X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fauxiliary%2Fvl%2Fvl_idct.c;h=79adb045dad4ed6cf21dee3447c28c72e8a047b3;hb=a3ed98f7aa85636579a5696bf036ec13e5c9104a;hp=b418aea951409728aa205f068a743e24d3bfb58d;hpb=310eea52ca1e997295c84163066cc5d0fd4f8cf6;p=mesa.git diff --git a/src/gallium/auxiliary/vl/vl_idct.c b/src/gallium/auxiliary/vl/vl_idct.c index b418aea9514..79adb045dad 100644 --- a/src/gallium/auxiliary/vl/vl_idct.c +++ b/src/gallium/auxiliary/vl/vl_idct.c @@ -25,43 +25,47 @@ * **************************************************************************/ -#include "vl_idct.h" -#include "vl_vertex_buffers.h" -#include "util/u_draw.h" #include -#include -#include -#include -#include -#include -#include -#include "vl_types.h" -#define BLOCK_WIDTH 8 -#define BLOCK_HEIGHT 8 +#include "pipe/p_context.h" +#include "pipe/p_screen.h" -#define SCALE_FACTOR_16_TO_9 (32768.0f / 256.0f) +#include "util/u_draw.h" +#include "util/u_sampler.h" +#include "util/u_memory.h" -#define NR_RENDER_TARGETS 4 +#include "tgsi/tgsi_ureg.h" + +#include "vl_defines.h" +#include "vl_types.h" +#include "vl_vertex_buffers.h" +#include "vl_idct.h" enum VS_OUTPUT { - VS_O_VPOS, - VS_O_L_ADDR0, + VS_O_VPOS = 0, + VS_O_L_ADDR0 = 0, VS_O_L_ADDR1, VS_O_R_ADDR0, VS_O_R_ADDR1 }; -static const float const_matrix[8][8] = { - { 0.3535530f, 0.3535530f, 0.3535530f, 0.3535530f, 0.3535530f, 0.3535530f, 0.353553f, 0.3535530f }, - { 0.4903930f, 0.4157350f, 0.2777850f, 0.0975451f, -0.0975452f, -0.2777850f, -0.415735f, -0.4903930f }, - { 0.4619400f, 0.1913420f, -0.1913420f, -0.4619400f, -0.4619400f, -0.1913420f, 0.191342f, 0.4619400f }, - { 0.4157350f, -0.0975452f, -0.4903930f, -0.2777850f, 0.2777850f, 0.4903930f, 0.097545f, -0.4157350f }, - { 0.3535530f, -0.3535530f, -0.3535530f, 0.3535540f, 0.3535530f, -0.3535540f, -0.353553f, 0.3535530f }, - { 0.2777850f, -0.4903930f, 0.0975452f, 0.4157350f, -0.4157350f, -0.0975451f, 0.490393f, -0.2777850f }, - { 0.1913420f, -0.4619400f, 0.4619400f, -0.1913420f, -0.1913410f, 0.4619400f, -0.461940f, 0.1913420f }, - { 0.0975451f, -0.2777850f, 0.4157350f, -0.4903930f, 0.4903930f, -0.4157350f, 0.277786f, -0.0975458f } +/** + * The DCT matrix stored as hex representation of floats. Equal to the following equation: + * for (i = 0; i < 8; ++i) + * for (j = 0; j < 8; ++j) + * if (i == 0) const_matrix[i][j] = 1.0f / sqrtf(8.0f); + * else const_matrix[i][j] = sqrtf(2.0f / 8.0f) * cosf((2 * j + 1) * i * M_PI / (2.0f * 8.0f)); + */ +static const uint32_t const_matrix[8][8] = { + { 0x3eb504f3, 0x3eb504f3, 0x3eb504f3, 0x3eb504f3, 0x3eb504f3, 0x3eb504f3, 0x3eb504f3, 0x3eb504f3 }, + { 0x3efb14be, 0x3ed4db31, 0x3e8e39da, 0x3dc7c5c4, 0xbdc7c5c2, 0xbe8e39d9, 0xbed4db32, 0xbefb14bf }, + { 0x3eec835f, 0x3e43ef15, 0xbe43ef14, 0xbeec835e, 0xbeec835f, 0xbe43ef1a, 0x3e43ef1b, 0x3eec835f }, + { 0x3ed4db31, 0xbdc7c5c2, 0xbefb14bf, 0xbe8e39dd, 0x3e8e39d7, 0x3efb14bf, 0x3dc7c5d0, 0xbed4db34 }, + { 0x3eb504f3, 0xbeb504f3, 0xbeb504f4, 0x3eb504f1, 0x3eb504f3, 0xbeb504f0, 0xbeb504ef, 0x3eb504f4 }, + { 0x3e8e39da, 0xbefb14bf, 0x3dc7c5c8, 0x3ed4db32, 0xbed4db34, 0xbdc7c5bb, 0x3efb14bf, 0xbe8e39d7 }, + { 0x3e43ef15, 0xbeec835f, 0x3eec835f, 0xbe43ef07, 0xbe43ef23, 0x3eec8361, 0xbeec835c, 0x3e43ef25 }, + { 0x3dc7c5c4, 0xbe8e39dd, 0x3ed4db32, 0xbefb14c0, 0x3efb14be, 0xbed4db31, 0x3e8e39ce, 0xbdc7c596 }, }; static void @@ -83,194 +87,261 @@ calc_addr(struct ureg_program *shader, struct ureg_dst addr[2], */ ureg_MOV(shader, ureg_writemask(addr[0], wm_start), ureg_scalar(start, sw_start)); ureg_MOV(shader, ureg_writemask(addr[0], wm_tc), ureg_scalar(tc, sw_tc)); - ureg_MOV(shader, ureg_writemask(addr[0], TGSI_WRITEMASK_Z), tc); ureg_ADD(shader, ureg_writemask(addr[1], wm_start), ureg_scalar(start, sw_start), ureg_imm1f(shader, 1.0f / size)); ureg_MOV(shader, ureg_writemask(addr[1], wm_tc), ureg_scalar(tc, sw_tc)); - ureg_MOV(shader, ureg_writemask(addr[1], TGSI_WRITEMASK_Z), tc); +} + +static void +increment_addr(struct ureg_program *shader, struct ureg_dst daddr[2], + struct ureg_src saddr[2], bool right_side, bool transposed, + int pos, float size) +{ + unsigned wm_start = (right_side == transposed) ? TGSI_WRITEMASK_X : TGSI_WRITEMASK_Y; + unsigned wm_tc = (right_side == transposed) ? TGSI_WRITEMASK_Y : TGSI_WRITEMASK_X; + + /* + * daddr[0..1].(start) = saddr[0..1].(start) + * daddr[0..1].(tc) = saddr[0..1].(tc) + */ + + ureg_MOV(shader, ureg_writemask(daddr[0], wm_start), saddr[0]); + ureg_ADD(shader, ureg_writemask(daddr[0], wm_tc), saddr[0], ureg_imm1f(shader, pos / size)); + ureg_MOV(shader, ureg_writemask(daddr[1], wm_start), saddr[1]); + ureg_ADD(shader, ureg_writemask(daddr[1], wm_tc), saddr[1], ureg_imm1f(shader, pos / size)); +} + +static void +fetch_four(struct ureg_program *shader, struct ureg_dst m[2], struct ureg_src addr[2], + struct ureg_src sampler, bool resource3d) +{ + ureg_TEX(shader, m[0], resource3d ? TGSI_TEXTURE_3D : TGSI_TEXTURE_2D, addr[0], sampler); + ureg_TEX(shader, m[1], resource3d ? TGSI_TEXTURE_3D : TGSI_TEXTURE_2D, addr[1], sampler); +} + +static void +matrix_mul(struct ureg_program *shader, struct ureg_dst dst, struct ureg_dst l[2], struct ureg_dst r[2]) +{ + struct ureg_dst tmp; + + tmp = ureg_DECL_temporary(shader); + + /* + * tmp.xy = dot4(m[0][0..1], m[1][0..1]) + * dst = tmp.x + tmp.y + */ + ureg_DP4(shader, ureg_writemask(tmp, TGSI_WRITEMASK_X), ureg_src(l[0]), ureg_src(r[0])); + ureg_DP4(shader, ureg_writemask(tmp, TGSI_WRITEMASK_Y), ureg_src(l[1]), ureg_src(r[1])); + ureg_ADD(shader, dst, + ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_X), + ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_Y)); + + ureg_release_temporary(shader, tmp); } static void * -create_vert_shader(struct vl_idct *idct, bool matrix_stage, int color_swizzle) +create_mismatch_vert_shader(struct vl_idct *idct) { struct ureg_program *shader; - struct ureg_src vrect, vpos, vblock, eb[4]; - struct ureg_src scale, blocks_xy, t_eb; - struct ureg_dst t_tex, t_start; - struct ureg_dst o_vpos, o_l_addr[2], o_r_addr[2]; - unsigned label; + struct ureg_src vpos; + struct ureg_src scale; + struct ureg_dst t_tex; + struct ureg_dst o_vpos, o_addr[2]; shader = ureg_create(TGSI_PROCESSOR_VERTEX); if (!shader) return NULL; - t_tex = ureg_DECL_temporary(shader); - t_start = ureg_DECL_temporary(shader); - - vrect = ureg_DECL_vs_input(shader, VS_I_RECT); vpos = ureg_DECL_vs_input(shader, VS_I_VPOS); - vblock = ureg_swizzle(vrect, TGSI_SWIZZLE_Z, TGSI_SWIZZLE_W, TGSI_SWIZZLE_X, TGSI_SWIZZLE_X); - o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS); - - eb[0] = ureg_DECL_vs_input(shader, VS_I_EB_0_0); - eb[1] = ureg_DECL_vs_input(shader, VS_I_EB_1_0); - eb[2] = ureg_DECL_vs_input(shader, VS_I_EB_0_1); - eb[3] = ureg_DECL_vs_input(shader, VS_I_EB_1_1); + t_tex = ureg_DECL_temporary(shader); - o_l_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0); - o_l_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1); + o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS); - o_r_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR0); - o_r_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR1); + o_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0); + o_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1); /* - * scale = (BLOCK_WIDTH, BLOCK_HEIGHT) / (dst.width, dst.height) - * blocks_xy = (blocks_x, blocks_y) + * scale = (VL_BLOCK_WIDTH, VL_BLOCK_HEIGHT) / (dst.width, dst.height) * - * ar = vblock.y * blocks.x + vblock.x - * if eb[ar].(color_swizzle) - * o_vpos.xy = -1 - * else - * t_tex = vpos * blocks_xy + vblock - * t_start = t_tex * scale - * t_tex = t_tex + vrect - * o_vpos.xy = t_tex * scale + * t_vpos = vpos + 7 / VL_BLOCK_WIDTH + * o_vpos.xy = t_vpos * scale * - * o_l_addr = calc_addr(...) - * o_r_addr = calc_addr(...) - * endif - * o_vpos.zw = vpos + * o_addr = calc_addr(...) * */ scale = ureg_imm2f(shader, - (float)BLOCK_WIDTH / idct->buffer_width, - (float)BLOCK_HEIGHT / idct->buffer_height); + (float)VL_BLOCK_WIDTH / idct->buffer_width, + (float)VL_BLOCK_HEIGHT / idct->buffer_height); - blocks_xy = ureg_imm2f(shader, idct->blocks_x, idct->blocks_y); + ureg_MAD(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_XY), vpos, scale, scale); + ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_ZW), ureg_imm1f(shader, 1.0f)); - if (idct->blocks_x > 1 || idct->blocks_y > 1) { - struct ureg_dst ar = ureg_DECL_address(shader); + ureg_MUL(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), vpos, scale); + calc_addr(shader, o_addr, ureg_src(t_tex), ureg_src(t_tex), false, false, idct->buffer_width / 4); - ureg_MAD(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_X), - ureg_scalar(vblock, TGSI_SWIZZLE_Y), blocks_xy, vblock); + ureg_release_temporary(shader, t_tex); - ureg_ARL(shader, ureg_writemask(ar, TGSI_WRITEMASK_X), ureg_src(t_tex)); - t_eb = ureg_src_indirect(eb[0], ureg_src(ar)); - } else { - t_eb = eb[0]; - } + ureg_END(shader); + + return ureg_create_shader_and_destroy(shader, idct->pipe); +} - ureg_IF(shader, ureg_scalar(t_eb, color_swizzle), &label); +static void * +create_mismatch_frag_shader(struct vl_idct *idct) +{ + struct ureg_program *shader; - ureg_MOV(shader, o_vpos, ureg_imm1f(shader, -1.0f)); + struct ureg_src addr[2]; - ureg_fixup_label(shader, label, ureg_get_instruction_number(shader)); - ureg_ELSE(shader, &label); + struct ureg_dst m[8][2]; + struct ureg_dst fragment; - ureg_MAD(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), vpos, blocks_xy, vblock); - ureg_MUL(shader, ureg_writemask(t_start, TGSI_WRITEMASK_XY), ureg_src(t_tex), scale); + unsigned i; - ureg_ADD(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), ureg_src(t_tex), vrect); + shader = ureg_create(TGSI_PROCESSOR_FRAGMENT); + if (!shader) + return NULL; - ureg_MUL(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), ureg_src(t_tex), scale); - ureg_MUL(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_Z), - ureg_scalar(vrect, TGSI_SWIZZLE_X), - ureg_imm1f(shader, BLOCK_WIDTH / NR_RENDER_TARGETS)); + addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0, TGSI_INTERPOLATE_LINEAR); + addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1, TGSI_INTERPOLATE_LINEAR); - ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_XY), ureg_src(t_tex)); + fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0); - if(matrix_stage) { - calc_addr(shader, o_l_addr, ureg_src(t_tex), ureg_src(t_start), false, false, idct->buffer_width / 4); - calc_addr(shader, o_r_addr, vrect, ureg_imm1f(shader, 0.0f), true, true, BLOCK_WIDTH / 4); - } else { - calc_addr(shader, o_l_addr, vrect, ureg_imm1f(shader, 0.0f), false, false, BLOCK_WIDTH / 4); - calc_addr(shader, o_r_addr, ureg_src(t_tex), ureg_src(t_start), true, false, idct->buffer_height / 4); - } + for (i = 0; i < 8; ++i) { + m[i][0] = ureg_DECL_temporary(shader); + m[i][1] = ureg_DECL_temporary(shader); + } - ureg_fixup_label(shader, label, ureg_get_instruction_number(shader)); - ureg_ENDIF(shader); + for (i = 0; i < 