arb_program_shader.c

Go to the documentation of this file.

/*
 * Pixel and vertex shaders implementation using ARB_vertex_program
 * and ARB_fragment_program GL extensions.
 *
 * Copyright 2002-2003 Jason Edmeades
 * Copyright 2002-2003 Raphael Junqueira
 * Copyright 2004 Christian Costa
 * Copyright 2005 Oliver Stieber
 * Copyright 2006 Ivan Gyurdiev
 * Copyright 2006 Jason Green
 * Copyright 2006 Henri Verbeet
 * Copyright 2007-2011, 2013-2014 Stefan Dösinger for CodeWeavers
 * Copyright 2009 Henri Verbeet for CodeWeavers
 *
 * This library is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * This library is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with this library; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
 */
 
#include "config.h"
#include "wine/port.h"
 
#include <stdio.h>
 
#include "wined3d_private.h"
 
WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
WINE_DECLARE_DEBUG_CHANNEL(d3d_constants);
WINE_DECLARE_DEBUG_CHANNEL(d3d);
WINE_DECLARE_DEBUG_CHANNEL(d3d_perf);
 
static BOOL shader_is_pshader_version(enum wined3d_shader_type type)
{
    return type == WINED3D_SHADER_TYPE_PIXEL;
}
 
static BOOL shader_is_vshader_version(enum wined3d_shader_type type)
{
    return type == WINED3D_SHADER_TYPE_VERTEX;
}
 
static const char *get_line(const char **ptr)
{
    const char *p, *q;
 
    p = *ptr;
    if (!(q = strstr(p, "\n")))
    {
        if (!*p) return NULL;
        *ptr += strlen(p);
        return p;
    }
    *ptr = q + 1;
 
    return p;
}
 
enum arb_helper_value
{
    ARB_ZERO,
    ARB_ONE,
    ARB_TWO,
    ARB_0001,
    ARB_EPS,
 
    ARB_VS_REL_OFFSET
};
 
static const char *arb_get_helper_value(enum wined3d_shader_type shader, enum arb_helper_value value)
{
    if (shader != WINED3D_SHADER_TYPE_VERTEX && shader != WINED3D_SHADER_TYPE_PIXEL)
    {
        ERR("Unsupported shader type '%s'.\n", debug_shader_type(shader));
        return "bad";
    }
 
    if (shader == WINED3D_SHADER_TYPE_PIXEL)
    {
        switch (value)
        {
            case ARB_ZERO: return "ps_helper_const.x";
            case ARB_ONE: return "ps_helper_const.y";
            case ARB_TWO: return "coefmul.x";
            case ARB_0001: return "ps_helper_const.xxxy";
            case ARB_EPS: return "ps_helper_const.z";
            default: break;
        }
    }
    else
    {
        switch (value)
        {
            case ARB_ZERO: return "helper_const.x";
            case ARB_ONE: return "helper_const.y";
            case ARB_TWO: return "helper_const.z";
            case ARB_EPS: return "helper_const.w";
            case ARB_0001: return "helper_const.xxxy";
            case ARB_VS_REL_OFFSET: return "rel_addr_const.y";
        }
    }
    FIXME("Unmanaged %s shader helper constant requested: %u.\n",
          shader == WINED3D_SHADER_TYPE_PIXEL ? "pixel" : "vertex", value);
    switch (value)
    {
        case ARB_ZERO: return "0.0";
        case ARB_ONE: return "1.0";
        case ARB_TWO: return "2.0";
        case ARB_0001: return "{0.0, 0.0, 0.0, 1.0}";
        case ARB_EPS: return "1e-8";
        default: return "bad";
    }
}
 
static inline BOOL ffp_clip_emul(const struct wined3d_context *context)
{
    return context->lowest_disabled_stage < 7;
}
 
/* ARB_program_shader private data */
 
struct control_frame
{
    struct                          list entry;
    enum
    {
        IF,
        IFC,
        LOOP,
        REP
    } type;
    BOOL                            muting;
    BOOL                            outer_loop;
    union
    {
        unsigned int                loop;
        unsigned int                ifc;
    } no;
    struct wined3d_shader_loop_control loop_control;
    BOOL                            had_else;
};
 
struct arb_ps_np2fixup_info
{
    struct ps_np2fixup_info         super;
    /* For ARB we need an offset value:
     * With both GLSL and ARB mode the NP2 fixup information (the texture dimensions) are stored in a
     * consecutive way (GLSL uses a uniform array). Since ARB doesn't know the notion of a "standalone"
     * array we need an offset to the index inside the program local parameter array. */
    UINT                            offset;
};
 
struct arb_ps_compile_args
{
    struct ps_compile_args          super;
    WORD                            bools;
    WORD                            clip;  /* only a boolean, use a WORD for alignment */
    unsigned char                   loop_ctrl[WINED3D_MAX_CONSTS_I][3];
};
 
struct stb_const_desc
{
    unsigned char           texunit;
    UINT                    const_num;
};
 
struct arb_ps_compiled_shader
{
    struct arb_ps_compile_args      args;
    struct arb_ps_np2fixup_info     np2fixup_info;
    struct stb_const_desc           bumpenvmatconst[MAX_TEXTURES];
    struct stb_const_desc           luminanceconst[MAX_TEXTURES];
    UINT                            int_consts[WINED3D_MAX_CONSTS_I];
    GLuint                          prgId;
    UINT                            ycorrection;
    unsigned char                   numbumpenvmatconsts;
    char                            num_int_consts;
};
 
