From adc1f88fc9278bdbb3b24a6d48f91a0bd98e9f1c Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Fri, 23 May 2008 09:10:59 +0100 Subject: mesa: do object-space lighting in ffvertex_prog.c Start pulling over some of the optimizations from the fixed function paths. --- src/mesa/main/ffvertex_prog.c | 79 +++++++++++++++++++++++++++---------------- 1 file changed, 50 insertions(+), 29 deletions(-) (limited to 'src/mesa/main/ffvertex_prog.c') diff --git a/src/mesa/main/ffvertex_prog.c b/src/mesa/main/ffvertex_prog.c index 810af9e33e6..adf15b03c2e 100644 --- a/src/mesa/main/ffvertex_prog.c +++ b/src/mesa/main/ffvertex_prog.c @@ -54,6 +54,7 @@ struct state_key { unsigned light_color_material_mask:12; unsigned light_material_mask:12; + unsigned need_eye_coords:1; unsigned normalize:1; unsigned rescale_normals:1; unsigned fog_source_is_depth:1; @@ -167,6 +168,8 @@ static struct state_key *make_state_key( GLcontext *ctx ) */ assert(fp); + key->need_eye_coords = ctx->_NeedEyeCoords; + key->fragprog_inputs_read = fp->Base.InputsRead; if (ctx->RenderMode == GL_FEEDBACK) { @@ -310,7 +313,7 @@ struct tnl_program { struct ureg eye_position; struct ureg eye_position_normalized; - struct ureg eye_normal; + struct ureg transformed_normal; struct ureg identity; GLuint materials; @@ -653,9 +656,9 @@ static void emit_normalize_vec3( struct tnl_program *p, struct ureg src ) { struct ureg tmp = get_temp(p); - emit_op2(p, OPCODE_DP3, tmp, 0, src, src); - emit_op1(p, OPCODE_RSQ, tmp, 0, tmp); - emit_op2(p, OPCODE_MUL, dest, 0, src, tmp); + emit_op2(p, OPCODE_DP3, tmp, WRITEMASK_X, src, src); + emit_op1(p, OPCODE_RSQ, tmp, WRITEMASK_X, tmp); + emit_op2(p, OPCODE_MUL, dest, 0, src, swizzle1(tmp, X)); release_temp(p, tmp); } @@ -705,36 +708,53 @@ static struct ureg get_eye_position_normalized( struct tnl_program *p ) } -static struct ureg get_eye_normal( struct tnl_program *p ) +static struct ureg get_transformed_normal( struct tnl_program *p ) { - if (is_undef(p->eye_normal)) { + if (is_undef(p->transformed_normal) && + !p->state->need_eye_coords && + !p->state->normalize && + !(p->state->need_eye_coords == p->state->rescale_normals)) + { + p->transformed_normal = register_input(p, VERT_ATTRIB_NORMAL ); + } + else if (is_undef(p->transformed_normal)) + { struct ureg normal = register_input(p, VERT_ATTRIB_NORMAL ); struct ureg mvinv[3]; + struct ureg transformed_normal = reserve_temp(p); - register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 2, - STATE_MATRIX_INVTRANS, mvinv ); + if (p->state->need_eye_coords) { + register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 2, + STATE_MATRIX_INVTRANS, mvinv ); - p->eye_normal = reserve_temp(p); - - /* Transform to eye space: - */ - emit_matrix_transform_vec3( p, p->eye_normal, mvinv, normal ); + /* Transform to eye space: + */ + emit_matrix_transform_vec3( p, transformed_normal, mvinv, normal ); + normal = transformed_normal; + } /* Normalize/Rescale: */ if (p->state->normalize) { - emit_normalize_vec3( p, p->eye_normal, p->eye_normal ); + emit_normalize_vec3( p, transformed_normal, normal ); + normal = transformed_normal; } - else if (p->state->rescale_normals) { + else if (p->state->need_eye_coords == p->state->rescale_normals) { + /* This is already