this field (Width) must be + * 11 if the Surface Format is RAW (the size of the buffer must be a + * multiple of 4 bytes)." + * + * From the Ivy Bridge PRM, volume 4 part 1, page 70: + * + * "For surfaces of type SURFTYPE_BUFFER and SURFTYPE_STRBUF, this + * field (Surface Pitch) indicates the size of the structure." + * + * "For linear surfaces with Surface Type of SURFTYPE_STRBUF, the pitch + * must be a multiple of 4 bytes." + */ + if (structured) + assert(struct_size % 4 == 0); + else if (!typed) + assert(num_entries % 4 == 0); + + pitch = struct_size; + + pitch--; + num_entries--; + /* bits [6:0] */ + width = (num_entries & 0x0000007f); + /* bits [20:7] */ + height = (num_entries & 0x001fff80) >> 7; + /* bits [30:21] */ + depth = (num_entries & 0x7fe00000) >> 21; + /* limit to [26:21] */ + if (typed || structured) + depth &= 0x3f; + + STATIC_ASSERT(Elements(surf->payload) >= 8); + dw = surf->payload; + + dw[0] = surface_type << GEN7_SURFACE_DW0_TYPE__SHIFT | + surface_format << GEN7_SURFACE_DW0_FORMAT__SHIFT; + if (render_cache_rw) + dw[0] |= GEN7_SURFACE_DW0_RENDER_CACHE_RW; + + dw[1] = offset; + + dw[2] = GEN_SHIFT32(height, GEN7_SURFACE_DW2_HEIGHT) | + GEN_SHIFT32(width, GEN7_SURFACE_DW2_WIDTH); + + dw[3] = GEN_SHIFT32(depth, GEN7_SURFACE_DW3_DEPTH) | + pitch; + + dw[4] = 0; + dw[5] = 0; + + dw[6] = 0; + dw[7] = 0; + + if (ilo_dev_gen(dev) >= ILO_GEN(7.5)) { + dw[7] |= GEN_SHIFT32(GEN75_SCS_RED, GEN75_SURFACE_DW7_SCS_R) | + GEN_SHIFT32(GEN75_SCS_GREEN, GEN75_SURFACE_DW7_SCS_G) | + GEN_SHIFT32(GEN75_SCS_BLUE, GEN75_SURFACE_DW7_SCS_B) | + GEN_SHIFT32(GEN75_SCS_ALPHA, GEN75_SURFACE_DW7_SCS_A); + } + + /* do not increment reference count */ + surf->bo = buf->bo; +} + +void +ilo_gpe_init_view_surface_for_texture_gen7(const struct ilo_dev_info *dev, + const struct ilo_texture *tex, + enum pipe_format format, + unsigned first_level, + unsigned num_levels, + unsigned first_layer, + unsigned num_layers, + bool is_rt, + struct ilo_view_surface *surf) +{ + int surface_type, surface_format; + int width, height, depth, pitch, lod; + uint32_t *dw; + + ILO_DEV_ASSERT(dev, 7, 7.5); + + surface_type = ilo_gpe_gen6_translate_texture(tex->base.target); + assert(surface_type != GEN6_SURFTYPE_BUFFER); + + if (format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT && tex->separate_s8) + format = PIPE_FORMAT_Z32_FLOAT; + + if (is_rt) + surface_format = ilo_translate_render_format(dev, format); + else + surface_format = ilo_translate_texture_format(dev, format); + assert(surface_format >= 0); + + width = tex->layout.width0; + height = tex->layout.height0; + depth = (tex->base.target == PIPE_TEXTURE_3D) ? + tex->base.depth0 : num_layers; + pitch = tex->layout.bo_stride; + + if (surface_type == GEN6_SURFTYPE_CUBE) { + /* + * From the Ivy Bridge PRM, volume 4 part 1, page 70: + * + * "For SURFTYPE_CUBE:For Sampling Engine Surfaces, the range of + * this field is [0,340], indicating the number of cube array + * elements (equal to the number of underlying 2D array elements + * divided by 6). For other surfaces, this field must be zero." + * + * When is_rt is true, we treat the texture as a 2D one to avoid the + * restriction. + */ + if (is_rt) { + surface_type = GEN6_SURFTYPE_2D; + } + else { + assert(num_layers % 6 == 0); + depth = num_layers / 6; + } + } + + /* sanity check the size */ + assert(width >= 1 && height >= 1 && depth >= 1 && pitch >= 1); + assert(first_layer < 2048 && num_layers <= 2048); + switch (surface_type) { + case GEN6_SURFTYPE_1D: + assert(width <= 16384 && height == 1 && depth <= 2048); + break; + case GEN6_SURFTYPE_2D: + assert(width <= 16384 && height <= 16384 && depth <= 2048); + break; + case GEN6_SURFTYPE_3D: + assert(width <= 2048 && height <= 2048 && depth <= 2048); + if (!is_rt) + assert(first_layer == 0); + break; + case GEN6_SURFTYPE_CUBE: + assert(width <= 16384 && height <= 16384 && depth <= 86); + assert(width == height); + if (is_rt) + assert(first_layer == 0); + break; + default: + assert(!"unexpected surface type"); + break; + } + + if (is_rt) { + assert(num_levels == 1); + lod = first_level; + } + else { + lod = num_levels - 1; + } + + /* + * From the Ivy Bridge PRM, volume 4 part 1, page 68: + * + * "The Base Address for linear render target surfaces and surfaces + * accessed with the typed surface read/write data port messages must + * be element-size aligned, for non-YUV surface formats, or a multiple + * of 2 element-sizes for YUV surface formats. Other linear surfaces + * have no alignment requirements (byte alignment is sufficient)." + * + * From the Ivy Bridge PRM, volume 4 part 1, page 70: + * + * "For linear render target surfaces and surfaces accessed with the + * typed data port messages, the pitch must be a multiple of the + * element size for non-YUV surface formats. Pitch must be a multiple + * of 2 * element size for YUV surface formats. For linear surfaces + * with Surface Type of SURFTYPE_STRBUF, the pitch must be a multiple + * of 4 bytes.For other linear surfaces, the pitch can be any multiple + * of bytes." + * + * From the Ivy Bridge PRM, volume 4 part 1, page 74: + * + * "For linear surfaces, this