8; ++i) { + increment_addr(shader, m[i], addr, false, false, i, idct->buffer_height); + } - ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_ZW), vpos); + for (i = 0; i < 8; ++i) { + struct ureg_src s_addr[2]; + s_addr[0] = ureg_src(m[i][0]); + s_addr[1] = ureg_src(m[i][1]); + fetch_four(shader, m[i], s_addr, ureg_DECL_sampler(shader, 0), false); + } - ureg_release_temporary(shader, t_tex); - ureg_release_temporary(shader, t_start); + for (i = 1; i < 8; ++i) { + ureg_ADD(shader, m[0][0], ureg_src(m[0][0]), ureg_src(m[i][0])); + ureg_ADD(shader, m[0][1], ureg_src(m[0][1]), ureg_src(m[i][1])); + } + + ureg_ADD(shader, m[0][0], ureg_src(m[0][0]), ureg_src(m[0][1])); + ureg_DP4(shader, m[0][0], ureg_abs(ureg_src(m[0][0])), ureg_imm1f(shader, 1 << 14)); + + ureg_MUL(shader, ureg_writemask(m[0][0], TGSI_WRITEMASK_W), ureg_abs(ureg_src(m[7][1])), ureg_imm1f(shader, 1 << 14)); + ureg_FRC(shader, m[0][0], ureg_src(m[0][0])); + ureg_SGT(shader, m[0][0], ureg_imm1f(shader, 0.5f), ureg_abs(ureg_src(m[0][0]))); + + ureg_CMP(shader, ureg_writemask(m[0][0], TGSI_WRITEMASK_W), ureg_negate(ureg_src(m[0][0])), + ureg_imm1f(shader, 1.0f / (1 << 15)), ureg_imm1f(shader, -1.0f / (1 << 15))); + ureg_MUL(shader, ureg_writemask(m[0][0], TGSI_WRITEMASK_W), ureg_src(m[0][0]), + ureg_scalar(ureg_src(m[0][0]), TGSI_SWIZZLE_X)); + + ureg_MOV(shader, ureg_writemask(fragment, TGSI_WRITEMASK_XYZ), ureg_src(m[7][1])); + ureg_ADD(shader, ureg_writemask(fragment, TGSI_WRITEMASK_W), ureg_src(m[0][0]), ureg_src(m[7][1])); + + for (i = 0; i < 8; ++i) { + ureg_release_temporary(shader, m[i][0]); + ureg_release_temporary(shader, m[i][1]); + } ureg_END(shader); return ureg_create_shader_and_destroy(shader, idct->pipe); } -static void -increment_addr(struct ureg_program *shader, struct ureg_dst daddr[2], - struct ureg_src saddr[2], bool right_side, bool transposed, - int pos, float size) +static void * +create_stage1_vert_shader(struct vl_idct *idct) { - unsigned wm_start = (right_side == transposed) ? TGSI_WRITEMASK_X : TGSI_WRITEMASK_Y; - unsigned wm_tc = (right_side == transposed) ? TGSI_WRITEMASK_Y : TGSI_WRITEMASK_X; + struct ureg_program *shader; + struct ureg_src vrect, vpos; + struct ureg_src scale; + struct ureg_dst t_tex, t_start; + struct ureg_dst o_vpos, o_l_addr[2], o_r_addr[2]; - /* - * daddr[0..1].(start) = saddr[0..1].(start) - * daddr[0..1].(tc) = saddr[0..1].(tc) - */ + shader = ureg_create(TGSI_PROCESSOR_VERTEX); + if (!shader) + return NULL; - ureg_MOV(shader, ureg_writemask(daddr[0], wm_start), saddr[0]); - ureg_ADD(shader, ureg_writemask(daddr[0], wm_tc), saddr[0], ureg_imm1f(shader, pos / size)); - ureg_MOV(shader, ureg_writemask(daddr[1], wm_start), saddr[1]); - ureg_ADD(shader, ureg_writemask(daddr[1], wm_tc), saddr[1], ureg_imm1f(shader, pos / size)); -} + vrect = ureg_DECL_vs_input(shader, VS_I_RECT); + vpos = ureg_DECL_vs_input(shader, VS_I_VPOS); -static void -fetch_four(struct ureg_program *shader, struct ureg_dst m[2], struct ureg_src addr[2], struct ureg_src sampler) -{ - ureg_TEX(shader, m[0], TGSI_TEXTURE_3D, addr[0], sampler); - ureg_TEX(shader, m[1], TGSI_TEXTURE_3D, addr[1], sampler); -} + t_tex = ureg_DECL_temporary(shader); + t_start = ureg_DECL_temporary(shader); -static void -matrix_mul(struct ureg_program *shader, struct ureg_dst dst, struct ureg_dst l[2], struct ureg_dst r[2]) -{ - struct ureg_dst tmp; + o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS); - tmp = ureg_DECL_temporary(shader); + o_l_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0); + o_l_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1); + + o_r_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR0); + o_r_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR1); /* - * tmp.xy = dot4(m[0][0..1], m[1][0..1]) - * dst = tmp.x + tmp.y + * scale = (VL_BLOCK_WIDTH, VL_BLOCK_HEIGHT) / (dst.width, dst.height) + * + * t_vpos = vpos + vrect + * o_vpos.xy = t_vpos * scale + * o_vpos.zw = vpos + * + * o_l_addr = calc_addr(...) + * o_r_addr = calc_addr(...) + * */ - ureg_DP4(shader, ureg_writemask(tmp, TGSI_WRITEMASK_X), ureg_src(l[0]), ureg_src(r[0])); - ureg_DP4(shader, ureg_writemask(tmp, TGSI_WRITEMASK_Y), ureg_src(l[1]), ureg_src(r[1])); - ureg_ADD(shader, dst, - ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_X), - ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_Y)); - ureg_release_temporary(shader, tmp); + scale = ureg_imm2f(shader, + (float)VL_BLOCK_WIDTH / idct->buffer_width, + (float)VL_BLOCK_HEIGHT / idct->buffer_height); + + ureg_ADD(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), vpos, vrect); + ureg_MUL(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), ureg_src(t_tex), scale); + + ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_XY), ureg_src(t_tex)); + ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_ZW), ureg_imm1f(shader, 1.0f)); + + ureg_MUL(shader, ureg_writemask(t_start, TGSI_WRITEMASK_XY), vpos, scale); + + calc_addr(shader, o_l_addr, ureg_src(t_tex), ureg_src(t_start), false, false, idct->buffer_width / 4); + calc_addr(shader, o_r_addr, vrect, ureg_imm1f(shader, 0.0f), true, true, VL_BLOCK_WIDTH / 4); + + ureg_release_temporary(shader, t_tex); + ureg_release_temporary(shader, t_start); + + ureg_END(shader); + + return ureg_create_shader_and_destroy(shader, idct->pipe); } static void * -create_matrix_frag_shader(struct vl_idct *idct) +create_stage1_frag_shader(struct vl_idct *idct) { struct ureg_program *shader; struct ureg_src l_addr[2], r_addr[2]; struct ureg_dst l[4][2], r[2]; - struct ureg_dst