adjusted for eye/non-eye rendering: + */ struct ureg rescale = register_param2(p, STATE_INTERNAL, - STATE_NORMAL_SCALE); + STATE_NORMAL_SCALE); - emit_op2( p, OPCODE_MUL, p->eye_normal, 0, p->eye_normal, + emit_op2( p, OPCODE_MUL, transformed_normal, 0, normal, swizzle1(rescale, X)); + normal = transformed_normal; } + + assert(normal.file == PROGRAM_TEMPORARY); + p->transformed_normal = normal; } - return p->eye_normal; + return p->transformed_normal; } @@ -856,7 +876,7 @@ static struct ureg calculate_light_attenuation( struct tnl_program *p, */ if (!p->state->unit[i].light_spotcutoff_is_180) { struct ureg spot_dir_norm = register_param3(p, STATE_INTERNAL, - STATE_SPOT_DIR_NORMALIZED, i); + STATE_LIGHT_SPOT_DIR_NORMALIZED, i); struct ureg spot = get_temp(p); struct ureg slt = get_temp(p); @@ -907,7 +927,7 @@ static void build_lighting( struct tnl_program *p ) const GLboolean twoside = p->state->light_twoside; const GLboolean separate = p->state->separate_specular; GLuint nr_lights = 0, count = 0; - struct ureg normal = get_eye_normal(p); + struct ureg normal = get_transformed_normal(p); struct ureg lit = get_temp(p); struct ureg dots = get_temp(p); struct ureg _col0 = undef, _col1 = undef; @@ -984,20 +1004,21 @@ static void build_lighting( struct tnl_program *p ) /* Can used precomputed constants in this case. * Attenuation never applies to infinite lights. */ - VPpli = register_param3(p, STATE_LIGHT, i, - STATE_POSITION_NORMALIZED); + VPpli = register_param3(p, STATE_INTERNAL, + STATE_LIGHT_POSITION_NORMALIZED, i); if (p->state->light_local_viewer) { struct ureg eye_hat = get_eye_position_normalized(p); half = get_temp(p); emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat); emit_normalize_vec3(p, half, half); } else { - half = register_param3(p, STATE_LIGHT, i, STATE_HALF_VECTOR); + half = register_param3(p, STATE_INTERNAL, + STATE_LIGHT_HALF_VECTOR, i); } } else { - struct ureg Ppli = register_param3(p, STATE_LIGHT, i, - STATE_POSITION); + struct ureg Ppli = register_param3(p, STATE_INTERNAL, + STATE_LIGHT_POSITION, i); struct ureg V = get_eye_position(p); struct ureg dist = get_temp(p); @@ -1201,7 +1222,7 @@ static void build_reflect_texgen( struct tnl_program *p, struct ureg dest, GLuint writemask ) { - struct ureg normal = get_eye_normal(p); + struct ureg normal = get_transformed_normal(p); struct ureg eye_hat = get_eye_position_normalized(p); struct ureg tmp = get_temp(p); @@ -1219,7 +1240,7 @@ static void build_sphere_texgen( struct tnl_program *p, struct ureg dest, GLuint writemask ) { - struct ureg normal = get_eye_normal(p); + struct ureg normal = get_transformed_normal(p); struct ureg eye_hat = get_eye_position_normalized(p); struct ureg tmp = get_temp(p); struct ureg half = register_scalar_const(p, .5); @@ -1338,7 +1359,7 @@ static void build_texture_transform( struct tnl_program *p ) } if (normal_mask) { - struct ureg normal = get_eye_normal(p); + struct ureg normal = get_transformed_normal(p); emit_op1(p, OPCODE_MOV, out_texgen, normal_mask, normal ); } @@ -1475,7 +1496,7 @@ create_new_program( const struct state_key *key, p.program = program; p.eye_position = undef; p.eye_position_normalized = undef; - p.eye_normal = undef; + p.transformed_normal = undef; p.identity = undef; p.temp_in_use = 0; -- cgit v1.2.3 From 0ac2f7955c01749e122f67ff03e79a0d8bd0f8e5 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Fri, 23 May 2008 19:17:02 +0100 Subject: mesa: don't emit LIT instruction when mat shininess known to be zero Use a faster path in that case & make gears go faster. --- src/mesa/main/ffvertex_prog.c | 133 ++++++++++++++++++++++++++++++++---------- 1 file changed, 102 insertions(+), 31 deletions(-) (limited to 'src/mesa/main/ffvertex_prog.c') diff --git a/src/mesa/main/ffvertex_prog.c b/src/mesa/main/ffvertex_prog.c index adf15b03c2e..623c2a64b5e 100644 --- a/src/mesa/main/ffvertex_prog.c +++ b/src/mesa/main/ffvertex_prog.c @@ -53,6 +53,7 @@ struct state_key { unsigned light_color_material:1; unsigned light_color_material_mask:12; unsigned light_material_mask:12; + unsigned material_shininess_is_zero:1; unsigned need_eye_coords:1; unsigned normalize:1; @@ -155,6 +156,26 @@ tnl_get_per_vertex_fog(GLcontext *ctx) #endif } +static GLboolean check_active_shininess( GLcontext *ctx, + const struct state_key *key, + GLuint side ) +{ + GLuint bit = 1 << (MAT_ATTRIB_FRONT_SHININESS + side); + + if (key->light_color_material_mask & bit) + return GL_TRUE; + + if (key->light_material_mask & bit) + return GL_TRUE; + + if (ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_SHININESS + side][0] != 0.0F) + return GL_TRUE; + + return GL_FALSE; +} + + + static struct state_key *make_state_key( GLcontext *ctx ) { @@ -214,6 +235,17 @@ static struct state_key *make_state_key( GLcontext *ctx ) key->unit[i].light_attenuated = 1; } } + + if (check_active_shininess(ctx, key, 0)) { + key->material_shininess_is_zero = 0; + } + else if (key->light_twoside && + check_active_shininess(ctx, key, 1)) { + key->material_shininess_is_zero = 0; + } + else { + key->material_shininess_is_zero = 1; + } } if (ctx->Transform.Normalize) @@ -915,7 +947,26 @@ static struct ureg calculate_light_attenuation( struct tnl_program *p, } +static void emit_degenerate_lit( struct tnl_program *p, + struct ureg lit, + struct ureg dots ) +{ + struct ureg id = get_identity_param(p); + + /* 1, 0, 0, 1 + */ + emit_op1(p, OPCODE_MOV, lit, 0, swizzle(id, Z, X, X, Z)); + /* 1, MAX2(in[0], 0), 0, 1 + */ + emit_op2(p, OPCODE_MAX, lit, WRITEMASK_Y, lit, swizzle1(dots, X)); + + /* 1, MAX2(in[0], 0), (in[0] > 0 ? 1 : 0), 1 + */ + emit_op2(p, OPCODE_SLT, lit, WRITEMASK_Z, + lit, /* 0 */ + swizzle1(dots, X)); /* in[0] */ +} /* Need to add some addtional parameters to allow lighting in object @@ -941,9 +992,11 @@ static void build_lighting( struct tnl_program *p ) set_material_flags(p); { - struct ureg shininess = get_material(p, 0, STATE_SHININESS); - emit_op1(p, OPCODE_MOV, dots, WRITEMASK_W, swizzle1(shininess,X)); - release_temp(p, shininess); + if (!p->state->material_shininess_is_zero) { + struct ureg shininess = get_material(p, 0, STATE_SHININESS); + emit_op1(p, OPCODE_MOV, dots, WRITEMASK_W, swizzle1(shininess,X)); + release_temp(p, shininess); + } _col0 = make_temp(p, get_scenecolor(p, 0)); if (separate) @@ -954,10 +1007,12 @@ static void build_lighting( struct tnl_program *p ) } if (twoside) { - struct ureg shininess = get_material(p, 1, STATE_SHININESS); - emit_op1(p, OPCODE_MOV, dots, WRITEMASK_Z, - negate(swizzle1(shininess,X))); - release_temp(p, shininess); + if (!p->state->material_shininess_is_zero) { + struct ureg shininess = get_material(p, 1, STATE_SHININESS); + emit_op1(p, OPCODE_MOV, dots, WRITEMASK_Z, + negate(swizzle1(shininess,X))); + release_temp(p, shininess); + } _bfc0 = make_temp(p, get_scenecolor(p, 1)); if (separate) @@ -1006,14 +1061,17 @@ static void build_lighting( struct tnl_program *p ) */ VPpli = register_param3(p, STATE_INTERNAL, STATE_LIGHT_POSITION_NORMALIZED, i); - if (p->state->light_local_viewer) { - struct ureg eye_hat = get_eye_position_normalized(p); - half = get_temp(p); - emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat); - emit_normalize_vec3(p, half, half); - } else { - half = register_param3(p, STATE_INTERNAL, - STATE_LIGHT_HALF_VECTOR, i); + + if (!p->state->material_shininess_is_zero) { + if (p->state->light_local_viewer) { + struct ureg eye_hat = get_eye_position_normalized(p); + half = get_temp(p); + emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat); + emit_normalize_vec3(p, half, half); + } else { + half = register_param3(p, STATE_INTERNAL, + STATE_LIGHT_HALF_VECTOR, i); + } } } else { @@ -1023,7 +1081,6 @@ static void build_lighting( struct tnl_program *p ) struct ureg dist = get_temp(p); VPpli = get_temp(p); - half = get_temp(p); /* Calculate VPpli vector */ @@ -1045,16 +1102,20 @@ static void build_lighting( struct tnl_program *p ) /* Calculate viewer direction, or use infinite viewer: */ - if (p->state->light_local_viewer) { - struct ureg eye_hat = get_eye_position_normalized(p); - emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat); - } - else { - struct ureg z_dir = swizzle(get_identity_param(p),X,Y,W,Z); - emit_op2(p, OPCODE_ADD, half, 0, VPpli, z_dir); - } - - emit_normalize_vec3(p, half, half); + if (!p->state->material_shininess_is_zero) { + half = get_temp(p); + + if (p->state->light_local_viewer) { + struct ureg eye_hat = get_eye_position_normalized(p); + emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat); + } + else { + struct ureg z_dir = swizzle(get_identity_param(p),X,Y,W,Z); + emit_op2(p, OPCODE_ADD, half, 0, VPpli, z_dir); + } + + emit_normalize_vec3(p, half, half); + } release_temp(p, dist); } @@ -1062,7 +1123,9 @@ static void build_lighting( struct tnl_program *p ) /* Calculate dot products: */ emit_op2(p, OPCODE_DP3, dots, WRITEMASK_X, normal, VPpli); - emit_op2(p, OPCODE_DP3, dots, WRITEMASK_Y, normal, half); + + if (!p->state->material_shininess_is_zero) + emit_op2(p, OPCODE_DP3, dots, WRITEMASK_Y, normal, half); /* Front face lighting: */ @@ -1073,7 +1136,11 @@ static void build_lighting( struct tnl_program *p ) struct ureg res0, res1; GLuint mask0, mask1; - emit_op1(p, OPCODE_LIT, lit, 0, dots); + if (p->state->material_shininess_is_zero) { + emit_degenerate_lit(p, lit, dots); + } else { + emit_op1(p, OPCODE_LIT, lit, 0, dots); + } if (!is_undef(att)) emit_op2(p, OPCODE_MUL, lit, 0, lit, att); @@ -1099,7 +1166,7 @@ static void build_lighting( struct tnl_program *p ) res1 = _col1; } - emit_op3(p, OPCODE_MAD, _col0, 0, swizzle1(lit,X), ambient, _col0); + emit_op2(p, OPCODE_ADD, _col0, 0, ambient, _col0); emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _col0); emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _col1); @@ -1117,7 +1184,11 @@ static void build_lighting( struct tnl_program *p ) struct ureg res0, res1; GLuint mask0, mask1; - emit_op1(p, OPCODE_LIT, lit, 0, negate(swizzle(dots,X,Y,W,Z))); + if (p->state->material_shininess_is_zero) { + emit_degenerate_lit(p, lit, negate(swizzle(dots,X,Y,W,Z))); + } else { + emit_op1(p, OPCODE_LIT, lit, 0, negate(swizzle(dots,X,Y,W,Z))); + } if (!is_undef(att)) emit_op2(p, OPCODE_MUL, lit, 0, lit, att); @@ -1142,7 +1213,7 @@ static void build_lighting( struct tnl_program *p ) mask1 = 0; } - emit_op3(p, OPCODE_MAD, _bfc0, 0, swizzle1(lit,X), ambient, _bfc0); + emit_op2(p, OPCODE_ADD, _bfc0, 0, ambient, _bfc0); emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _bfc0); emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _bfc1); -- cgit v1.2.3 From e841b92d9c8bf48085b4996df828ae745977f931 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Fri, 23 May 2008 20:05:36 +0100 Subject: mesa: further degenerate the special case lit substitute --- src/mesa/main/ffvertex_prog.c | 21 ++++++++++++--------- 1 file changed, 12 insertions(+), 9 deletions(-) (limited to 'src/mesa/main/ffvertex_prog.c') diff --git a/src/mesa/main/ffvertex_prog.c b/src/mesa/main/ffvertex_prog.c index 623c2a64b5e..90b156f8128 100644 --- a/src/mesa/main/ffvertex_prog.c +++ b/src/mesa/main/ffvertex_prog.c @@ -953,19 +953,19 @@ static void emit_degenerate_lit( struct tnl_program *p, { struct ureg id = get_identity_param(p); - /* 1, 0, 0, 1 + /* Note that result.x & result.w will not be examined. Note also that + * dots.xyzw == dots.xxxx. */ - emit_op1(p, OPCODE_MOV, lit, 0, swizzle(id, Z, X, X, Z)); - /* 1, MAX2(in[0], 0), 0, 1 + /* result[1] = MAX2(in, 0) */ - emit_op2(p, OPCODE_MAX, lit, WRITEMASK_Y, lit, swizzle1(dots, X)); + emit_op2(p, OPCODE_MAX, lit, 0, id, dots); - /* 1, MAX2(in[0], 0), (in[0] > 0 ? 1 : 0), 1 + /* result[2] = (in > 0 ? 1 : 0) */ emit_op2(p, OPCODE_SLT, lit, WRITEMASK_Z, lit, /* 0 */ - swizzle1(dots, X)); /* in[0] */ + dots); /* in[0] */ } @@ -1122,10 +1122,13 @@ static void build_lighting( struct tnl_program *p ) /* Calculate dot products: */ - emit_op2(p, OPCODE_DP3, dots, WRITEMASK_X, normal, VPpli); - - if (!p->state->material_shininess_is_zero) + if (p->state->material_shininess_is_zero) { + emit_op2(p, OPCODE_DP3, dots, 0, normal, VPpli); + } + else { + emit_op2(p, OPCODE_DP3, dots, WRITEMASK_X, normal, VPpli); emit_op2(p, OPCODE_DP3, dots, WRITEMASK_Y, normal, half); + } /* Front face lighting: */ -- cgit v1.2.3 From feceb43948f76cc4d4c8ecbb86b1b1f438c6daee Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Fri, 23 May 2008 20:37:50 +0100 Subject: mesa: save a temp on normalizes --- src/mesa/main/ffvertex_prog.c | 10 ++++------ 1 file changed, 4 insertions(+), 6 deletions(-) (limited to 'src/mesa/main/ffvertex_prog.c') diff --git a/src/mesa/main/ffvertex_prog.c b/src/mesa/main/ffvertex_prog.c index 90b156f8128..e36f1f69a40 100644 --- a/src/mesa/main/ffvertex_prog.c +++ b/src/mesa/main/ffvertex_prog.c @@ -305,7 +305,7 @@ static struct state_key *make_state_key( GLcontext *ctx ) * generated program with line/function references for each * instruction back into this file: */ -#define DISASSEM (MESA_VERBOSE&VERBOSE_DISASSEM) +#define