struct arb_vs_compile_args
{
    struct vs_compile_args          super;
    union
    {
        struct
        {
            WORD                    bools;
            unsigned char           clip_texcoord;
            unsigned char           clipplane_mask;
        }                           boolclip;
        DWORD                       boolclip_compare;
    } clip;
    DWORD                           ps_signature;
    union
    {
        unsigned char               samplers[4];
        DWORD                       samplers_compare;
    } vertex;
    unsigned char                   loop_ctrl[WINED3D_MAX_CONSTS_I][3];
};
 
struct arb_vs_compiled_shader
{
    struct arb_vs_compile_args      args;
    GLuint                          prgId;
    UINT                            int_consts[WINED3D_MAX_CONSTS_I];
    char                            num_int_consts;
    char                            need_color_unclamp;
    UINT                            pos_fixup;
};
 
struct recorded_instruction
{
    struct wined3d_shader_instruction ins;
    struct list entry;
};
 
struct shader_arb_ctx_priv
{
    char addr_reg[20];
    enum
    {
        /* plain GL_ARB_vertex_program or GL_ARB_fragment_program */
        ARB,
        /* GL_NV_vertex_program2_option or GL_NV_fragment_program_option */
        NV2,
        /* GL_NV_vertex_program3 or GL_NV_fragment_program2 */
        NV3
    } target_version;
 
    const struct arb_vs_compile_args    *cur_vs_args;
    const struct arb_ps_compile_args    *cur_ps_args;
    const struct arb_ps_compiled_shader *compiled_fprog;
    const struct arb_vs_compiled_shader *compiled_vprog;
    struct arb_ps_np2fixup_info         *cur_np2fixup_info;
    struct list                         control_frames;
    struct list                         record;
    BOOL                                recording;
    BOOL                                muted;
    unsigned int                        num_loops, loop_depth, num_ifcs;
    int                                 aL;
    BOOL                                ps_post_process;
 
    unsigned int                        vs_clipplanes;
    BOOL                                footer_written;
    BOOL                                in_main_func;
 
    /* For 3.0 vertex shaders */
    const char                          *vs_output[MAX_REG_OUTPUT];
    /* For 2.x and earlier vertex shaders */
    const char                          *texcrd_output[8], *color_output[2], *fog_output;
 
    /* 3.0 pshader input for compatibility with fixed function */
    const char                          *ps_input[MAX_REG_INPUT];
};
 
struct ps_signature
{
    struct wined3d_shader_signature sig;
    DWORD                               idx;
    struct wine_rb_entry                entry;
};
 
struct arb_pshader_private {
    struct arb_ps_compiled_shader   *gl_shaders;
    UINT                            num_gl_shaders, shader_array_size;
    DWORD                           input_signature_idx;
    DWORD                           clipplane_emulation;
    BOOL                            clamp_consts;
};
 
struct arb_vshader_private {
    struct arb_vs_compiled_shader   *gl_shaders;
    UINT                            num_gl_shaders, shader_array_size;
    UINT rel_offset;
};
 
struct shader_arb_priv
{
    GLuint                  current_vprogram_id;
    GLuint                  current_fprogram_id;
    const struct arb_ps_compiled_shader *compiled_fprog;
    const struct arb_vs_compiled_shader *compiled_vprog;
    BOOL                    use_arbfp_fixed_func;
    struct wine_rb_tree     fragment_shaders;
    BOOL                    last_ps_const_clamped;
    BOOL                    last_vs_color_unclamp;
 
    struct wine_rb_tree     signature_tree;
    DWORD ps_sig_number;
 
    unsigned int highest_dirty_ps_const, highest_dirty_vs_const;
    char vshader_const_dirty[WINED3D_MAX_VS_CONSTS_F];
    char pshader_const_dirty[WINED3D_MAX_PS_CONSTS_F];
    const struct wined3d_context *last_context;
 
    const struct wined3d_vertex_pipe_ops *vertex_pipe;
    const struct fragment_pipeline *fragment_pipe;
    BOOL ffp_proj_control;
};
 
/* Context activation for state handlers is done by the caller. */
 
static BOOL need_rel_addr_const(const struct arb_vshader_private *shader_data,
        const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
{
    if (shader_data->rel_offset) return TRUE;
    if (!reg_maps->usesmova) return FALSE;
    return !gl_info->supported[NV_VERTEX_PROGRAM2_OPTION];
}
 
/* Returns TRUE if result.clip from GL_NV_vertex_program2 should be used and FALSE otherwise */
static inline BOOL use_nv_clip(const struct wined3d_gl_info *gl_info)
{
    return gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]
            && !(gl_info->quirks & WINED3D_QUIRK_NV_CLIP_BROKEN);
}
 
static BOOL need_helper_const(const struct arb_vshader_private *shader_data,
        const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
{
    if (need_rel_addr_const(shader_data, reg_maps, gl_info)) return TRUE;
    if (!gl_info->supported[NV_VERTEX_PROGRAM]) return TRUE; /* Need to init colors. */
    if (gl_info->quirks & WINED3D_QUIRK_ARB_VS_OFFSET_LIMIT) return TRUE; /* Load the immval offset. */
    if (gl_info->quirks & WINED3D_QUIRK_SET_TEXCOORD_W) return TRUE; /* Have to init texcoords. */
    if (!use_nv_clip(gl_info)) return TRUE; /* Init the clip texcoord */
    if (reg_maps->usesnrm) return TRUE; /* 0.0 */
    if (reg_maps->usespow) return TRUE; /* EPS, 0.0 and 1.0 */
    if (reg_maps->fog) return TRUE; /* Clamping fog coord, 0.0 and 1.0 */
    return FALSE;
}
 
static unsigned int reserved_vs_const(const struct arb_vshader_private *shader_data,
        const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info)
{
    unsigned int ret = 1;
    /* We use one PARAM for the pos fixup, and in some cases one to load
     * some immediate values into the shader. */
    if (need_helper_const(shader_data, reg_maps, gl_info)) ++ret;
    if (need_rel_addr_const(shader_data, reg_maps, gl_info)) ++ret;
    return ret;
}
 