field (X Offset) must be zero." + */ + if (tex->layout.tiling == INTEL_TILING_NONE) { + if (is_rt) { + const int elem_size = util_format_get_blocksize(format); + assert(pitch % elem_size == 0); + } + } + + STATIC_ASSERT(Elements(surf->payload) >= 8); + dw = surf->payload; + + dw[0] = surface_type << GEN7_SURFACE_DW0_TYPE__SHIFT | + surface_format << GEN7_SURFACE_DW0_FORMAT__SHIFT | + ilo_gpe_gen6_translate_winsys_tiling(tex->layout.tiling) << 13; + + /* + * From the Ivy Bridge PRM, volume 4 part 1, page 63: + * + * "If this field (Surface Array) is enabled, the Surface Type must be + * SURFTYPE_1D, SURFTYPE_2D, or SURFTYPE_CUBE. If this field is + * disabled and Surface Type is SURFTYPE_1D, SURFTYPE_2D, or + * SURFTYPE_CUBE, the Depth field must be set to zero." + * + * For non-3D sampler surfaces, resinfo (the sampler message) always + * returns zero for the number of layers when this field is not set. + */ + if (surface_type != GEN6_SURFTYPE_3D) { + if (util_resource_is_array_texture(&tex->base)) + dw[0] |= GEN7_SURFACE_DW0_IS_ARRAY; + else + assert(depth == 1); + } + + assert(tex->layout.align_i == 4 || tex->layout.align_i == 8); + assert(tex->layout.align_j == 2 || tex->layout.align_j == 4); + + if (tex->layout.align_j == 4) + dw[0] |= GEN7_SURFACE_DW0_VALIGN_4; + + if (tex->layout.align_i == 8) + dw[0] |= GEN7_SURFACE_DW0_HALIGN_8; + + if (tex->layout.walk == ILO_LAYOUT_WALK_LOD) + dw[0] |= GEN7_SURFACE_DW0_ARYSPC_LOD0; + else + dw[0] |= GEN7_SURFACE_DW0_ARYSPC_FULL; + + if (is_rt) + dw[0] |= GEN7_SURFACE_DW0_RENDER_CACHE_RW; + + if (surface_type == GEN6_SURFTYPE_CUBE && !is_rt) + dw[0] |= GEN7_SURFACE_DW0_CUBE_FACE_ENABLES__MASK; + + dw[1] = 0; + + dw[2] = GEN_SHIFT32(height - 1, GEN7_SURFACE_DW2_HEIGHT) | + GEN_SHIFT32(width - 1, GEN7_SURFACE_DW2_WIDTH); + + dw[3] = GEN_SHIFT32(depth - 1, GEN7_SURFACE_DW3_DEPTH) | + (pitch - 1); + + dw[4] = first_layer << 18 | + (num_layers - 1) << 7; + + /* + * MSFMT_MSS means the samples are not interleaved and MSFMT_DEPTH_STENCIL + * means the samples are interleaved. The layouts are the same when the + * number of samples is 1. + */ + if (tex->layout.interleaved_samples && tex->base.nr_samples > 1) { + assert(!is_rt); + dw[4] |= GEN7_SURFACE_DW4_MSFMT_DEPTH_STENCIL; + } + else { + dw[4] |= GEN7_SURFACE_DW4_MSFMT_MSS; + } + + if (tex->base.nr_samples > 4) + dw[4] |= GEN7_SURFACE_DW4_MULTISAMPLECOUNT_8; + else if (tex->base.nr_samples > 2) + dw[4] |= GEN7_SURFACE_DW4_MULTISAMPLECOUNT_4; + else + dw[4] |= GEN7_SURFACE_DW4_MULTISAMPLECOUNT_1; + + dw[5] = GEN_SHIFT32(first_level, GEN7_SURFACE_DW5_MIN_LOD) | + lod; + + dw[6] = 0; + dw[7] = 0; + + if (ilo_dev_gen(dev) >= ILO_GEN(7.5)) { + dw[7] |= GEN_SHIFT32(GEN75_SCS_RED, GEN75_SURFACE_DW7_SCS_R) | + GEN_SHIFT32(GEN75_SCS_GREEN, GEN75_SURFACE_DW7_SCS_G) | + GEN_SHIFT32(GEN75_SCS_BLUE, GEN75_SURFACE_DW7_SCS_B) | + GEN_SHIFT32(GEN75_SCS_ALPHA, GEN75_SURFACE_DW7_SCS_A); + } + + /* do not increment reference count */ + surf->bo = tex->bo; +} + +static void +sampler_init_border_color_gen6(const struct ilo_dev_info *dev, + const union pipe_color_union *color, + uint32_t *dw, int num_dwords) +{ + float rgba[4] = { + color->f[0], color->f[1], color->f[2], color->f[3], + }; + + ILO_DEV_ASSERT(dev, 6, 6); + + assert(num_dwords >= 12); + + /* + * This state is not documented in the Sandy Bridge PRM, but in the + * Ironlake PRM. SNORM8 seems to be in DW11 instead of DW1. + */ + + /* IEEE_FP */ + dw[1] = fui(rgba[0]); + dw[2] = fui(rgba[1]); + dw[3] = fui(rgba[2]); + dw[4] = fui(rgba[3]); + + /* FLOAT_16 */ + dw[5] = util_float_to_half(rgba[0]) | + util_float_to_half(rgba[1]) << 16; + dw[6] = util_float_to_half(rgba[2]) | + util_float_to_half(rgba[3]) << 16; + + /* clamp to [-1.0f, 1.0f] */ + rgba[0] = CLAMP(rgba[0], -1.0f, 1.0f); + rgba[1] = CLAMP(rgba[1], -1.0f, 1.0f); + rgba[2] = CLAMP(rgba[2], -1.0f, 1.0f); + rgba[3] = CLAMP(rgba[3], -1.0f, 1.0f); + + /* SNORM16 */ + dw[9] = (int16_t) util_iround(rgba[0] * 32767.0f) | + (int16_t) util_iround(rgba[1] * 32767.0f) << 16; + dw[10] = (int16_t) util_iround(rgba[2] * 32767.0f) | + (int16_t) util_iround(rgba[3] * 32767.0f) << 16; + + /* SNORM8 */ + dw[11] = (int8_t) util_iround(rgba[0] * 127.0f) | + (int8_t) util_iround(rgba[1] * 127.0f) << 8 | + (int8_t) util_iround(rgba[2] * 127.0f) << 16 | + (int8_t) util_iround(rgba[3] * 127.0f) << 24; + + /* clamp to [0.0f, 1.0f] */ + rgba[0] = CLAMP(rgba[0], 0.0f, 1.0f); + rgba[1] = CLAMP(rgba[1], 0.0f, 1.0f); + rgba[2] = CLAMP(rgba[2], 0.0f, 1.0f); + rgba[3] = CLAMP(rgba[3], 0.0f, 1.0f); + + /* UNORM8 */ + dw[0] = (uint8_t) util_iround(rgba[0] * 255.0f) | + (uint8_t) util_iround(rgba[1] * 255.0f) << 8 | + (uint8_t) util_iround(rgba[2] * 255.0f) << 16 | + (uint8_t) util_iround(rgba[3] * 255.0f) << 24; + + /* UNORM16 */ + dw[7] = (uint16_t) util_iround(rgba[0] * 65535.0f) | + (uint16_t) util_iround(rgba[1] * 65535.0f) << 16; + dw[8] = (uint16_t) util_iround(rgba[2] * 65535.0f) | + (uint16_t) util_iround(rgba[3] * 65535.0f) << 16; +} + +/** + * Translate a pipe texture mipfilter to the matching hardware mipfilter. + */ +static int +gen6_translate_tex_mipfilter(unsigned