DISASSEM 1 /* Should be tunable by the driver - do we want to do matrix * multiplications with DP4's or with MUL/MAD's? SSE works better @@ -687,11 +687,9 @@ static void emit_normalize_vec3( struct tnl_program *p, struct ureg dest, struct ureg src ) { - struct ureg tmp = get_temp(p); - emit_op2(p, OPCODE_DP3, tmp, WRITEMASK_X, src, src); - emit_op1(p, OPCODE_RSQ, tmp, WRITEMASK_X, tmp); - emit_op2(p, OPCODE_MUL, dest, 0, src, swizzle1(tmp, X)); - release_temp(p, tmp); + emit_op2(p, OPCODE_DP3, dest, WRITEMASK_X, src, src); + emit_op1(p, OPCODE_RSQ, dest, WRITEMASK_X, dest); + emit_op2(p, OPCODE_MUL, dest, 0, src, swizzle1(dest, X)); } static void emit_passthrough( struct tnl_program *p, -- cgit v1.2.3 From e1590abb17f1effd92c136207f363de6cf52df18 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Sat, 24 May 2008 13:23:06 +0100 Subject: mesa: pre-swizzle normal scale state value --- src/mesa/main/ffvertex_prog.c | 3 +-- src/mesa/shader/prog_statevars.c | 6 +++++- 2 files changed, 6 insertions(+), 3 deletions(-) (limited to 'src/mesa/main/ffvertex_prog.c') diff --git a/src/mesa/main/ffvertex_prog.c b/src/mesa/main/ffvertex_prog.c index e36f1f69a40..7a099b23767 100644 --- a/src/mesa/main/ffvertex_prog.c +++ b/src/mesa/main/ffvertex_prog.c @@ -775,8 +775,7 @@ static struct ureg get_transformed_normal( struct tnl_program *p ) struct ureg rescale = register_param2(p, STATE_INTERNAL, STATE_NORMAL_SCALE); - emit_op2( p, OPCODE_MUL, transformed_normal, 0, normal, - swizzle1(rescale, X)); + emit_op2( p, OPCODE_MUL, transformed_normal, 0, normal, rescale ); normal = transformed_normal; } diff --git a/src/mesa/shader/prog_statevars.c b/src/mesa/shader/prog_statevars.c index 37bd17ba4a0..44fbfdcd041 100644 --- a/src/mesa/shader/prog_statevars.c +++ b/src/mesa/shader/prog_statevars.c @@ -397,7 +397,11 @@ _mesa_fetch_state(GLcontext *ctx, const gl_state_index state[], case STATE_INTERNAL: switch (state[1]) { case STATE_NORMAL_SCALE: - ASSIGN_4V(value, ctx->_ModelViewInvScale, 0, 0, 1); + ASSIGN_4V(value, + ctx->_ModelViewInvScale, + ctx->_ModelViewInvScale, + ctx->_ModelViewInvScale, + 1); return; case STATE_TEXRECT_SCALE: { -- cgit v1.2.3 From 48a24f0ff7e3aad000b8acc55c16bbeaca58abe6 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Sat, 24 May 2008 16:32:08 +0100 Subject: Revert "mesa: save a temp on normalizes" This reverts commit feceb43948f76cc4d4c8ecbb86b1b1f438c6daee. --- src/mesa/main/ffvertex_prog.c | 10 ++++++---- 1 file changed, 6 insertions(+), 4 deletions(-) (limited to 'src/mesa/main/ffvertex_prog.c') diff --git a/src/mesa/main/ffvertex_prog.c b/src/mesa/main/ffvertex_prog.c index 7a099b23767..a627a21f650 100644 --- a/src/mesa/main/ffvertex_prog.c +++ b/src/mesa/main/ffvertex_prog.c @@ -305,7 +305,7 @@ static struct state_key *make_state_key( GLcontext *ctx ) * generated program with line/function references for each * instruction back into this file: */ -#define DISASSEM 1 +#define DISASSEM (MESA_VERBOSE&VERBOSE_DISASSEM) /* Should be tunable by the driver - do we want to do matrix * multiplications with DP4's or with MUL/MAD's? SSE works better @@ -687,9 +687,11 @@ static void emit_normalize_vec3( struct tnl_program *p, struct ureg dest, struct ureg src ) { - emit_op2(p, OPCODE_DP3, dest, WRITEMASK_X, src, src); - emit_op1(p, OPCODE_RSQ, dest, WRITEMASK_X, dest); - emit_op2(p, OPCODE_MUL, dest, 0, src, swizzle1(dest, X)); + struct ureg tmp = get_temp(p); + emit_op2(p, OPCODE_DP3, tmp, WRITEMASK_X, src, src); + emit_op1(p, OPCODE_RSQ, tmp, WRITEMASK_X, tmp); + emit_op2(p, OPCODE_MUL, dest, 0, src, swizzle1(tmp, X)); + release_temp(p, tmp); } static void emit_passthrough( struct tnl_program *p, -- cgit v1.2.3 From dc1537bc25c7cbff0a41034ece0830146616f036 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 27 May 2008 09:48:32 +0100 Subject: ffvertex: don't compute whole eye vector if only eye.z is required --- src/mesa/main/ffvertex_prog.c | 30 +++++++++++++++++++++++++++--- 1 file changed, 27 insertions(+), 3 deletions(-) (limited to 'src/mesa/main/ffvertex_prog.c') diff --git a/src/mesa/main/ffvertex_prog.c b/src/mesa/main/ffvertex_prog.c index a627a21f650..a790d4b1427 100644 --- a/src/mesa/main/ffvertex_prog.c +++ b/src/mesa/main/ffvertex_prog.c @@ -305,7 +305,7 @@ static struct state_key *make_state_key( GLcontext *ctx ) * generated program with line/function references for each * instruction back into this file: */ -#define DISASSEM (MESA_VERBOSE&VERBOSE_DISASSEM) +#define DISASSEM 1 /* Should be tunable by the driver - do we want to do matrix * multiplications with DP4's or with MUL/MAD's? SSE works better @@ -344,6 +344,7 @@ struct tnl_program { GLuint temp_reserved; struct ureg eye_position; + struct ureg eye_position_z; struct ureg eye_position_normalized; struct ureg transformed_normal; struct ureg identity; @@ -728,6 +729,28 @@ static struct ureg get_eye_position( struct tnl_program *p ) } +static struct ureg get_eye_position_z( struct tnl_program *p ) +{ + if (!is_undef(p->eye_position)) + return swizzle1(p->eye_position, Z); + + if (is_undef(p->eye_position_z)) { + struct ureg pos = register_input( p, VERT_ATTRIB_POS ); + struct ureg modelview[4]; + + p->eye_position_z = reserve_temp(p); + + register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 3, + 0, modelview ); + + emit_op2(p, OPCODE_DP4, p->eye_position_z, 0, pos, modelview[2]); + } + + return p->eye_position_z; +} + + + static struct ureg get_eye_position_normalized( struct tnl_program *p ) { if (is_undef(p->eye_position_normalized)) { @@ -1240,7 +1263,7 @@ static void build_fog( struct tnl_program *p ) struct ureg input; if (p->state->fog_source_is_depth) { - input = swizzle1(get_eye_position(p), Z); + input = get_eye_position_z(p); } else { input = swizzle1(register_input(p, VERT_ATTRIB_FOG), X); @@ -1470,7 +1493,7 @@ static void build_texture_transform( struct tnl_program *p ) static void build_pointsize( struct tnl_program *p ) { - struct ureg eye = get_eye_position(p); + struct ureg eye = get_eye_position_z(p); struct ureg state_size = register_param1(p, STATE_POINT_SIZE); struct ureg state_attenuation = register_param1(p, STATE_POINT_ATTENUATION); struct ureg out = register_output(p, VERT_RESULT_PSIZ); @@ -1568,6 +1591,7 @@ create_new_program( const struct state_key *key, p.state = key; p.program = program; p.eye_position = undef; + p.eye_position_z = undef; p.eye_position_normalized = undef; p.transformed_normal = undef; p.identity = undef; -- cgit v1.2.3 From 