/* Loads floating point constants into the currently set ARB_vertex/fragment_program.
 * When constant_list == NULL, it will load all the constants.
 *
 * @target_type should be either GL_VERTEX_PROGRAM_ARB (for vertex shaders)
 *  or GL_FRAGMENT_PROGRAM_ARB (for pixel shaders)
 */
/* Context activation is done by the caller. */
static unsigned int shader_arb_load_constants_f(const struct wined3d_shader *shader,
        const struct wined3d_gl_info *gl_info, GLuint target_type, unsigned int max_constants,
        const struct wined3d_vec4 *constants, char *dirty_consts)
{
    struct wined3d_shader_lconst *lconst;
    unsigned int ret, i, j;
 
    if (TRACE_ON(d3d_constants))
    {
        for (i = 0; i < max_constants; ++i)
        {
            if (!dirty_consts[i])
                continue;
            TRACE_(d3d_constants)("Loading constant %u: %s.\n", i, debug_vec4(&constants[i]));
        }
    }
 
    i = 0;
 
    /* In 1.X pixel shaders constants are implicitly clamped in the range [-1;1] */
    if (target_type == GL_FRAGMENT_PROGRAM_ARB && shader->reg_maps.shader_version.major == 1)
    {
        float lcl_const[4];
        /* ps 1.x supports only 8 constants, clamp only those. When switching between 1.x and higher
         * shaders, the first 8 constants are marked dirty for reload
         */
        for (; i < min(8, max_constants); ++i)
        {
            if (!dirty_consts[i])
                continue;
            dirty_consts[i] = 0;
 
            if (constants[i].x > 1.0f)
                lcl_const[0] = 1.0f;
            else if (constants[i].x < -1.0f)
                lcl_const[0] = -1.0f;
            else
                lcl_const[0] = constants[i].x;
 
            if (constants[i].y > 1.0f)
                lcl_const[1] = 1.0f;
            else if (constants[i].y < -1.0f)
                lcl_const[1] = -1.0f;
            else
                lcl_const[1] = constants[i].y;
 
            if (constants[i].z > 1.0f)
                lcl_const[2] = 1.0f;
            else if (constants[i].z < -1.0f)
                lcl_const[2] = -1.0f;
            else
                lcl_const[2] = constants[i].z;
 
            if (constants[i].w > 1.0f)
                lcl_const[3] = 1.0f;
            else if (constants[i].w < -1.0f)
                lcl_const[3] = -1.0f;
            else
                lcl_const[3] = constants[i].w;
 
            GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, lcl_const));
        }
 
        /* If further constants are dirty, reload them without clamping.
         *
         * The alternative is not to touch them, but then we cannot reset the dirty constant count
         * to zero. That's bad for apps that only use PS 1.x shaders, because in that case the code
         * above would always re-check the first 8 constants since max_constant remains at the init
         * value
         */
    }
 
    if (gl_info->supported[EXT_GPU_PROGRAM_PARAMETERS])
    {
        /* TODO: Benchmark if we're better of with finding the dirty constants ourselves,
         * or just reloading *all* constants at once
         *
        GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, i, max_constants, constants + (i * 4)));
         */
        for (; i < max_constants; ++i)
        {
            if (!dirty_consts[i])
                continue;
 
            /* Find the next block of dirty constants */
            dirty_consts[i] = 0;
            j = i;
            for (++i; (i < max_constants) && dirty_consts[i]; ++i)
            {
                dirty_consts[i] = 0;
            }
 
            GL_EXTCALL(glProgramEnvParameters4fvEXT(target_type, j, i - j, &constants[j].x));
        }
    }
    else
    {
        for (; i < max_constants; ++i)
        {
            if (dirty_consts[i])
            {
                dirty_consts[i] = 0;
                GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, i, &constants[i].x));
            }
        }
    }
    checkGLcall("glProgramEnvParameter4fvARB()");
 
    /* Load immediate constants */
    if (shader->load_local_constsF)
    {
        if (TRACE_ON(d3d_shader))
        {
            LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
            {
                GLfloat* values = (GLfloat*)lconst->value;
                TRACE_(d3d_constants)("Loading local constants %i: %f, %f, %f, %f\n", lconst->idx,
                        values[0], values[1], values[2], values[3]);
            }
        }
        /* Immediate constants are clamped for 1.X shaders at loading times */
        ret = 0;
        LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
        {
            dirty_consts[lconst->idx] = 1; /* Dirtify so the non-immediate constant overwrites it next time */
            ret = max(ret, lconst->idx + 1);
            GL_EXTCALL(glProgramEnvParameter4fvARB(target_type, lconst->idx, (GLfloat*)lconst->value));
        }
        checkGLcall("glProgramEnvParameter4fvARB()");
        return ret; /* The loaded immediate constants need reloading for the next shader */
    } else {
        return 0; /* No constants are dirty now */
    }
}
 