fragment[NR_RENDER_TARGETS]; + struct ureg_dst *fragment; - unsigned i, j; + int i, j; shader = ureg_create(TGSI_PROCESSOR_FRAGMENT); if (!shader) return NULL; + fragment = MALLOC(idct->nr_of_render_targets * sizeof(struct ureg_dst)); + l_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0, TGSI_INTERPOLATE_LINEAR); l_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1, TGSI_INTERPOLATE_LINEAR); r_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR0, TGSI_INTERPOLATE_LINEAR); r_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR1, TGSI_INTERPOLATE_LINEAR); - for (i = 0; i < NR_RENDER_TARGETS; ++i) + for (i = 0; i < idct->nr_of_render_targets; ++i) fragment[i] = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, i); for (i = 0; i < 4; ++i) { @@ -281,25 +352,25 @@ create_matrix_frag_shader(struct vl_idct *idct) r[0] = ureg_DECL_temporary(shader); r[1] = ureg_DECL_temporary(shader); - for (i = 1; i < 4; ++i) { - increment_addr(shader, l[i], l_addr, false, false, i, idct->buffer_height); + for (i = 0; i < 4; ++i) { + increment_addr(shader, l[i], l_addr, false, false, i - 2, idct->buffer_height); } for (i = 0; i < 4; ++i) { struct ureg_src s_addr[2]; - s_addr[0] = i == 0 ? l_addr[0] : ureg_src(l[i][0]); - s_addr[1] = i == 0 ? l_addr[1] : ureg_src(l[i][1]); - fetch_four(shader, l[i], s_addr, ureg_DECL_sampler(shader, 1)); + s_addr[0] = ureg_src(l[i][0]); + s_addr[1] = ureg_src(l[i][1]); + fetch_four(shader, l[i], s_addr, ureg_DECL_sampler(shader, 0), false); } - for (i = 0; i < NR_RENDER_TARGETS; ++i) { - if(i > 0) - increment_addr(shader, r, r_addr, true, true, i, BLOCK_HEIGHT); + for (i = 0; i < idct->nr_of_render_targets; ++i) { + struct ureg_src s_addr[2]; + + increment_addr(shader, r, r_addr, true, true, i - (signed)idct->nr_of_render_targets / 2, VL_BLOCK_HEIGHT); - struct ureg_src s_addr[2] = { ureg_src(r[0]), ureg_src(r[1]) }; - s_addr[0] = i == 0 ? r_addr[0] : ureg_src(r[0]); - s_addr[1] = i == 0 ? r_addr[1] : ureg_src(r[1]); - fetch_four(shader, r, s_addr, ureg_DECL_sampler(shader, 0)); + s_addr[0] = ureg_src(r[0]); + s_addr[1] = ureg_src(r[1]); + fetch_four(shader, r, s_addr, ureg_DECL_sampler(shader, 1), false); for (j = 0; j < 4; ++j) { matrix_mul(shader, ureg_writemask(fragment[i], TGSI_WRITEMASK_X << j), l[j], r); @@ -315,91 +386,163 @@ create_matrix_frag_shader(struct vl_idct *idct) ureg_END(shader); + FREE(fragment); + return ureg_create_shader_and_destroy(shader, idct->pipe); } -static void * -create_transpose_frag_shader(struct vl_idct *idct) +void +vl_idct_stage2_vert_shader(struct vl_idct *idct, struct ureg_program *shader, + unsigned first_output, struct ureg_dst tex) { - struct ureg_program *shader; + struct ureg_src vrect, vpos; + struct ureg_src scale; + struct ureg_dst t_start; + struct ureg_dst o_l_addr[2], o_r_addr[2]; + + vrect = ureg_DECL_vs_input(shader, VS_I_RECT); + vpos = ureg_DECL_vs_input(shader, VS_I_VPOS); + + t_start = ureg_DECL_temporary(shader); + + --first_output; + + o_l_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, first_output + VS_O_L_ADDR0); + o_l_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, first_output + VS_O_L_ADDR1); + + o_r_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, first_output + VS_O_R_ADDR0); + o_r_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, first_output + VS_O_R_ADDR1); + + scale = ureg_imm2f(shader, + (float)VL_BLOCK_WIDTH / idct->buffer_width, + (float)VL_BLOCK_HEIGHT / idct->buffer_height); + + ureg_MUL(shader, ureg_writemask(tex, TGSI_WRITEMASK_Z), + ureg_scalar(vrect, TGSI_SWIZZLE_X), + ureg_imm1f(shader, VL_BLOCK_WIDTH / idct->nr_of_render_targets)); + ureg_MUL(shader, ureg_writemask(t_start, TGSI_WRITEMASK_XY), vpos, scale); + + calc_addr(shader, o_l_addr, vrect, ureg_imm1f(shader, 0.0f), false, false, VL_BLOCK_WIDTH / 4); + calc_addr(shader, o_r_addr, ureg_src(tex), ureg_src(t_start), true, false, idct->buffer_height / 4); + ureg_MOV(shader, ureg_writemask(o_r_addr[0], TGSI_WRITEMASK_Z), ureg_src(tex)); + ureg_MOV(shader, ureg_writemask(o_r_addr[1], TGSI_WRITEMASK_Z), ureg_src(tex)); +} + +void +vl_idct_stage2_frag_shader(struct vl_idct *idct, struct ureg_program *shader, + unsigned first_input, struct ureg_dst fragment) +{ struct ureg_src l_addr[2], r_addr[2]; struct ureg_dst l[2], r[2]; - struct ureg_dst fragment; - shader = ureg_create(TGSI_PROCESSOR_FRAGMENT); - if (!shader) - return NULL; + --first_input; - l_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0, TGSI_INTERPOLATE_LINEAR); - l_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1, TGSI_INTERPOLATE_LINEAR); + l_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, first_input + VS_O_L_ADDR0, TGSI_INTERPOLATE_LINEAR); + l_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, first_input + VS_O_L_ADDR1, TGSI_INTERPOLATE_LINEAR); - r_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR0, TGSI_INTERPOLATE_LINEAR); - r_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR1, TGSI_INTERPOLATE_LINEAR); + r_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, first_input + VS_O_R_ADDR0, TGSI_INTERPOLATE_LINEAR); + r_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, first_input + VS_O_R_ADDR1, TGSI_INTERPOLATE_LINEAR); l[0] = ureg_DECL_temporary(shader); l[1] = ureg_DECL_temporary(shader); r[0] = ureg_DECL_temporary(shader); r[1] = ureg_DECL_temporary(shader); - fetch_four(shader, l, l_addr, ureg_DECL_sampler(shader, 0)); - fetch_four(shader, r, r_addr, ureg_DECL_sampler(shader, 