filter) +{ + switch (filter) { + case PIPE_TEX_MIPFILTER_NEAREST: return GEN6_MIPFILTER_NEAREST; + case PIPE_TEX_MIPFILTER_LINEAR: return GEN6_MIPFILTER_LINEAR; + case PIPE_TEX_MIPFILTER_NONE: return GEN6_MIPFILTER_NONE; + default: + assert(!"unknown mipfilter"); + return GEN6_MIPFILTER_NONE; + } +} + +/** + * Translate a pipe texture filter to the matching hardware mapfilter. + */ +static int +gen6_translate_tex_filter(unsigned filter) +{ + switch (filter) { + case PIPE_TEX_FILTER_NEAREST: return GEN6_MAPFILTER_NEAREST; + case PIPE_TEX_FILTER_LINEAR: return GEN6_MAPFILTER_LINEAR; + default: + assert(!"unknown sampler filter"); + return GEN6_MAPFILTER_NEAREST; + } +} + +/** + * Translate a pipe texture coordinate wrapping mode to the matching hardware + * wrapping mode. + */ +static int +gen6_translate_tex_wrap(unsigned wrap, bool clamp_to_edge) +{ + /* clamp to edge or border? */ + if (wrap == PIPE_TEX_WRAP_CLAMP) { + wrap = (clamp_to_edge) ? + PIPE_TEX_WRAP_CLAMP_TO_EDGE : PIPE_TEX_WRAP_CLAMP_TO_BORDER; + } + + switch (wrap) { + case PIPE_TEX_WRAP_REPEAT: return GEN6_TEXCOORDMODE_WRAP; + case PIPE_TEX_WRAP_CLAMP_TO_EDGE: return GEN6_TEXCOORDMODE_CLAMP; + case PIPE_TEX_WRAP_CLAMP_TO_BORDER: return GEN6_TEXCOORDMODE_CLAMP_BORDER; + case PIPE_TEX_WRAP_MIRROR_REPEAT: return GEN6_TEXCOORDMODE_MIRROR; + case PIPE_TEX_WRAP_CLAMP: + case PIPE_TEX_WRAP_MIRROR_CLAMP: + case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE: + case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER: + default: + assert(!"unknown sampler wrap mode"); + return GEN6_TEXCOORDMODE_WRAP; + } +} + +/** + * Translate a pipe shadow compare function to the matching hardware shadow + * function. + */ +static int +gen6_translate_shadow_func(unsigned func) +{ + /* + * For PIPE_FUNC_x, the reference value is on the left-hand side of the + * comparison, and 1.0 is returned when the comparison is true. + * + * For GEN6_COMPAREFUNCTION_x, the reference value is on the right-hand side of + * the comparison, and 0.0 is returned when the comparison is true. + */ + switch (func) { + case PIPE_FUNC_NEVER: return GEN6_COMPAREFUNCTION_ALWAYS; + case PIPE_FUNC_LESS: return GEN6_COMPAREFUNCTION_LEQUAL; + case PIPE_FUNC_EQUAL: return GEN6_COMPAREFUNCTION_NOTEQUAL; + case PIPE_FUNC_LEQUAL: return GEN6_COMPAREFUNCTION_LESS; + case PIPE_FUNC_GREATER: return GEN6_COMPAREFUNCTION_GEQUAL; + case PIPE_FUNC_NOTEQUAL: return GEN6_COMPAREFUNCTION_EQUAL; + case PIPE_FUNC_GEQUAL: return GEN6_COMPAREFUNCTION_GREATER; + case PIPE_FUNC_ALWAYS: return GEN6_COMPAREFUNCTION_NEVER; + default: + assert(!"unknown shadow compare function"); + return GEN6_COMPAREFUNCTION_NEVER; + } +} + +void +ilo_gpe_init_sampler_cso(const struct ilo_dev_info *dev, + const struct pipe_sampler_state *state, + struct ilo_sampler_cso *sampler) +{ + int mip_filter, min_filter, mag_filter, max_aniso; + int lod_bias, max_lod, min_lod; + int wrap_s, wrap_t, wrap_r, wrap_cube; + bool clamp_is_to_edge; + uint32_t dw0, dw1, dw3; + + ILO_DEV_ASSERT(dev, 6, 7.5); + + memset(sampler, 0, sizeof(*sampler)); + + mip_filter = gen6_translate_tex_mipfilter(state->min_mip_filter); + min_filter = gen6_translate_tex_filter(state->min_img_filter); + mag_filter = gen6_translate_tex_filter(state->mag_img_filter); + + sampler->anisotropic = state->max_anisotropy; + + if (state->max_anisotropy >= 2 && state->max_anisotropy <= 16) + max_aniso = state->max_anisotropy / 2 - 1; + else if (state->max_anisotropy > 16) + max_aniso = GEN6_ANISORATIO_16; + else + max_aniso = GEN6_ANISORATIO_2; + + /* + * + * Here is how the hardware calculate per-pixel LOD, from my reading of the + * PRMs: + * + * 1) LOD is set to log2(ratio of texels to pixels) if not specified in + * other ways. The number of texels is measured using level + * SurfMinLod. + * 2) Bias is added to LOD. + * 3) LOD is clamped to [MinLod, MaxLod], and the clamped value is + * compared with Base to determine whether magnification or + * minification is needed. (if preclamp is disabled, LOD is compared + * with Base before clamping) + * 4) If magnification is needed, or no mipmapping is requested, LOD is + * set to floor(MinLod). + * 5) LOD is clamped to [0, MIPCnt], and SurfMinLod is added to LOD. + * + * With Gallium interface, Base is always zero and + * pipe_sampler_view::u.tex.first_level specifies SurfMinLod. + */ + if (ilo_dev_gen(dev) >= ILO_GEN(7)) { + const float scale = 256.0f; + + /* [-16.0, 16.0) in S4.8 */ + lod_bias = (int) + (CLAMP(state->lod_bias, -16.0f, 15.9f) * scale); + lod_bias &= 0x1fff; + + /* [0.0, 14.0] in U4.8 */ + max_lod = (int) (CLAMP(state->max_lod, 0.0f, 14.0f) * scale); + min_lod = (int) (CLAMP(state->min_lod, 0.0f, 14.0f) * scale); + } + else { + const float scale = 64.0f; + + /* [-16.0, 16.0) in S4.6 */ + lod_bias = (int) + (CLAMP(state->lod_bias, -16.0f, 15.9f) * scale); + lod_bias &= 0x7ff; + + /* [0.0, 13.0] in U4.6 */ + max_lod = (int) (CLAMP(state->max_lod, 0.0f, 13.0f) * scale); + min_lod = (int) (CLAMP(state->min_lod, 0.0f, 13.0f) * scale); + } + + /* + * We want LOD to be clamped to determine magnification/minification, and + * get set to zero when it is