2109ddafefde26dd20a1c6a25f594984143944a3 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 27 May 2008 10:35:33 +0100 Subject: ffvertex: emit full LIT when attenuating (needs the 1 in X position) --- src/mesa/main/ffvertex_prog.c | 50 +++++++++++++++++++++++++++---------------- src/mesa/main/light.c | 1 + 2 files changed, 32 insertions(+), 19 deletions(-) (limited to 'src/mesa/main/ffvertex_prog.c') diff --git a/src/mesa/main/ffvertex_prog.c b/src/mesa/main/ffvertex_prog.c index a790d4b1427..2ef3286e579 100644 --- a/src/mesa/main/ffvertex_prog.c +++ b/src/mesa/main/ffvertex_prog.c @@ -1161,15 +1161,6 @@ static void build_lighting( struct tnl_program *p ) struct ureg res0, res1; GLuint mask0, mask1; - if (p->state->material_shininess_is_zero) { - emit_degenerate_lit(p, lit, dots); - } else { - emit_op1(p, OPCODE_LIT, lit, 0, dots); - } - - if (!is_undef(att)) - emit_op2(p, OPCODE_MUL, lit, 0, lit, att); - if (count == nr_lights) { if (separate) { @@ -1191,7 +1182,21 @@ static void build_lighting( struct tnl_program *p ) res1 = _col1; } - emit_op2(p, OPCODE_ADD, _col0, 0, ambient, _col0); + + if (!is_undef(att)) { + emit_op1(p, OPCODE_LIT, lit, 0, dots); + emit_op2(p, OPCODE_MUL, lit, 0, lit, att); + emit_op3(p, OPCODE_MAD, _col0, 0, swizzle1(lit,X), ambient, _col0); + } + else if (!p->state->material_shininess_is_zero) { + emit_op1(p, OPCODE_LIT, lit, 0, dots); + emit_op2(p, OPCODE_ADD, _col0, 0, ambient, _col0); + } + else { + emit_degenerate_lit(p, lit, dots); + emit_op2(p, OPCODE_ADD, _col0, 0, ambient, _col0); + } + emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _col0); emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _col1); @@ -1209,15 +1214,6 @@ static void build_lighting( struct tnl_program *p ) struct ureg res0, res1; GLuint mask0, mask1; - if (p->state->material_shininess_is_zero) { - emit_degenerate_lit(p, lit, negate(swizzle(dots,X,Y,W,Z))); - } else { - emit_op1(p, OPCODE_LIT, lit, 0, negate(swizzle(dots,X,Y,W,Z))); - } - - if (!is_undef(att)) - emit_op2(p, OPCODE_MUL, lit, 0, lit, att); - if (count == nr_lights) { if (separate) { mask0 = WRITEMASK_XYZ; @@ -1238,6 +1234,22 @@ static void build_lighting( struct tnl_program *p ) mask1 = 0; } + dots = negate(swizzle(dots,X,Y,W,Z)); + + if (!is_undef(att)) { + emit_op1(p, OPCODE_LIT, lit, 0, dots); + emit_op2(p, OPCODE_MUL, lit, 0, lit, att); + emit_op3(p, OPCODE_MAD, _bfc0, 0, swizzle1(lit,X), ambient, _bfc0); + } + else if (!p->state->material_shininess_is_zero) { + emit_op1(p, OPCODE_LIT, lit, 0, dots); + emit_op2(p, OPCODE_ADD, _col0, 0, ambient, _col0); + } + else { + emit_degenerate_lit(p, lit, dots); + emit_op2(p, OPCODE_ADD, _col0, 0, ambient, _col0); + } + emit_op2(p, OPCODE_ADD, _bfc0, 0, ambient, _bfc0); emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _bfc0); emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _bfc1); diff --git a/src/mesa/main/light.c b/src/mesa/main/light.c index 6e057614bad..0af647e13c0 100644 --- a/src/mesa/main/light.c +++ b/src/mesa/main/light.c @@ -1357,6 +1357,7 @@ _mesa_init_lighting( GLcontext *ctx ) /* Miscellaneous */ ctx->Light._NeedEyeCoords = GL_FALSE; ctx->_NeedEyeCoords = GL_FALSE; + ctx->_ForceEyeCoords = GL_TRUE; ctx->_ModelViewInvScale = 1.0; } -- cgit v1.2.3