1)); - - fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0); + fetch_four(shader, l, l_addr, ureg_DECL_sampler(shader, 1), false); + fetch_four(shader, r, r_addr, ureg_DECL_sampler(shader, 0), true); - matrix_mul(shader, ureg_writemask(fragment, TGSI_WRITEMASK_X), l, r); + matrix_mul(shader, fragment, l, r); ureg_release_temporary(shader, l[0]); ureg_release_temporary(shader, l[1]); ureg_release_temporary(shader, r[0]); ureg_release_temporary(shader, r[1]); - - ureg_END(shader); - - return ureg_create_shader_and_destroy(shader, idct->pipe); } static bool -init_shaders(struct vl_idct *idct, int color_swizzle) +init_shaders(struct vl_idct *idct) { - idct->matrix_vs = create_vert_shader(idct, true, color_swizzle); - idct->matrix_fs = create_matrix_frag_shader(idct); + idct->vs_mismatch = create_mismatch_vert_shader(idct); + if (!idct->vs_mismatch) + goto error_vs_mismatch; - idct->transpose_vs = create_vert_shader(idct, false, color_swizzle); - idct->transpose_fs = create_transpose_frag_shader(idct); + idct->fs_mismatch = create_mismatch_frag_shader(idct); + if (!idct->fs_mismatch) + goto error_fs_mismatch; - return - idct->matrix_vs != NULL && - idct->matrix_fs != NULL && - idct->transpose_vs != NULL && - idct->transpose_fs != NULL; + idct->vs = create_stage1_vert_shader(idct); + if (!idct->vs) + goto error_vs; + + idct->fs = create_stage1_frag_shader(idct); + if (!idct->fs) + goto error_fs; + + return true; + +error_fs: + idct->pipe->delete_vs_state(idct->pipe, idct->vs); + +error_vs: + idct->pipe->delete_vs_state(idct->pipe, idct->vs_mismatch); + +error_fs_mismatch: + idct->pipe->delete_vs_state(idct->pipe, idct->fs); + +error_vs_mismatch: + return false; } static void cleanup_shaders(struct vl_idct *idct) { - idct->pipe->delete_vs_state(idct->pipe, idct->matrix_vs); - idct->pipe->delete_fs_state(idct->pipe, idct->matrix_fs); - idct->pipe->delete_vs_state(idct->pipe, idct->transpose_vs); - idct->pipe->delete_fs_state(idct->pipe, idct->transpose_fs); + idct->pipe->delete_vs_state(idct->pipe, idct->vs_mismatch); + idct->pipe->delete_fs_state(idct->pipe, idct->fs_mismatch); + idct->pipe->delete_vs_state(idct->pipe, idct->vs); + idct->pipe->delete_fs_state(idct->pipe, idct->fs); } static bool init_state(struct vl_idct *idct) { - struct pipe_sampler_state sampler; + struct pipe_blend_state blend; struct pipe_rasterizer_state rs_state; + struct pipe_sampler_state sampler; unsigned i; assert(idct); - idct->quad = vl_vb_upload_quads(idct->pipe, idct->blocks_x, idct->blocks_y); - - if(idct->quad.buffer == NULL) - return false; - - for (i = 0; i < 4; ++i) { + memset(&rs_state, 0, sizeof(rs_state)); + rs_state.point_size = 1; + rs_state.half_pixel_center = true; + rs_state.bottom_edge_rule = true; + rs_state.depth_clip = 1; + idct->rs_state = idct->pipe->create_rasterizer_state(idct->pipe, &rs_state); + if (!idct->rs_state) + goto error_rs_state; + + memset(&blend, 0, sizeof blend); + + blend.independent_blend_enable = 0; + blend.rt[0].blend_enable = 0; + blend.rt[0].rgb_func = PIPE_BLEND_ADD; + blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE; + blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ONE; + blend.rt[0].alpha_func = PIPE_BLEND_ADD; + blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE; + blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ONE; + blend.logicop_enable = 0; + blend.logicop_func = PIPE_LOGICOP_CLEAR; + /* Needed to allow color writes to FB, even if blending disabled */ + blend.rt[0].colormask = PIPE_MASK_RGBA; + blend.dither = 0; + idct->blend = idct->pipe->create_blend_state(idct->pipe, &blend); + if (!idct->blend) + goto error_blend; + + for (i = 0; i < 2; ++i) { memset(&sampler, 0, sizeof(sampler)); sampler.wrap_s = PIPE_TEX_WRAP_REPEAT; sampler.wrap_t = PIPE_TEX_WRAP_REPEAT; @@ -410,26 +553,25 @@ init_state(struct vl_idct *idct) sampler.compare_mode = PIPE_TEX_COMPARE_NONE; sampler.compare_func = PIPE_FUNC_ALWAYS; sampler.normalized_coords = 1; - /*sampler.shadow_ambient = ; */ - /*sampler.lod_bias = ; */ - sampler.min_lod = 0; - /*sampler.max_lod = ; */ - /*sampler.border_color[0] = ; */ - /*sampler.max_anisotropy = ; */ - idct->samplers.all[i] = idct->pipe->create_sampler_state(idct->pipe, &sampler); + idct->samplers[i] = idct->pipe->create_sampler_state(idct->pipe, &sampler); + if (!idct->samplers[i]) + goto error_samplers; } - memset(&rs_state, 0, sizeof(rs_state)); - /*rs_state.sprite_coord_enable */ - rs_state.sprite_coord_mode = PIPE_SPRITE_COORD_UPPER_LEFT; - rs_state.point_quad_rasterization = true; - rs_state.point_size = BLOCK_WIDTH; - rs_state.gl_rasterization_rules = false; - idct->rs_state = idct->pipe->create_rasterizer_state(idct->pipe, &rs_state); + return true; - idct->vertex_elems_state = vl_vb_get_elems_state(idct->pipe, false); +error_samplers: + for (i = 0; i < 2; ++i) + if (idct->samplers[i]) + idct->pipe->delete_sampler_state(idct->pipe, idct->samplers[i]); - return true; + idct->pipe->delete_rasterizer_state(idct->pipe, idct->rs_state); + +error_blend: + idct->pipe->delete_blend_state(idct->pipe, idct->blend); + +error_rs_state: + return false; } static void @@ -437,98 +579,109 @@ cleanup_state(struct vl_idct *idct) { unsigned i; - for (i = 0; i < 4; ++i) - idct->pipe->delete_sampler_state(idct->pipe, idct->samplers.all[i]); + for (i = 0; i < 2; ++i) + idct->pipe->delete_sampler_state(idct->pipe, idct->samplers[i]); idct->pipe->delete_rasterizer_state(idct->pipe, idct->rs_state); - idct->pipe->delete_vertex_elements_state(idct->pipe, idct->vertex_elems_state); + idct->pipe->delete_blend_state(idct->pipe, idct->blend); } static