/* Loads the texture dimensions for NP2 fixup into the currently set
 * ARB_[vertex/fragment]_programs. */
static void shader_arb_load_np2fixup_constants(const struct arb_ps_np2fixup_info *fixup,
        const struct wined3d_gl_info *gl_info, const struct wined3d_state *state)
{
    GLfloat np2fixup_constants[4 * MAX_FRAGMENT_SAMPLERS];
    WORD active = fixup->super.active;
    UINT i;
 
    if (!active)
        return;
 
    for (i = 0; active; active >>= 1, ++i)
    {
        const struct wined3d_texture *tex = state->textures[i];
        unsigned char idx = fixup->super.idx[i];
        GLfloat *tex_dim = &np2fixup_constants[(idx >> 1) * 4];
 
        if (!(active & 1))
            continue;
 
        if (!tex)
        {
            ERR("Nonexistent texture is flagged for NP2 texcoord fixup.\n");
            continue;
        }
 
        if (idx % 2)
        {
            tex_dim[2] = tex->pow2_matrix[0];
            tex_dim[3] = tex->pow2_matrix[5];
        }
        else
        {
            tex_dim[0] = tex->pow2_matrix[0];
            tex_dim[1] = tex->pow2_matrix[5];
        }
    }
 
    for (i = 0; i < fixup->super.num_consts; ++i)
    {
        GL_EXTCALL(glProgramEnvParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
                fixup->offset + i, &np2fixup_constants[i * 4]));
    }
}
 
/* Context activation is done by the caller. */
static void shader_arb_ps_local_constants(const struct arb_ps_compiled_shader *gl_shader,
        const struct wined3d_context *context, const struct wined3d_state *state, UINT rt_height)
{
    const struct wined3d_gl_info *gl_info = context->gl_info;
    unsigned char i;
 
    for(i = 0; i < gl_shader->numbumpenvmatconsts; i++)
    {
        int texunit = gl_shader->bumpenvmatconst[i].texunit;
 
        /* The state manager takes care that this function is always called if the bump env matrix changes */
        const float *data = (const float *)&state->texture_states[texunit][WINED3D_TSS_BUMPENV_MAT00];
        GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
                gl_shader->bumpenvmatconst[i].const_num, data));
 
        if (gl_shader->luminanceconst[i].const_num != WINED3D_CONST_NUM_UNUSED)
        {
            /* WINED3D_TSS_BUMPENVLSCALE and WINED3D_TSS_BUMPENVLOFFSET are next to each other.
             * point gl to the scale, and load 4 floats. x = scale, y = offset, z and w are junk, we
             * don't care about them. The pointers are valid for sure because the stateblock is bigger.
             * (they're WINED3D_TSS_TEXTURETRANSFORMFLAGS and WINED3D_TSS_ADDRESSW, so most likely 0 or NaN
            */
            const float *scale = (const float *)&state->texture_states[texunit][WINED3D_TSS_BUMPENV_LSCALE];
            GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB,
                    gl_shader->luminanceconst[i].const_num, scale));
        }
    }
    checkGLcall("Load bumpmap consts");
 
    if(gl_shader->ycorrection != WINED3D_CONST_NUM_UNUSED)
    {
        /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
        * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
        * ycorrection.z: 1.0
        * ycorrection.w: 0.0
        */
        float val[4];
        val[0] = context->render_offscreen ? 0.0f : (float) rt_height;
        val[1] = context->render_offscreen ? 1.0f : -1.0f;
        val[2] = 1.0f;
        val[3] = 0.0f;
        GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->ycorrection, val));
        checkGLcall("y correction loading");
    }
 
    if (!gl_shader->num_int_consts) return;
 
    for (i = 0; i < WINED3D_MAX_CONSTS_I; ++i)
    {
        if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
        {
            float val[4];
            val[0] = (float)state->ps_consts_i[i].x;
            val[1] = (float)state->ps_consts_i[i].y;
            val[2] = (float)state->ps_consts_i[i].z;
            val[3] = -1.0f;
 
            GL_EXTCALL(glProgramLocalParameter4fvARB(GL_FRAGMENT_PROGRAM_ARB, gl_shader->int_consts[i], val));
        }
    }
    checkGLcall("Load ps int consts");
}
 
/* Context activation is done by the caller. */
static void shader_arb_vs_local_constants(const struct arb_vs_compiled_shader *gl_shader,
        const struct wined3d_context *context, const struct wined3d_state *state)
{
    const struct wined3d_gl_info *gl_info = context->gl_info;
    float position_fixup[4];
    unsigned char i;
 
    /* Upload the position fixup */
    shader_get_position_fixup(context, state, position_fixup);
    GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->pos_fixup, position_fixup));
 
    if (!gl_shader->num_int_consts) return;
 
    for (i = 0; i < WINED3D_MAX_CONSTS_I; ++i)
    {
        if(gl_shader->int_consts[i] != WINED3D_CONST_NUM_UNUSED)
        {
            float val[4];
            val[0] = (float)state->vs_consts_i[i].x;
            val[1] = (float)state->vs_consts_i[i].y;
            val[2] = (float)state->vs_consts_i[i].z;
            val[3] = -1.0f;
 
            GL_EXTCALL(glProgramLocalParameter4fvARB(GL_VERTEX_PROGRAM_ARB, gl_shader->int_consts[i], val));
        }
    }
    checkGLcall("Load vs int consts");
}
 
static void shader_arb_select(void *shader_priv, struct wined3d_context *context,
        const struct wined3d_state *state);
 