magnification or when mipmapping is disabled. + * The hardware would set LOD to floor(MinLod) and that is a problem when + * MinLod is greater than or equal to 1.0f. + * + * With Base being zero, it is always minification when MinLod is non-zero. + * To achieve our goal, we just need to set MinLod to zero and set + * MagFilter to MinFilter when mipmapping is disabled. + */ + if (state->min_mip_filter == PIPE_TEX_MIPFILTER_NONE && min_lod) { + min_lod = 0; + mag_filter = min_filter; + } + + /* + * For nearest filtering, PIPE_TEX_WRAP_CLAMP means + * PIPE_TEX_WRAP_CLAMP_TO_EDGE; for linear filtering, PIPE_TEX_WRAP_CLAMP + * means PIPE_TEX_WRAP_CLAMP_TO_BORDER while additionally clamping the + * texture coordinates to [0.0, 1.0]. + * + * The clamping will be taken care of in the shaders. There are two + * filters here, but let the minification one has a say. + */ + clamp_is_to_edge = (state->min_img_filter == PIPE_TEX_FILTER_NEAREST); + if (!clamp_is_to_edge) { + sampler->saturate_s = (state->wrap_s == PIPE_TEX_WRAP_CLAMP); + sampler->saturate_t = (state->wrap_t == PIPE_TEX_WRAP_CLAMP); + sampler->saturate_r = (state->wrap_r == PIPE_TEX_WRAP_CLAMP); + } + + /* determine wrap s/t/r */ + wrap_s = gen6_translate_tex_wrap(state->wrap_s, clamp_is_to_edge); + wrap_t = gen6_translate_tex_wrap(state->wrap_t, clamp_is_to_edge); + wrap_r = gen6_translate_tex_wrap(state->wrap_r, clamp_is_to_edge); + + /* + * From the Sandy Bridge PRM, volume 4 part 1, page 107: + * + * "When using cube map texture coordinates, only TEXCOORDMODE_CLAMP + * and TEXCOORDMODE_CUBE settings are valid, and each TC component + * must have the same Address Control mode." + * + * From the Ivy Bridge PRM, volume 4 part 1, page 96: + * + * "This field (Cube Surface Control Mode) must be set to + * CUBECTRLMODE_PROGRAMMED" + * + * Therefore, we cannot use "Cube Surface Control Mode" for semless cube + * map filtering. + */ + if (state->seamless_cube_map && + (state->min_img_filter != PIPE_TEX_FILTER_NEAREST || + state->mag_img_filter != PIPE_TEX_FILTER_NEAREST)) { + wrap_cube = GEN6_TEXCOORDMODE_CUBE; + } + else { + wrap_cube = GEN6_TEXCOORDMODE_CLAMP; + } + + if (!state->normalized_coords) { + /* + * From the Ivy Bridge PRM, volume 4 part 1, page 98: + * + * "The following state must be set as indicated if this field + * (Non-normalized Coordinate Enable) is enabled: + * + * - TCX/Y/Z Address Control Mode must be TEXCOORDMODE_CLAMP, + * TEXCOORDMODE_HALF_BORDER, or TEXCOORDMODE_CLAMP_BORDER. + * - Surface Type must be SURFTYPE_2D or SURFTYPE_3D. + * - Mag Mode Filter must be MAPFILTER_NEAREST or + * MAPFILTER_LINEAR. + * - Min Mode Filter must be MAPFILTER_NEAREST or + * MAPFILTER_LINEAR. + * - Mip Mode Filter must be MIPFILTER_NONE. + * - Min LOD must be 0. + * - Max LOD must be 0. + * - MIP Count must be 0. + * - Surface Min LOD must be 0. + * - Texture LOD Bias must be 0." + */ + assert(wrap_s == GEN6_TEXCOORDMODE_CLAMP || + wrap_s == GEN6_TEXCOORDMODE_CLAMP_BORDER); + assert(wrap_t == GEN6_TEXCOORDMODE_CLAMP || + wrap_t == GEN6_TEXCOORDMODE_CLAMP_BORDER); + assert(wrap_r == GEN6_TEXCOORDMODE_CLAMP || + wrap_r == GEN6_TEXCOORDMODE_CLAMP_BORDER); + + assert(mag_filter == GEN6_MAPFILTER_NEAREST || + mag_filter == GEN6_MAPFILTER_LINEAR); + assert(min_filter == GEN6_MAPFILTER_NEAREST || + min_filter == GEN6_MAPFILTER_LINEAR); + + /* work around a bug in util_blitter */ + mip_filter = GEN6_MIPFILTER_NONE; + + assert(mip_filter == GEN6_MIPFILTER_NONE); + } + + if (ilo_dev_gen(dev) >= ILO_GEN(7)) { + dw0 = 1 << 28 | + mip_filter << 20 | + lod_bias << 1; + + sampler->dw_filter = mag_filter << 17 | + min_filter << 14; + + sampler->dw_filter_aniso = GEN6_MAPFILTER_ANISOTROPIC << 17 | + GEN6_MAPFILTER_ANISOTROPIC << 14 | + 1; + + dw1 = min_lod << 20 | + max_lod << 8; + + if (state->compare_mode != PIPE_TEX_COMPARE_NONE) + dw1 |= gen6_translate_shadow_func(state->compare_func) << 1; + + dw3 = max_aniso << 19; + + /* round the coordinates for linear filtering */ + if (min_filter != GEN6_MAPFILTER_NEAREST) { + dw3 |= (GEN6_SAMPLER_DW3_U_MIN_ROUND | + GEN6_SAMPLER_DW3_V_MIN_ROUND | + GEN6_SAMPLER_DW3_R_MIN_ROUND); + } + if (mag_filter != GEN6_MAPFILTER_NEAREST) { + dw3 |= (GEN6_SAMPLER_DW3_U_MAG_ROUND | + GEN6_SAMPLER_DW3_V_MAG_ROUND | + GEN6_SAMPLER_DW3_R_MAG_ROUND); + } + + if (!state->normalized_coords) + dw3 |= 1 << 10; + + sampler->dw_wrap = wrap_s << 6 | + wrap_t << 3 | + wrap_r; + + /* + * As noted in the classic i965 driver, the HW may still reference + * wrap_t and wrap_r for 1D textures. We need to set them to a safe + * mode + */ + sampler->dw_wrap_1d = wrap_s << 6 | + GEN6_TEXCOORDMODE_WRAP << 3 | + GEN6_TEXCOORDMODE_WRAP; + + sampler->dw_wrap_cube = wrap_cube << 6 | + wrap_cube << 3 | + wrap_cube; + + STATIC_ASSERT(Elements(sampler->payload) >= 7); + + sampler->payload[0] = dw0; + sampler->payload[1] = dw1; + sampler->payload[2] = dw3; + + memcpy(&sampler->payload[3], + state->border_color.ui, sizeof(state->border_color.ui)); + } + else { + dw0 = 1 << 28 | + mip_filter << 20 | + lod_bias << 3; + + if (state->compare_mode != PIPE_TEX_COMPARE_NONE) + dw0 |= gen6_translate_shadow_func(state->compare_func); + + sampler->dw_filter = (min_filter != mag_filter) << 27 | + mag_filter << 17 | + min_filter << 14; + + sampler->dw_filter_aniso = GEN6_MAPFILTER_ANISOTROPIC << 17 | + GEN6_MAPFILTER_ANISOTROPIC << 14; + + dw1 = min_lod << 22 | + max_lod << 12; + + sampler->dw_wrap = wrap_s << 6 | + wrap_t << 3 | + wrap_r; + + sampler->dw_wrap_1d = wrap_s << 6 | + GEN6_TEXCOORDMODE_WRAP << 3 | + GEN6_TEXCOORDMODE_WRAP; + + sampler->dw_wrap_cube = wrap_cube << 6 | + wrap_cube << 3 | + wrap_cube; + + dw3 = max_aniso << 19; + + /* round the coordinates for linear filtering */ + if (min_filter != GEN6_MAPFILTER_NEAREST) { + dw3 |= (GEN6_SAMPLER_DW3_U_MIN_ROUND | + GEN6_SAMPLER_DW3_V_MIN_ROUND | + GEN6_SAMPLER_DW3_R_MIN_ROUND); + } + if (mag_filter != GEN6_MAPFILTER_NEAREST) { + dw3 |= (GEN6_SAMPLER_DW3_U_MAG_ROUND | + GEN6_SAMPLER_DW3_V_MAG_ROUND | + GEN6_SAMPLER_DW3_R_MAG_ROUND); + } + + if (!state->normalized_coords) + dw3 |= 1; + + STATIC_ASSERT(Elements(sampler->payload) >= 15); + + sampler->payload[0] = dw0; + sampler->payload[1] = dw1; + sampler->payload[2] = dw3; + + sampler_init_border_color_gen6(dev, + &state->border_color, &sampler->payload[3], 12); + } +} diff --git a/src/gallium/drivers/ilo/ilo_state_gen7.c b/src/gallium/drivers/ilo/ilo_state_gen7.c deleted file mode 100644 index f91a088b059..00000000000 --- a/src/gallium/drivers/ilo/ilo_state_gen7.c +++ /dev/null @@ -1,641 +0,0 @@ -/* - * Mesa 3-D graphics library - * - * Copyright (C) 2013 LunarG, Inc. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included - * in all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER - * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING - * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER - * DEALINGS IN THE SOFTWARE. - * - * Authors: - * Chia-I Wu <olv@lunarg.com> - */ - -#include "genhw/genhw.h" -#include "util/u_resource.h" - -#include "ilo_format.h" -#include "ilo_resource.h" -#include "ilo_shader.h" -#include "ilo_state.h" -#include "ilo_state_gen.h" - -void -ilo_gpe_init_gs_cso_gen7(const struct ilo_dev_info *dev, - const struct ilo_shader_state *gs, - struct ilo_shader_cso *cso) -{ - int start_grf, vue_read_len, sampler_count, max_threads; - uint32_t dw2, dw4, dw5; - - ILO_DEV_ASSERT(dev, 7, 7.5); - - start_grf = ilo_shader_get_kernel_param(gs, ILO_KERNEL_URB_DATA_START_REG); - vue_read_len = ilo_shader_get_kernel_param(gs, ILO_KERNEL_INPUT_COUNT); - sampler_count = ilo_shader_get_kernel_param(gs, ILO_KERNEL_SAMPLER_COUNT); - - /* in pairs */ - vue_read_len = (vue_read_len + 1) / 2; - - switch (ilo_dev_gen(dev)) { - case ILO_GEN(7.5): - max_threads = (dev->gt >= 2) ? 256 : 70; - break; - case ILO_GEN(7): - max_threads = (dev->gt == 2) ? 128 : 36; - break; - default: - max_threads = 1; - break; - } - - dw2 = (true) ? 0 : GEN6_THREADDISP_FP_MODE_ALT; - dw2 |= ((sampler_count + 3) / 4) << GEN6_THREADDISP_SAMPLER_COUNT__SHIFT; - - dw4 = vue_read_len << GEN7_GS_DW4_URB_READ_LEN__SHIFT | - GEN7_GS_DW4_INCLUDE_VERTEX_HANDLES | - 0 << GEN7_GS_DW4_URB_READ_OFFSET__SHIFT | - start_grf << GEN7_GS_DW4_URB_GRF_START__SHIFT; - - dw5 = (max_threads - 1) << GEN7_GS_DW5_MAX_THREADS__SHIFT | - GEN7_GS_DW5_STATISTICS | - GEN7_GS_DW5_GS_ENABLE; - - STATIC_ASSERT(Elements(cso->payload) >= 3); - cso->payload[0] = dw2; - cso->payload[1] = dw4; - cso->payload[2] = dw5; -} - -void -ilo_gpe_init_rasterizer_wm_gen7(const struct ilo_dev_info *dev, - const struct pipe_rasterizer_state *state, - struct ilo_rasterizer_wm *wm) -{ - uint32_t dw1, dw2; - - ILO_DEV_ASSERT(dev, 7, 7.5); - - dw1 = GEN7_WM_DW1_ZW_INTERP_PIXEL | - GEN7_WM_DW1_AA_LINE_WIDTH_2_0 | - GEN7_WM_DW1_MSRASTMODE_OFF_PIXEL; - - /* same value as in 3DSTATE_SF */ - if (state->line_smooth) - dw1 |= GEN7_WM_DW1_AA_LINE_CAP_1_0; - - if (state->poly_stipple_enable) - dw1 |= GEN7_WM_DW1_POLY_STIPPLE_ENABLE; - if (state->line_stipple_enable) - dw1 |= GEN7_WM_DW1_LINE_STIPPLE_ENABLE; - - if (state->bottom_edge_rule) - dw1 |= GEN7_WM_DW1_POINT_RASTRULE_UPPER_RIGHT; - - dw2 = GEN7_WM_DW2_MSDISPMODE_PERSAMPLE; - - /* - * assertion that makes sure - * - * dw1 |= wm->dw_msaa_rast; - * dw2 |= wm->dw_msaa_disp; - * - * is valid - */ - STATIC_ASSERT(GEN7_WM_DW1_MSRASTMODE_OFF_PIXEL == 0 && - GEN7_WM_DW2_MSDISPMODE_PERSAMPLE == 0); - - wm->dw_msaa_rast = - (state->multisample) ? GEN7_WM_DW1_MSRASTMODE_ON_PATTERN : 0; - wm->dw_msaa_disp = GEN7_WM_DW2_MSDISPMODE_PERPIXEL; - - STATIC_ASSERT(Elements(wm->payload) >= 2); - wm->payload[0] = dw1; - wm->payload[1] = dw2; -} - -void -ilo_gpe_init_fs_cso_gen7(const struct ilo_dev_info *dev, - const struct ilo_shader_state *fs, - struct ilo_shader_cso *cso) -{ - int start_grf, sampler_count, max_threads; - uint32_t dw2, dw4, dw5; - uint32_t wm_interps, wm_dw1; - - ILO_DEV_ASSERT(dev, 7, 7.5); - - start_grf = ilo_shader_get_kernel_param(fs, ILO_KERNEL_URB_DATA_START_REG); - sampler_count = ilo_shader_get_kernel_param(fs, ILO_KERNEL_SAMPLER_COUNT); - - dw2 = (true) ? 0 : GEN6_THREADDISP_FP_MODE_ALT; - dw2 |= ((sampler_count + 3) / 4) << GEN6_THREADDISP_SAMPLER_COUNT__SHIFT; - - dw4 = GEN7_PS_DW4_POSOFFSET_NONE; - - /* see brwCreateContext() */ - switch (ilo_dev_gen(dev)) { - case ILO_GEN(7.5): - max_threads = (dev->gt == 3) ? 