bool -init_textures(struct vl_idct *idct, struct vl_idct_buffer *buffer) +init_source(struct vl_idct *idct, struct vl_idct_buffer *buffer) { - struct pipe_resource template; - struct pipe_sampler_view sampler_view; - unsigned i; + struct pipe_resource *tex; + struct pipe_surface surf_templ; assert(idct && buffer); - /* create textures */ - memset(&template, 0, sizeof(struct pipe_resource)); - template.last_level = 0; - template.bind = PIPE_BIND_SAMPLER_VIEW; - template.flags = 0; - - template.target = PIPE_TEXTURE_2D; - template.format = PIPE_FORMAT_R16G16B16A16_SNORM; - template.width0 = idct->buffer_width / 4; - template.height0 = idct->buffer_height; - template.depth0 = 1; - template.array_size = 1; - template.usage = PIPE_USAGE_STREAM; - buffer->textures.individual.source = idct->pipe->screen->resource_create(idct->pipe->screen, &template); - - template.target = PIPE_TEXTURE_3D; - template.format = PIPE_FORMAT_R16G16B16A16_SNORM; - template.width0 = idct->buffer_width / NR_RENDER_TARGETS; - template.height0 = idct->buffer_height / 4; - template.depth0 = NR_RENDER_TARGETS; - template.usage = PIPE_USAGE_STATIC; - buffer->textures.individual.intermediate = idct->pipe->screen->resource_create(idct->pipe->screen, &template); + tex = buffer->sampler_views.individual.source->texture; - for (i = 0; i < 4; ++i) { - if(buffer->textures.all[i] == NULL) - return false; /* a texture failed to allocate */ + buffer->fb_state_mismatch.width = tex->width0; + buffer->fb_state_mismatch.height = tex->height0; + buffer->fb_state_mismatch.nr_cbufs = 1; - u_sampler_view_default_template(&sampler_view, buffer->textures.all[i], buffer->textures.all[i]->format); - buffer->sampler_views.all[i] = idct->pipe->create_sampler_view(idct->pipe, buffer->textures.all[i], &sampler_view); - } + memset(&surf_templ, 0, sizeof(surf_templ)); + surf_templ.format = tex->format; + surf_templ.u.tex.first_layer = 0; + surf_templ.u.tex.last_layer = 0; + buffer->fb_state_mismatch.cbufs[0] = idct->pipe->create_surface(idct->pipe, tex, &surf_templ); + + buffer->viewport_mismatch.scale[0] = tex->width0; + buffer->viewport_mismatch.scale[1] = tex->height0; + buffer->viewport_mismatch.scale[2] = 1; + buffer->viewport_mismatch.scale[3] = 1; return true; } static void -cleanup_textures(struct vl_idct *idct, struct vl_idct_buffer *buffer) +cleanup_source(struct vl_idct_buffer *buffer) { - unsigned i; + assert(buffer); - assert(idct && buffer); + pipe_surface_reference(&buffer->fb_state_mismatch.cbufs[0], NULL); - for (i = 0; i < 4; ++i) { - pipe_sampler_view_reference(&buffer->sampler_views.all[i], NULL); - pipe_resource_reference(&buffer->textures.all[i], NULL); - } + pipe_sampler_view_reference(&buffer->sampler_views.individual.source, NULL); } static bool -init_vertex_buffers(struct vl_idct *idct, struct vl_idct_buffer *buffer, struct pipe_vertex_buffer stream) +init_intermediate(struct vl_idct *idct, struct vl_idct_buffer *buffer) { + struct pipe_resource *tex; + struct pipe_surface surf_templ; + unsigned i; + assert(idct && buffer); - buffer->vertex_bufs.individual.quad.stride = idct->quad.stride; - buffer->vertex_bufs.individual.quad.buffer_offset = idct->quad.buffer_offset; - pipe_resource_reference(&buffer->vertex_bufs.individual.quad.buffer, idct->quad.buffer); + tex = buffer->sampler_views.individual.intermediate->texture; + + buffer->fb_state.width = tex->width0; + buffer->fb_state.height = tex->height0; + buffer->fb_state.nr_cbufs = idct->nr_of_render_targets; + for(i = 0; i < idct->nr_of_render_targets; ++i) { + memset(&surf_templ, 0, sizeof(surf_templ)); + surf_templ.format = tex->format; + surf_templ.u.tex.first_layer = i; + surf_templ.u.tex.last_layer = i; + buffer->fb_state.cbufs[i] = idct->pipe->create_surface( + idct->pipe, tex, &surf_templ); + + if (!buffer->fb_state.cbufs[i]) + goto error_surfaces; + } - buffer->vertex_bufs.individual.stream.stride = stream.stride; - buffer->vertex_bufs.individual.stream.buffer_offset = stream.buffer_offset; - pipe_resource_reference(&buffer->vertex_bufs.individual.stream.buffer, stream.buffer); + buffer->viewport.scale[0] = tex->width0; + buffer->viewport.scale[1] = tex->height0; + buffer->viewport.scale[2] = 1; + buffer->viewport.scale[3] = 1; return true; + +error_surfaces: + for(i = 0; i < idct->nr_of_render_targets; ++i) + pipe_surface_reference(&buffer->fb_state.cbufs[i], NULL); + + return false; } static void -cleanup_vertex_buffers(struct vl_idct *idct, struct vl_idct_buffer *buffer) +cleanup_intermediate(struct vl_idct_buffer *buffer) { - assert(idct && buffer); + unsigned i; + + assert(buffer); + + for(i = 0; i < PIPE_MAX_COLOR_BUFS; ++i) + pipe_surface_reference(&buffer->fb_state.cbufs[i], NULL); - pipe_resource_reference(&buffer->vertex_bufs.individual.quad.buffer, NULL); - pipe_resource_reference(&buffer->vertex_bufs.individual.stream.buffer, NULL); + pipe_sampler_view_reference(&buffer->sampler_views.individual.intermediate, NULL); } -struct pipe_resource * -vl_idct_upload_matrix(struct pipe_context *pipe) +struct pipe_sampler_view * +vl_idct_upload_matrix(struct pipe_context *pipe, float scale) { - struct pipe_resource template, *matrix; + struct pipe_resource tex_templ, *matrix; + struct pipe_sampler_view sv_templ, *sv; struct pipe_transfer *buf_transfer; unsigned i, j, pitch; float *f; @@ -536,61 +689,79 @@ vl_idct_upload_matrix(struct pipe_context *pipe) struct pipe_box rect = { 0, 0, 0, - BLOCK_WIDTH / 