/* Context activation is done by the caller (state handler). */
static void shader_arb_load_constants_internal(struct shader_arb_priv *priv,
        struct wined3d_context *context, const struct wined3d_state *state,
        BOOL usePixelShader, BOOL useVertexShader, BOOL from_shader_select)
{
    const struct wined3d_d3d_info *d3d_info = context->d3d_info;
    const struct wined3d_gl_info *gl_info = context->gl_info;
 
    if (!from_shader_select)
    {
        const struct wined3d_shader *vshader = state->shader[WINED3D_SHADER_TYPE_VERTEX];
        const struct wined3d_shader *pshader = state->shader[WINED3D_SHADER_TYPE_PIXEL];
 
        if (vshader
                && (vshader->reg_maps.boolean_constants
                || (!gl_info->supported[NV_VERTEX_PROGRAM2_OPTION]
                && (vshader->reg_maps.integer_constants & ~vshader->reg_maps.local_int_consts))))
        {
            TRACE("bool/integer vertex shader constants potentially modified, forcing shader reselection.\n");
            shader_arb_select(priv, context, state);
        }
        else if (pshader
                && (pshader->reg_maps.boolean_constants
                || (!gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION]
                && (pshader->reg_maps.integer_constants & ~pshader->reg_maps.local_int_consts))))
        {
            TRACE("bool/integer pixel shader constants potentially modified, forcing shader reselection.\n");
            shader_arb_select(priv, context, state);
        }
    }
 
    if (context != priv->last_context)
    {
        memset(priv->vshader_const_dirty, 1,
                sizeof(*priv->vshader_const_dirty) * d3d_info->limits.vs_uniform_count);
        priv->highest_dirty_vs_const = d3d_info->limits.vs_uniform_count;
 
        memset(priv->pshader_const_dirty, 1,
                sizeof(*priv->pshader_const_dirty) * d3d_info->limits.ps_uniform_count);
        priv->highest_dirty_ps_const = d3d_info->limits.ps_uniform_count;
 
        priv->last_context = context;
    }
 
    if (useVertexShader)
    {
        const struct wined3d_shader *vshader = state->shader[WINED3D_SHADER_TYPE_VERTEX];
        const struct arb_vs_compiled_shader *gl_shader = priv->compiled_vprog;
 
        /* Load DirectX 9 float constants for vertex shader */
        priv->highest_dirty_vs_const = shader_arb_load_constants_f(vshader, gl_info, GL_VERTEX_PROGRAM_ARB,
                priv->highest_dirty_vs_const, state->vs_consts_f, priv->vshader_const_dirty);
        shader_arb_vs_local_constants(gl_shader, context, state);
    }
 
    if (usePixelShader)
    {
        const struct wined3d_shader *pshader = state->shader[WINED3D_SHADER_TYPE_PIXEL];
        const struct arb_ps_compiled_shader *gl_shader = priv->compiled_fprog;
        UINT rt_height = state->fb->render_targets[0]->height;
 
        /* Load DirectX 9 float constants for pixel shader */
        priv->highest_dirty_ps_const = shader_arb_load_constants_f(pshader, gl_info, GL_FRAGMENT_PROGRAM_ARB,
                priv->highest_dirty_ps_const, state->ps_consts_f, priv->pshader_const_dirty);
        shader_arb_ps_local_constants(gl_shader, context, state, rt_height);
 
        if (context->constant_update_mask & WINED3D_SHADER_CONST_PS_NP2_FIXUP)
            shader_arb_load_np2fixup_constants(&gl_shader->np2fixup_info, gl_info, state);
    }
}
 
static void shader_arb_load_constants(void *shader_priv, struct wined3d_context *context,
        const struct wined3d_state *state)
{
    BOOL vs = use_vs(state);
    BOOL ps = use_ps(state);
 
    shader_arb_load_constants_internal(shader_priv, context, state, ps, vs, FALSE);
}
 
static void shader_arb_update_float_vertex_constants(struct wined3d_device *device, UINT start, UINT count)
{
    struct wined3d_context *context = context_get_current();
    struct shader_arb_priv *priv = device->shader_priv;
 
    /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
     * context. On a context switch the old context will be fully dirtified */
    if (!context || context->device != device)
        return;
 
    memset(priv->vshader_const_dirty + start, 1, sizeof(*priv->vshader_const_dirty) * count);
    priv->highest_dirty_vs_const = max(priv->highest_dirty_vs_const, start + count);
}
 
static void shader_arb_update_float_pixel_constants(struct wined3d_device *device, UINT start, UINT count)
{
    struct wined3d_context *context = context_get_current();
    struct shader_arb_priv *priv = device->shader_priv;
 
    /* We don't want shader constant dirtification to be an O(contexts), so just dirtify the active
     * context. On a context switch the old context will be fully dirtified */
    if (!context || context->device != device)
        return;
 
    memset(priv->pshader_const_dirty + start, 1, sizeof(*priv->pshader_const_dirty) * count);
    priv->highest_dirty_ps_const = max(priv->highest_dirty_ps_const, start + count);
}
 
static void shader_arb_append_imm_vec4(struct wined3d_string_buffer *buffer, const float *values)
{
    char str[4][17];
 
    wined3d_ftoa(values[0], str[0]);
    wined3d_ftoa(values[1], str[1]);
    wined3d_ftoa(values[2], str[2]);
    wined3d_ftoa(values[3], str[3]);
    shader_addline(buffer, "{%s, %s, %s, %s}", str[0], str[1], str[2], str[3]);
}
 
/* Generate the variable & register declarations for the ARB_vertex_program output target */
static void shader_generate_arb_declarations(const struct wined3d_shader *shader,
        const struct wined3d_shader_reg_maps *reg_maps, struct wined3d_string_buffer *buffer,
        const struct wined3d_gl_info *gl_info, DWORD *num_clipplanes,
        const struct shader_arb_ctx_priv *ctx)
{
    DWORD i;
    char pshader = shader_is_pshader_version(reg_maps->shader_version.type);
    const struct wined3d_shader_lconst *lconst;
    unsigned max_constantsF;
    DWORD map;
 