408 : (dev->gt == 2) ? 204 : 102; - dw4 |= (max_threads - 1) << GEN75_PS_DW4_MAX_THREADS__SHIFT; - dw4 |= 1 << GEN75_PS_DW4_SAMPLE_MASK__SHIFT; - break; - case ILO_GEN(7): - default: - max_threads = (dev->gt == 2) ? 172 : 48; - dw4 |= (max_threads - 1) << GEN7_PS_DW4_MAX_THREADS__SHIFT; - break; - } - - if (ilo_shader_get_kernel_param(fs, ILO_KERNEL_PCB_CBUF0_SIZE)) - dw4 |= GEN7_PS_DW4_PUSH_CONSTANT_ENABLE; - - if (ilo_shader_get_kernel_param(fs, ILO_KERNEL_INPUT_COUNT)) - dw4 |= GEN7_PS_DW4_ATTR_ENABLE; - - assert(!ilo_shader_get_kernel_param(fs, ILO_KERNEL_FS_DISPATCH_16_OFFSET)); - dw4 |= GEN7_PS_DW4_8_PIXEL_DISPATCH; - - dw5 = start_grf << GEN7_PS_DW5_URB_GRF_START0__SHIFT | - 0 << GEN7_PS_DW5_URB_GRF_START1__SHIFT | - 0 << GEN7_PS_DW5_URB_GRF_START2__SHIFT; - - /* FS affects 3DSTATE_WM too */ - wm_dw1 = 0; - - /* - * TODO set this bit only when - * - * a) fs writes colors and color is not masked, or - * b) fs writes depth, or - * c) fs or cc kills - */ - wm_dw1 |= GEN7_WM_DW1_PS_ENABLE; - - /* - * From the Ivy Bridge PRM, volume 2 part 1, page 278: - * - * "This bit (Pixel Shader Kill Pixel), if ENABLED, indicates that - * the PS kernel or color calculator has the ability to kill - * (discard) pixels or samples, other than due to depth or stencil - * testing. This bit is required to be ENABLED in the following - * situations: - * - * - The API pixel shader program contains "killpix" or "discard" - * instructions, or other code in the pixel shader kernel that - * can cause the final pixel mask to differ from the pixel mask - * received on dispatch. - * - * - A sampler with chroma key enabled with kill pixel mode is used - * by the pixel shader. - * - * - Any render target has Alpha Test Enable or AlphaToCoverage - * Enable enabled. - * - * - The pixel shader kernel generates and outputs oMask. - * - * Note: As ClipDistance clipping is fully supported in hardware - * and therefore not via PS instructions, there should be no need - * to ENABLE this bit due to ClipDistance clipping." - */ - if (ilo_shader_get_kernel_param(fs, ILO_KERNEL_FS_USE_KILL)) - wm_dw1 |= GEN7_WM_DW1_PS_KILL; - - if (ilo_shader_get_kernel_param(fs, ILO_KERNEL_FS_OUTPUT_Z)) - wm_dw1 |= GEN7_WM_DW1_PSCDEPTH_ON; - - if (ilo_shader_get_kernel_param(fs, ILO_KERNEL_FS_INPUT_Z)) - wm_dw1 |= GEN7_WM_DW1_PS_USE_DEPTH; - - if (ilo_shader_get_kernel_param(fs, ILO_KERNEL_FS_INPUT_W)) - wm_dw1 |= GEN7_WM_DW1_PS_USE_W; - - wm_interps = ilo_shader_get_kernel_param(fs, - ILO_KERNEL_FS_BARYCENTRIC_INTERPOLATIONS); - - wm_dw1 |= wm_interps << GEN7_WM_DW1_BARYCENTRIC_INTERP__SHIFT; - - STATIC_ASSERT(Elements(cso->payload) >= 4); - cso->payload[0] = dw2; - cso->payload[1] = dw4; - cso->payload[2] = dw5; - cso->payload[3] = wm_dw1; -} - -void -ilo_gpe_init_view_surface_null_gen7(const struct ilo_dev_info *dev, - unsigned width, unsigned height, - unsigned depth, unsigned level, - struct ilo_view_surface *surf) -{ - uint32_t *dw; - - ILO_DEV_ASSERT(dev, 7, 7.5); - - assert(width >= 1 && height >= 1 && depth >= 1); - - /* - * From the Ivy Bridge PRM, volume 4 part 1, page 62: - * - * "A null surface is used in instances where an actual surface is not - * bound. When a write message is generated to a null surface, no - * actual surface is written to. When a read message (including any - * sampling engine message) is generated to a null surface, the result - * is all zeros. Note that a null surface type is allowed to be used - * with all messages, even if it is not specificially indicated as - * supported. All of the remaining fields in surface state are ignored - * for null surfaces, with the following exceptions: - * - * * Width, Height, Depth, LOD, and Render Target View Extent fields - * must match the depth buffer's corresponding state for all render - * target surfaces, including null. - * * All sampling engine and data port messages support null surfaces - * with the above behavior, even if not mentioned as specifically - * supported, except for the following: - * * Data Port Media Block Read/Write messages. - * * The Surface Type of a surface used as a render target (accessed - * via the Data Port's Render Target Write message) must be the same - * as the Surface Type of all other render targets and of the depth - * buffer (defined in 3DSTATE_DEPTH_BUFFER), unless either the depth - * buffer or render targets are SURFTYPE_NULL." - * - * From the Ivy Bridge PRM, volume 4 part 1, page 65: - * - * "If Surface Type is SURFTYPE_NULL, this field (Tiled Surface) must be - * true" - */ - - STATIC_ASSERT(Elements(surf->payload) >= 8); - dw = surf->payload; - - dw[0] = GEN6_SURFTYPE_NULL << GEN7_SURFACE_DW0_TYPE__SHIFT | - GEN6_FORMAT_B8G8R8A8_UNORM << GEN7_SURFACE_DW0_FORMAT__SHIFT | - GEN6_TILING_X << 13; - - dw[1] = 0; - - dw[2] = GEN_SHIFT32(height - 1, GEN7_SURFACE_DW2_HEIGHT) | - GEN_SHIFT32(width - 1, GEN7_SURFACE_DW2_WIDTH); - - dw[3] = GEN_SHIFT32(depth - 1, GEN7_SURFACE_DW3_DEPTH); - - dw[4] = 0; - dw[5] = level; - - dw[6] = 0; - dw[7] = 0; - - surf->bo = NULL; -} - -void -ilo_gpe_init_view_surface_for_buffer_gen7(const