4, - BLOCK_HEIGHT, + VL_BLOCK_WIDTH / 4, + VL_BLOCK_HEIGHT, 1 }; - memset(&template, 0, sizeof(struct pipe_resource)); - template.target = PIPE_TEXTURE_2D; - template.format = PIPE_FORMAT_R32G32B32A32_FLOAT; - template.last_level = 0; - template.width0 = 2; - template.height0 = 8; - template.depth0 = 1; - template.array_size = 1; - template.usage = PIPE_USAGE_IMMUTABLE; - template.bind = PIPE_BIND_SAMPLER_VIEW; - template.flags = 0; - - matrix = pipe->screen->resource_create(pipe->screen, &template); - - /* matrix */ - buf_transfer = pipe->get_transfer - ( - pipe, matrix, - 0, PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD, - &rect - ); + assert(pipe); + + memset(&tex_templ, 0, sizeof(tex_templ)); + tex_templ.target = PIPE_TEXTURE_2D; + tex_templ.format = PIPE_FORMAT_R32G32B32A32_FLOAT; + tex_templ.last_level = 0; + tex_templ.width0 = 2; + tex_templ.height0 = 8; + tex_templ.depth0 = 1; + tex_templ.array_size = 1; + tex_templ.usage = PIPE_USAGE_IMMUTABLE; + tex_templ.bind = PIPE_BIND_SAMPLER_VIEW; + tex_templ.flags = 0; + + matrix = pipe->screen->resource_create(pipe->screen, &tex_templ); + if (!matrix) + goto error_matrix; + + f = pipe->transfer_map(pipe, matrix, 0, + PIPE_TRANSFER_WRITE | + PIPE_TRANSFER_DISCARD_RANGE, + &rect, &buf_transfer); + if (!f) + goto error_map; + pitch = buf_transfer->stride / sizeof(float); - f = pipe->transfer_map(pipe, buf_transfer); - for(i = 0; i < BLOCK_HEIGHT; ++i) - for(j = 0; j < BLOCK_WIDTH; ++j) + for(i = 0; i < VL_BLOCK_HEIGHT; ++i) + for(j = 0; j < VL_BLOCK_WIDTH; ++j) // transpose and scale - f[i * pitch + j] = const_matrix[j][i] * sqrtf(SCALE_FACTOR_16_TO_9); + f[i * pitch + j] = ((const float (*)[8])const_matrix)[j][i] * scale; pipe->transfer_unmap(pipe, buf_transfer); - pipe->transfer_destroy(pipe, buf_transfer); - return matrix; + memset(&sv_templ, 0, sizeof(sv_templ)); + u_sampler_view_default_template(&sv_templ, matrix, matrix->format); + sv = pipe->create_sampler_view(pipe, matrix, &sv_templ); + pipe_resource_reference(&matrix, NULL); + if (!sv) + goto error_map; + + return sv; + +error_map: + pipe_resource_reference(&matrix, NULL); + +error_matrix: + return NULL; } bool vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe, unsigned buffer_width, unsigned buffer_height, - unsigned blocks_x, unsigned blocks_y, - int color_swizzle, struct pipe_resource *matrix) + unsigned nr_of_render_targets, + struct pipe_sampler_view *matrix, + struct pipe_sampler_view *transpose) { - assert(idct && pipe && matrix); + assert(idct && pipe); + assert(matrix && transpose); idct->pipe = pipe; idct->buffer_width = buffer_width; idct->buffer_height = buffer_height; - idct->blocks_x = blocks_x; - idct->blocks_y = blocks_y; - pipe_resource_reference(&idct->matrix, matrix); + idct->nr_of_render_targets = nr_of_render_targets; - if(!init_shaders(idct, color_swizzle)) + pipe_sampler_view_reference(&idct->matrix, matrix); + pipe_sampler_view_reference(&idct->transpose, transpose); + + if(!init_shaders(idct)) return false; if(!init_state(idct)) { @@ -607,176 +778,85 @@ vl_idct_cleanup(struct vl_idct *idct) cleanup_shaders(idct); cleanup_state(idct); - pipe_resource_reference(&idct->matrix, NULL); + pipe_sampler_view_reference(&idct->matrix, NULL); + pipe_sampler_view_reference(&idct->transpose, NULL); } bool vl_idct_init_buffer(struct vl_idct *idct, struct vl_idct_buffer *buffer, - struct pipe_resource *dst, struct pipe_vertex_buffer stream) + struct pipe_sampler_view *source, + struct pipe_sampler_view *intermediate) { - struct pipe_surface template; - - unsigned i; + assert(buffer && idct); + assert(source && intermediate); - assert(buffer); - assert(idct); - assert(dst); + memset(buffer, 0, sizeof(struct vl_idct_buffer)); - pipe_resource_reference(&buffer->textures.individual.matrix, idct->matrix); - pipe_resource_reference(&buffer->textures.individual.transpose, idct->matrix); - pipe_resource_reference(&buffer->destination, dst); + pipe_sampler_view_reference(&buffer->sampler_views.individual.matrix, idct->matrix); + pipe_sampler_view_reference(&buffer->sampler_views.individual.source, source); + pipe_sampler_view_reference(&buffer->sampler_views.individual.transpose, idct->transpose); + pipe_sampler_view_reference(&buffer->sampler_views.individual.intermediate, intermediate); - if (!init_textures(idct, buffer)) + if (!init_source(idct, buffer)) return false; - if (!init_vertex_buffers(idct, buffer, stream)) + if (!init_intermediate(idct, buffer)) return false; - /* init state */ - buffer->viewport[0].scale[0] = buffer->textures.individual.intermediate->width0; - buffer->viewport[0].scale[1] = buffer->textures.individual.intermediate->height0; - - buffer->viewport[1].scale[0] = buffer->destination->width0; - buffer->viewport[1].scale[1] = buffer->destination->height0; - - buffer->fb_state[0].width = buffer->textures.individual.intermediate->width0; - buffer->fb_state[0].height = buffer->textures.individual.intermediate->height0; - - buffer->fb_state[0].nr_cbufs = NR_RENDER_TARGETS; - for(i = 0; i < NR_RENDER_TARGETS; ++i) { - memset(&template, 0, sizeof(template)); - template.format = buffer->textures.individual.intermediate->format; - template.u.tex.first_layer = i; - template.u.tex.last_layer = i; - template.usage = PIPE_BIND_RENDER_TARGET; - buffer->fb_state[0].cbufs[i] = idct->pipe->create_surface( - idct->pipe, buffer->textures.individual.intermediate, - &template); - } - - buffer->fb_state[1].width = buffer->destination->width0; - buffer->fb_state[1].height = buffer->destination->height0; - - buffer->fb_state[1].nr_cbufs = 1; - - memset(&template, 0, sizeof(template)); - template.format = buffer->destination->format; - template.usage = PIPE_BIND_RENDER_TARGET; - buffer->fb_state[1].cbufs[0] = idct->pipe->create_surface( - idct->pipe, buffer->destination, &template); - - for(i = 0; i < 2; ++i) { - buffer->viewport[i].scale[2] = 1; - buffer->viewport[i].scale[3] = 1; - buffer->viewport[i].translate[0] = 0; - buffer->viewport[i].translate[1] = 0; - buffer->viewport[i].translate[2] = 0; - buffer->viewport[i].translate[3] = 0; - - buffer->fb_state[i].zsbuf = NULL; - } - return true; } void -vl_idct_cleanup_buffer(struct vl_idct *idct, struct vl_idct_buffer *buffer) +vl_idct_cleanup_buffer(struct vl_idct_buffer *buffer) { - unsigned i; - assert(buffer); - for(i = 0; i < NR_RENDER_TARGETS; ++i) { - idct->pipe->surface_destroy(idct->pipe, buffer->fb_state[0].cbufs[i]); - } - - idct->pipe->surface_destroy(idct->pipe, buffer->fb_state[1].cbufs[0]); + cleanup_source(buffer); + cleanup_intermediate(buffer); - cleanup_textures(idct, buffer); - cleanup_vertex_buffers(idct, buffer); + pipe_sampler_view_reference(&buffer->sampler_views.individual.matrix, NULL); + pipe_sampler_view_reference(&buffer->sampler_views.individual.transpose, NULL); } void -vl_idct_map_buffers(struct vl_idct *idct, struct vl_idct_buffer *buffer) -{ - assert(idct); - - struct pipe_box rect = - { - 0, 0, 0, - buffer->textures.individual.source->width0, - buffer->textures.individual.source->height0, - 1 - }; - - buffer->tex_transfer = idct->pipe->get_transfer - ( - idct->pipe, buffer->textures.individual.source, - 0, PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD, - &rect - ); - - buffer->texels = idct->pipe->transfer_map(idct->pipe, buffer->tex_transfer); -} - -void -vl_idct_add_block(struct vl_idct_buffer *buffer, unsigned x, unsigned y, short *block) +vl_idct_flush(struct vl_idct *idct, struct vl_idct_buffer *buffer, unsigned num_instances) { - //struct vertex2s v; - unsigned tex_pitch; - short *texels; - - unsigned i; - assert(buffer); - tex_pitch = buffer->tex_transfer->stride / sizeof(short); - texels = buffer->texels + y * tex_pitch * BLOCK_HEIGHT + x * BLOCK_WIDTH; + idct->pipe->bind_rasterizer_state(idct->pipe, idct->rs_state); + idct->pipe->bind_blend_state(idct->pipe, idct->blend); - for (i = 0; i < BLOCK_HEIGHT; ++i) - memcpy(texels + i * tex_pitch, block + i * BLOCK_WIDTH, BLOCK_WIDTH * sizeof(short)); -} + idct->pipe->bind_sampler_states(idct->pipe, PIPE_SHADER_FRAGMENT, + 0, 2, idct->samplers); -void -vl_idct_unmap_buffers(struct vl_idct *idct, struct vl_idct_buffer *buffer) -{ - assert(idct && buffer); + idct->pipe->set_sampler_views(idct->pipe, PIPE_SHADER_FRAGMENT, 0, 2, + buffer->sampler_views.stage[0]); - idct->pipe->transfer_unmap(idct->pipe, buffer->tex_transfer); - idct->pipe->transfer_destroy(idct->pipe, buffer->tex_transfer); + /* mismatch control */ + idct->pipe->set_framebuffer_state(idct->pipe, &buffer->fb_state_mismatch); + idct->pipe->set_viewport_states(idct->pipe, 0, 1, &buffer->viewport_mismatch); + idct->pipe->bind_vs_state(idct->pipe, idct->vs_mismatch); + idct->pipe->bind_fs_state(idct->pipe, idct->fs_mismatch); + util_draw_arrays_instanced(idct->pipe, PIPE_PRIM_POINTS, 0, 1, 0, num_instances); + + /* first stage */ + idct->pipe->set_framebuffer_state(idct->pipe, &buffer->fb_state); + idct->pipe->set_viewport_states(idct->pipe, 0, 1, &buffer->viewport); + idct->pipe->bind_vs_state(idct->pipe, idct->vs); + idct->pipe->bind_fs_state(idct->pipe, idct->fs); + util_draw_arrays_instanced(idct->pipe, PIPE_PRIM_QUADS, 0, 4, 0, num_instances); } void -vl_idct_flush(struct vl_idct *idct, struct vl_idct_buffer *buffer, unsigned num_instances) +vl_idct_prepare_stage2(struct vl_idct *idct, struct vl_idct_buffer *buffer) { - unsigned num_verts; - - assert(idct); assert(buffer); - if(num_instances > 0) { - num_verts = idct->blocks_x * idct->blocks_y * 4; - - idct->pipe->bind_rasterizer_state(idct->pipe, idct->rs_state); - idct->pipe->set_vertex_buffers(idct->pipe, 2, buffer->vertex_bufs.all); - idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state); - - /* first stage */ - idct->pipe->set_framebuffer_state(idct->pipe, &buffer->fb_state[0]); - idct->pipe->set_viewport_state(idct->pipe, &buffer->viewport[0]); - idct->pipe->set_fragment_sampler_views(idct->pipe, 2, buffer->sampler_views.stage[0]); - idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[0]); - idct->pipe->bind_vs_state(idct->pipe, idct->matrix_vs); - idct->pipe->bind_fs_state(idct->pipe, idct->matrix_fs); - util_draw_arrays_instanced(idct->pipe, PIPE_PRIM_QUADS, 0, num_verts, 0, num_instances); - - /* second stage */ - idct->pipe->set_framebuffer_state(idct->pipe, &buffer->fb_state[1]); - idct->pipe->set_viewport_state(idct->pipe, &buffer->viewport[1]); - idct->pipe->set_fragment_sampler_views(idct->pipe, 2, buffer->sampler_views.stage[1]); - idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[1]); - idct->pipe->bind_vs_state(idct->pipe, idct->transpose_vs); - idct->pipe->bind_fs_state(idct->pipe, idct->transpose_fs); - util_draw_arrays_instanced(idct->pipe, PIPE_PRIM_QUADS, 0, num_verts, 0, num_instances); - } + /* second stage */ + idct->pipe->bind_rasterizer_state(idct->pipe, idct->rs_state); + idct->pipe->bind_sampler_states(idct->pipe, PIPE_SHADER_FRAGMENT, + 0, 2, idct->samplers); + idct->pipe->set_sampler_views(idct->pipe, PIPE_SHADER_FRAGMENT, + 0, 2, buffer->sampler_views.stage[1]); } +