    /* In pixel shaders, all private constants are program local, we don't need anything
     * from program.env. Thus we can advertise the full set of constants in pixel shaders.
     * If we need a private constant the GL implementation will squeeze it in somewhere
     *
     * With vertex shaders we need the posFixup and on some GL implementations 4 helper
     * immediate values. The posFixup is loaded using program.env for now, so always
     * subtract one from the number of constants. If the shader uses indirect addressing,
     * account for the helper const too because we have to declare all available d3d constants
     * and don't know which are actually used.
     */
    if (pshader)
    {
        max_constantsF = gl_info->limits.arb_ps_native_constants;
        /* 24 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value. */
        if (max_constantsF < 24)
            max_constantsF = gl_info->limits.arb_ps_float_constants;
    }
    else
    {
        const struct arb_vshader_private *shader_data = shader->backend_data;
        max_constantsF = gl_info->limits.arb_vs_native_constants;
        /* 96 is the minimum MAX_PROGRAM_ENV_PARAMETERS_ARB value.
         * Also prevents max_constantsF from becoming less than 0 and
         * wrapping . */
        if (max_constantsF < 96)
            max_constantsF = gl_info->limits.arb_vs_float_constants;
 
        if (reg_maps->usesrelconstF)
        {
            DWORD highest_constf = 0, clip_limit;
 
            max_constantsF -= reserved_vs_const(shader_data, reg_maps, gl_info);
            max_constantsF -= wined3d_popcount(reg_maps->integer_constants);
            max_constantsF -= gl_info->reserved_arb_constants;
 
            for (i = 0; i < shader->limits->constant_float; ++i)
            {
                DWORD idx = i >> 5;
                DWORD shift = i & 0x1f;
                if (reg_maps->constf[idx] & (1u << shift))
                    highest_constf = i;
            }
 
            if(use_nv_clip(gl_info) && ctx->target_version >= NV2)
            {
                if(ctx->cur_vs_args->super.clip_enabled)
                    clip_limit = gl_info->limits.user_clip_distances;
                else
                    clip_limit = 0;
            }
            else
            {
                unsigned int mask = ctx->cur_vs_args->clip.boolclip.clipplane_mask;
                clip_limit = min(wined3d_popcount(mask), 4);
            }
            *num_clipplanes = min(clip_limit, max_constantsF - highest_constf - 1);
            max_constantsF -= *num_clipplanes;
            if(*num_clipplanes < clip_limit)
            {
                WARN("Only %u clip planes out of %u enabled.\n", *num_clipplanes,
                        gl_info->limits.user_clip_distances);
            }
        }
        else
        {
            if (ctx->target_version >= NV2)
                *num_clipplanes = gl_info->limits.user_clip_distances;
            else
                *num_clipplanes = min(gl_info->limits.user_clip_distances, 4);
        }
    }
 
    for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
    {
        if (map & 1) shader_addline(buffer, "TEMP R%u;\n", i);
    }
 
    for (i = 0, map = reg_maps->address; map; map >>= 1, ++i)
    {
        if (map & 1) shader_addline(buffer, "ADDRESS A%u;\n", i);
    }
 
    if (pshader && reg_maps->shader_version.major == 1 && reg_maps->shader_version.minor <= 3)
    {
        for (i = 0, map = reg_maps->texcoord; map; map >>= 1, ++i)
        {
            if (map & 1) shader_addline(buffer, "TEMP T%u;\n", i);
        }
    }
 
    if (!shader->load_local_constsF)
    {
        LIST_FOR_EACH_ENTRY(lconst, &shader->constantsF, struct wined3d_shader_lconst, entry)
        {
            const float *value;
            value = (const float *)lconst->value;
            shader_addline(buffer, "PARAM C%u = ", lconst->idx);
            shader_arb_append_imm_vec4(buffer, value);
            shader_addline(buffer, ";\n");
        }
    }
 
    /* After subtracting privately used constants from the hardware limit(they are loaded as
     * local constants), make sure the shader doesn't violate the env constant limit
     */
    if (pshader)
    {
        max_constantsF = min(max_constantsF, gl_info->limits.arb_ps_float_constants);
    }
    else
    {
        max_constantsF = min(max_constantsF, gl_info->limits.arb_vs_float_constants);
    }
 
    /* Avoid declaring more constants than needed */
    max_constantsF = min(max_constantsF, shader->limits->constant_float);
 
    /* we use the array-based constants array if the local constants are marked for loading,
     * because then we use indirect addressing, or when the local constant list is empty,
     * because then we don't know if we're using indirect addressing or not. If we're hardcoding
     * local constants do not declare the loaded constants as an array because ARB compilers usually
     * do not optimize unused constants away
     */
    if (reg_maps->usesrelconstF)
    {
        /* Need to PARAM the environment parameters (constants) so we can use relative addressing */
        shader_addline(buffer, "PARAM C[%d] = { program.env[0..%d] };\n",
                    max_constantsF, max_constantsF - 1);
    }
    else
    {
        for (i = 0; i < max_constantsF; ++i)
        {
            if (!shader_constant_is_local(shader, i) && wined3d_extract_bits(reg_maps->constf, i, 1))
            {
                shader_addline(buffer, "PARAM C%d = program.env[%d];\n",i, i);
            }
        }
    }
}
 
static const char * const shift_tab[] = {
    "dummy",     /*  0 (none) */
    "coefmul.x", /*  1 (x2)   */
    "coefmul.y", /*  2 (x4)   */
    "coefmul.z", /*  3 (x8)   */
    "coefmul.w", /*  4 (x16)  */
    "dummy",     /*  5 (x32)  */
    "dummy",     /*  6 (x64)  */
    "dummy",     /*  7 (x128) */
    "dummy",     /*  8 (d256) */
    "dummy",     /*  9 (d128) */
    "dummy",     /* 10 (d64)  */
    "dummy",     /* 11 (d32)  */
    "coefdiv.w", /* 12 (d16)  */
    "coefdiv.z", /* 13 (d8)   */
    "coefdiv.y", /* 14 (d4)   */
    "coefdiv.x"  /* 15 (d2)   */
};
 
static void shader_arb_get_write_mask(const struct wined3d_shader_instruction *ins,
        const struct wined3d_shader_dst_param *dst, char *write_mask)
{
    char *ptr = write_mask;
 
    if (dst->write_mask != WINED3DSP_WRITEMASK_ALL)
    {
        *ptr++ = '.';
        if (dst->write_mask & WINED3DSP_WRITEMASK_0) *ptr++ = 'x';
        if (dst->write_mask & WINED3DSP_WRITEMASK_1) *ptr++ = 'y';
        if (dst->write_mask & WINED3DSP_WRITEMASK_2) *ptr++ = 'z';
        if (dst->write_mask & WINED3DSP_WRITEMASK_3) *ptr++ = 'w';
    }
 