struct ilo_dev_info *dev, - const struct ilo_buffer *buf, - unsigned offset, unsigned size, - unsigned struct_size, - enum pipe_format elem_format, - bool is_rt, bool render_cache_rw, - struct ilo_view_surface *surf) -{ - const bool typed = (elem_format != PIPE_FORMAT_NONE); - const bool structured = (!typed && struct_size > 1); - const int elem_size = (typed) ? - util_format_get_blocksize(elem_format) : 1; - int width, height, depth, pitch; - int surface_type, surface_format, num_entries; - uint32_t *dw; - - ILO_DEV_ASSERT(dev, 7, 7.5); - - surface_type = (structured) ? GEN7_SURFTYPE_STRBUF : GEN6_SURFTYPE_BUFFER; - - surface_format = (typed) ? - ilo_translate_color_format(dev, elem_format) : GEN6_FORMAT_RAW; - - num_entries = size / struct_size; - /* see if there is enough space to fit another element */ - if (size % struct_size >= elem_size && !structured) - num_entries++; - - /* - * From the Ivy Bridge PRM, volume 4 part 1, page 67: - * - * "For SURFTYPE_BUFFER render targets, this field (Surface Base - * Address) specifies the base address of first element of the - * surface. The surface is interpreted as a simple array of that - * single element type. The address must be naturally-aligned to the - * element size (e.g., a buffer containing R32G32B32A32_FLOAT elements - * must be 16-byte aligned) - * - * For SURFTYPE_BUFFER non-rendertarget surfaces, this field specifies - * the base address of the first element of the surface, computed in - * software by adding the surface base address to the byte offset of - * the element in the buffer." - */ - if (is_rt) - assert(offset % elem_size == 0); - - /* - * From the Ivy Bridge PRM, volume 4 part 1, page 68: - * - * "For typed buffer and structured buffer surfaces, the number of - * entries in the buffer ranges from 1 to 2^27. For raw buffer - * surfaces, the number of entries in the buffer is the number of - * bytes which can range from 1 to 2^30." - */ - assert(num_entries >= 1 && - num_entries <= 1 << ((typed || structured) ? 27 : 30)); - - /* - * From the Ivy Bridge PRM, volume 4 part 1, page 69: - * - * "For SURFTYPE_BUFFER: The low two bits of this field (Width) must be - * 11 if the Surface Format is RAW (the size of the buffer must be a - * multiple of 4 bytes)." - * - * From the Ivy Bridge PRM, volume 4 part 1, page 70: - * - * "For surfaces of type SURFTYPE_BUFFER and SURFTYPE_STRBUF, this - * field (Surface Pitch) indicates the size of the structure." - * - * "For linear surfaces with Surface Type of SURFTYPE_STRBUF, the pitch - * must be a multiple of 4 bytes." - */ - if (structured) - assert(struct_size % 4 == 0); - else if (!typed) - assert(num_entries % 4 == 0); - - pitch = struct_size; - - pitch--; - num_entries--; - /* bits [6:0] */ - width = (num_entries & 0x0000007f); - /* bits [20:7] */ - height = (num_entries & 0x001fff80) >> 7; - /* bits [30:21] */ - depth = (num_entries & 0x7fe00000) >> 21; - /* limit to [26:21] */ - if (typed || structured) - depth &= 0x3f; - - STATIC_ASSERT(Elements(surf->payload) >= 8); - dw = surf->payload; - - dw[0] = surface_type << GEN7_SURFACE_DW0_TYPE__SHIFT | - surface_format << GEN7_SURFACE_DW0_FORMAT__SHIFT; - if (render_cache_rw) - dw[0] |= GEN7_SURFACE_DW0_RENDER_CACHE_RW; - - dw[1] = offset; - - dw[2] = GEN_SHIFT32(height, GEN7_SURFACE_DW2_HEIGHT) | - GEN_SHIFT32(width, GEN7_SURFACE_DW2_WIDTH); - - dw[3] = GEN_SHIFT32(depth, GEN7_SURFACE_DW3_DEPTH) | - pitch; - - dw[4] = 0; - dw[5] = 0; - - dw[6] = 0; - dw[7] = 0; - - if (ilo_dev_gen(dev) >= ILO_GEN(7.5)) { - dw[7] |= GEN_SHIFT32(GEN75_SCS_RED, GEN75_SURFACE_DW7_SCS_R) | - GEN_SHIFT32(GEN75_SCS_GREEN, GEN75_SURFACE_DW7_SCS_G) | - GEN_SHIFT32(GEN75_SCS_BLUE, GEN75_SURFACE_DW7_SCS_B) | - GEN_SHIFT32(GEN75_SCS_ALPHA, GEN75_SURFACE_DW7_SCS_A); - } - - /* do not increment reference count */ - surf->bo = buf->bo; -} - -void -ilo_gpe_init_view_surface_for_texture_gen7(const struct ilo_dev_info *dev, - const struct ilo_texture *tex, - enum pipe_format format, - unsigned first_level, - unsigned num_levels, - unsigned first_layer, - unsigned num_layers, - bool is_rt, - struct ilo_view_surface *surf) -{ - int surface_type, surface_format; - int width, height, depth, pitch, lod; - uint32_t *dw; - - ILO_DEV_ASSERT(dev, 7, 7.5); - - surface_type = ilo_gpe_gen6_translate_texture(tex->base.target); - assert(surface_type != GEN6_SURFTYPE_BUFFER); - - if (format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT && tex->separate_s8) - format = PIPE_FORMAT_Z32_FLOAT; - - if (is_rt) - surface_format = ilo_translate_render_format(dev, format); - else - surface_format = ilo_translate_texture_format(dev, format); - assert(surface_format >= 0); - - width = tex->layout.width0; - height = tex->layout.height0; - depth = (tex->base.target == PIPE_TEXTURE_3D) ? - tex->base.depth0 : num_layers; - pitch = tex->layout.bo_stride; - - if (surface_type == GEN6_SURFTYPE_CUBE) { - /* - * From the Ivy Bridge PRM, volume 4 part 1, page 70: - * - * "For SURFTYPE_CUBE:For Sampling Engine Surfaces, the range of - * this field is [0,340], indicating the number of cube array - * elements (equal to the number of