    *ptr = '\0';
}
 
static void shader_arb_get_swizzle(const struct wined3d_shader_src_param *param, BOOL fixup, char *swizzle_str)
{
    /* For registers of type WINED3DDECLTYPE_D3DCOLOR, data is stored as "bgra",
     * but addressed as "rgba". To fix this we need to swap the register's x
     * and z components. */
    const char *swizzle_chars = fixup ? "zyxw" : "xyzw";
    char *ptr = swizzle_str;
 
    /* swizzle bits fields: wwzzyyxx */
    DWORD swizzle = param->swizzle;
    DWORD swizzle_x = swizzle & 0x03;
    DWORD swizzle_y = (swizzle >> 2) & 0x03;
    DWORD swizzle_z = (swizzle >> 4) & 0x03;
    DWORD swizzle_w = (swizzle >> 6) & 0x03;
 
    /* If the swizzle is the default swizzle (ie, "xyzw"), we don't need to
     * generate a swizzle string. Unless we need to our own swizzling. */
    if (swizzle != WINED3DSP_NOSWIZZLE || fixup)
    {
        *ptr++ = '.';
        if (swizzle_x == swizzle_y && swizzle_x == swizzle_z && swizzle_x == swizzle_w) {
            *ptr++ = swizzle_chars[swizzle_x];
        } else {
            *ptr++ = swizzle_chars[swizzle_x];
            *ptr++ = swizzle_chars[swizzle_y];
            *ptr++ = swizzle_chars[swizzle_z];
            *ptr++ = swizzle_chars[swizzle_w];
        }
    }
 
    *ptr = '\0';
}
 
static void shader_arb_request_a0(const struct wined3d_shader_instruction *ins, const char *src)
{
    struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
 
    if (!strcmp(priv->addr_reg, src)) return;
 
    strcpy(priv->addr_reg, src);
    shader_addline(buffer, "ARL A0.x, %s;\n", src);
}
 
static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
        const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr);
 
static void shader_arb_get_register_name(const struct wined3d_shader_instruction *ins,
        const struct wined3d_shader_register *reg, char *register_name, BOOL *is_color)
{
    /* oPos, oFog and oPts in D3D */
    static const char * const rastout_reg_names[] = {"TMP_OUT", "TMP_FOGCOORD", "result.pointsize"};
    const struct wined3d_shader *shader = ins->ctx->shader;
    const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
    BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
    struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
 
    *is_color = FALSE;
 
    switch (reg->type)
    {
        case WINED3DSPR_TEMP:
            sprintf(register_name, "R%u", reg->idx[0].offset);
            break;
 
        case WINED3DSPR_INPUT:
            if (pshader)
            {
                if (reg_maps->shader_version.major < 3)
                {
                    if (!reg->idx[0].offset)
                        strcpy(register_name, "fragment.color.primary");
                    else
                        strcpy(register_name, "fragment.color.secondary");
                }
                else
                {
                    if (reg->idx[0].rel_addr)
                    {
                        char rel_reg[50];
                        shader_arb_get_src_param(ins, reg->idx[0].rel_addr, 0, rel_reg);
 
                        if (!strcmp(rel_reg, "**aL_emul**"))
                        {
                            DWORD idx = ctx->aL + reg->idx[0].offset;
                            if(idx < MAX_REG_INPUT)
                            {
                                strcpy(register_name, ctx->ps_input[idx]);
                            }
                            else
                            {
                                ERR("Pixel shader input register out of bounds: %u\n", idx);
                                sprintf(register_name, "out_of_bounds_%u", idx);
                            }
                        }
                        else if (reg_maps->input_registers & 0x0300)
                        {
                            /* There are two ways basically:
                             *
                             * 1) Use the unrolling code that is used for loop emulation and unroll the loop.
                             *    That means trouble if the loop also contains a breakc or if the control values
                             *    aren't local constants.
                             * 2) Generate an if block that checks if aL.y < 8, == 8 or == 9 and selects the
                             *    source dynamically. The trouble is that we cannot simply read aL.y because it
                             *    is an ADDRESS register. We could however push it, load .zw with a value and use
                             *    ADAC to load the condition code register and pop it again afterwards
                             */
                            FIXME("Relative input register addressing with more than 8 registers\n");
 