underlying 2D array elements - * divided by 6). For other surfaces, this field must be zero." - * - * When is_rt is true, we treat the texture as a 2D one to avoid the - * restriction. - */ - if (is_rt) { - surface_type = GEN6_SURFTYPE_2D; - } - else { - assert(num_layers % 6 == 0); - depth = num_layers / 6; - } - } - - /* sanity check the size */ - assert(width >= 1 && height >= 1 && depth >= 1 && pitch >= 1); - assert(first_layer < 2048 && num_layers <= 2048); - switch (surface_type) { - case GEN6_SURFTYPE_1D: - assert(width <= 16384 && height == 1 && depth <= 2048); - break; - case GEN6_SURFTYPE_2D: - assert(width <= 16384 && height <= 16384 && depth <= 2048); - break; - case GEN6_SURFTYPE_3D: - assert(width <= 2048 && height <= 2048 && depth <= 2048); - if (!is_rt) - assert(first_layer == 0); - break; - case GEN6_SURFTYPE_CUBE: - assert(width <= 16384 && height <= 16384 && depth <= 86); - assert(width == height); - if (is_rt) - assert(first_layer == 0); - break; - default: - assert(!"unexpected surface type"); - break; - } - - if (is_rt) { - assert(num_levels == 1); - lod = first_level; - } - else { - lod = num_levels - 1; - } - - /* - * From the Ivy Bridge PRM, volume 4 part 1, page 68: - * - * "The Base Address for linear render target surfaces and surfaces - * accessed with the typed surface read/write data port messages must - * be element-size aligned, for non-YUV surface formats, or a multiple - * of 2 element-sizes for YUV surface formats. Other linear surfaces - * have no alignment requirements (byte alignment is sufficient)." - * - * From the Ivy Bridge PRM, volume 4 part 1, page 70: - * - * "For linear render target surfaces and surfaces accessed with the - * typed data port messages, the pitch must be a multiple of the - * element size for non-YUV surface formats. Pitch must be a multiple - * of 2 * element size for YUV surface formats. For linear surfaces - * with Surface Type of SURFTYPE_STRBUF, the pitch must be a multiple - * of 4 bytes.For other linear surfaces, the pitch can be any multiple - * of bytes." - * - * From the Ivy Bridge PRM, volume 4 part 1, page 74: - * - * "For linear surfaces, this field (X Offset) must be zero." - */ - if (tex->layout.tiling == INTEL_TILING_NONE) { - if (is_rt) { - const int elem_size = util_format_get_blocksize(format); - assert(pitch % elem_size == 0); - } - } - - STATIC_ASSERT(Elements(surf->payload) >= 8); - dw = surf->payload; - - dw[0] = surface_type << GEN7_SURFACE_DW0_TYPE__SHIFT | - surface_format << GEN7_SURFACE_DW0_FORMAT__SHIFT | - ilo_gpe_gen6_translate_winsys_tiling(tex->layout.tiling) << 13; - - /* - * From the Ivy Bridge PRM, volume 4 part 1, page 63: - * - * "If this field (Surface Array) is enabled, the Surface Type must be - * SURFTYPE_1D, SURFTYPE_2D, or SURFTYPE_CUBE. If this field is - * disabled and Surface Type is SURFTYPE_1D, SURFTYPE_2D, or - * SURFTYPE_CUBE, the Depth field must be set to zero." - * - * For non-3D sampler surfaces, resinfo (the sampler message) always - * returns zero for the number of layers when this field is not set. - */ - if (surface_type != GEN6_SURFTYPE_3D) { - if (util_resource_is_array_texture(&tex->base)) - dw[0] |= GEN7_SURFACE_DW0_IS_ARRAY; - else - assert(depth == 1); - } - - assert(tex->layout.align_i == 4 || tex->layout.align_i == 8); - assert(tex->layout.align_j == 2 || tex->layout.align_j == 4); - - if (tex->layout.align_j == 4) - dw[0] |= GEN7_SURFACE_DW0_VALIGN_4; - - if (tex->layout.align_i == 8) - dw[0] |= GEN7_SURFACE_DW0_HALIGN_8; - - if (tex->layout.walk == ILO_LAYOUT_WALK_LOD) - dw[0] |= GEN7_SURFACE_DW0_ARYSPC_LOD0; - else - dw[0] |= GEN7_SURFACE_DW0_ARYSPC_FULL; - - if (is_rt) - dw[0] |= GEN7_SURFACE_DW0_RENDER_CACHE_RW; - - if (surface_type == GEN6_SURFTYPE_CUBE && !is_rt) - dw[0] |= GEN7_SURFACE_DW0_CUBE_FACE_ENABLES__MASK; - - dw[1] = 0; - - dw[2] = GEN_SHIFT32(height - 1, GEN7_SURFACE_DW2_HEIGHT) | - GEN_SHIFT32(width - 1, GEN7_SURFACE_DW2_WIDTH); - - dw[3] = GEN_SHIFT32(depth - 1, GEN7_SURFACE_DW3_DEPTH) | - (pitch - 1); - - dw[4] = first_layer << 18 | - (num_layers - 1) << 7; - - /* - * MSFMT_MSS means the samples are not interleaved and MSFMT_DEPTH_STENCIL - * means the samples are interleaved. The layouts are the same when the - * number of samples is 1. - */ - if (tex->layout.interleaved_samples && tex->base.nr_samples > 1) { - assert(!is_rt); - dw[4] |= GEN7_SURFACE_DW4_MSFMT_DEPTH_STENCIL; - } - else { - dw[4] |= GEN7_SURFACE_DW4_MSFMT_MSS; - } - - if (tex->base.nr_samples > 4) - dw[4] |= GEN7_SURFACE_DW4_MULTISAMPLECOUNT_8; - else if (tex->base.nr_samples > 2) - dw[4] |= GEN7_SURFACE_DW4_MULTISAMPLECOUNT_4; - else - dw[4] |= GEN7_SURFACE_DW4_MULTISAMPLECOUNT_1; - - dw[5] = GEN_SHIFT32(first_level, GEN7_SURFACE_DW5_MIN_LOD) | - lod; - - dw[6] = 0; - dw[7] = 0; - - if (ilo_dev_gen(dev) >= ILO_GEN(7.5)) { - dw[7] |= GEN_SHIFT32(GEN75_SCS_RED, GEN75_SURFACE_DW7_SCS_R) | - GEN_SHIFT32(GEN75_SCS_GREEN, GEN75_SURFACE_DW7_SCS_G) | - GEN_SHIFT32(GEN75_SCS_BLUE, GEN75_SURFACE_DW7_SCS_B) | - GEN_SHIFT32(GEN75_SCS_ALPHA, GEN75_SURFACE_DW7_SCS_A); - } - - /* do not increment reference count */ - surf->bo = tex->bo; -} |