                            /* This is better than nothing for now */
                            sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx[0].offset);
                        }
                        else if(ctx->cur_ps_args->super.vp_mode != vertexshader)
                        {
                            /* This is problematic because we'd have to consult the ctx->ps_input strings
                             * for where to find the varying. Some may be "0.0", others can be texcoords or
                             * colors. This needs either a pipeline replacement to make the vertex shader feed
                             * proper varyings, or loop unrolling
                             *
                             * For now use the texcoords and hope for the best
                             */
                            FIXME("Non-vertex shader varying input with indirect addressing\n");
                            sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx[0].offset);
                        }
                        else
                        {
                            /* D3D supports indirect addressing only with aL in loop registers. The loop instruction
                             * pulls GL_NV_fragment_program2 in
                             */
                            sprintf(register_name, "fragment.texcoord[%s + %u]", rel_reg, reg->idx[0].offset);
                        }
                    }
                    else
                    {
                        if (reg->idx[0].offset < MAX_REG_INPUT)
                        {
                            strcpy(register_name, ctx->ps_input[reg->idx[0].offset]);
                        }
                        else
                        {
                            ERR("Pixel shader input register out of bounds: %u\n", reg->idx[0].offset);
                            sprintf(register_name, "out_of_bounds_%u", reg->idx[0].offset);
                        }
                    }
                }
            }
            else
            {
                if (ctx->cur_vs_args->super.swizzle_map & (1u << reg->idx[0].offset))
                    *is_color = TRUE;
                sprintf(register_name, "vertex.attrib[%u]", reg->idx[0].offset);
            }
            break;
 
        case WINED3DSPR_CONST:
            if (!pshader && reg->idx[0].rel_addr)
            {
                const struct arb_vshader_private *shader_data = shader->backend_data;
                UINT rel_offset = ctx->target_version == ARB ? shader_data->rel_offset : 0;
                BOOL aL = FALSE;
                char rel_reg[50];
                if (reg_maps->shader_version.major < 2)
                {
                    sprintf(rel_reg, "A0.x");
                }
                else
                {
                    shader_arb_get_src_param(ins, reg->idx[0].rel_addr, 0, rel_reg);
                    if (ctx->target_version == ARB)
                    {
                        if (!strcmp(rel_reg, "**aL_emul**"))
                        {
                            aL = TRUE;
                        } else {
                            shader_arb_request_a0(ins, rel_reg);
                            sprintf(rel_reg, "A0.x");
                        }
                    }
                }
                if (aL)
                    sprintf(register_name, "C[%u]", ctx->aL + reg->idx[0].offset);
                else if (reg->idx[0].offset >= rel_offset)
                    sprintf(register_name, "C[%s + %u]", rel_reg, reg->idx[0].offset - rel_offset);
                else
                    sprintf(register_name, "C[%s - %u]", rel_reg, rel_offset - reg->idx[0].offset);
            }
            else
            {
                if (reg_maps->usesrelconstF)
                    sprintf(register_name, "C[%u]", reg->idx[0].offset);
                else
                    sprintf(register_name, "C%u", reg->idx[0].offset);
            }
            break;
 
        case WINED3DSPR_TEXTURE: /* case WINED3DSPR_ADDR: */
            if (pshader)
            {
                if (reg_maps->shader_version.major == 1
                        && reg_maps->shader_version.minor <= 3)
                    /* In ps <= 1.3, Tx is a temporary register as destination
                     * to all instructions, and as source to most instructions.
                     * For some instructions it is the texcoord input. Those
                     * instructions know about the special use. */
                    sprintf(register_name, "T%u", reg->idx[0].offset);
                else
                    /* In ps 1.4 and 2.x Tx is always a (read-only) varying. */
                    sprintf(register_name, "fragment.texcoord[%u]", reg->idx[0].offset);
            }
            else
            {
                if (reg_maps->shader_version.major == 1 || ctx->target_version >= NV2)
                    sprintf(register_name, "A%u", reg->idx[0].offset);
                else
                    sprintf(register_name, "A%u_SHADOW", reg->idx[0].offset);
            }
            break;
 
        case WINED3DSPR_COLOROUT:
            if (ctx->ps_post_process && !reg->idx[0].offset)
            {
                strcpy(register_name, "TMP_COLOR");
            }
            else
            {
                if (ctx->cur_ps_args->super.srgb_correction)
                    FIXME("sRGB correction on higher render targets.\n");
                if (reg_maps->rt_mask > 1)
                    sprintf(register_name, "result.color[%u]", reg->idx[0].offset);
                else
                    strcpy(register_name, "result.color");
            }
            break;
 
        case WINED3DSPR_RASTOUT:
            if (reg->idx[0].offset == 1)
                sprintf(register_name, "%s", ctx->fog_output);
            else
                sprintf(register_name, "%s", rastout_reg_names[reg->idx[0].offset]);
            break;
 
        case WINED3DSPR_DEPTHOUT:
            strcpy(register_name, "result.depth");
            break;
 
        case WINED3DSPR_ATTROUT:
        /* case WINED3DSPR_OUTPUT: */
            if (pshader)
                sprintf(register_name, "oD[%u]", reg->idx[0].offset);
            else
                strcpy(register_name, ctx->color_output[reg->idx[0].offset]);
            break;
 
        case WINED3DSPR_TEXCRDOUT:
            if (pshader)
                sprintf(register_name, "oT[%u]", reg->idx[0].offset);
            else if (reg_maps->shader_version.major < 3)
                strcpy(register_name, ctx->texcrd_output[reg->idx[0].offset]);
            else
                strcpy(register_name, ctx->vs_output[reg->idx[0].offset]);
            break;
 
        case WINED3DSPR_LOOP:
            if(ctx->target_version >= NV2)
            {
                /* Pshader has an implicitly declared loop index counter A0.x that cannot be renamed */
                if(pshader) sprintf(register_name, "A0.x");
                else sprintf(register_name, "aL.y");
            }
            else
            {
                /* Unfortunately this code cannot return the value of ctx->aL here. An immediate value
                 * would be valid, but if aL is used for indexing(its only use), there's likely an offset,
                 * thus the result would be something like C[15 + 30], which is not valid in the ARB program
                 * grammar. So return a marker for the emulated aL and intercept it in constant and varying
                 * indexing
                 */
                sprintf(register_name, "**aL_emul**");
            }
 
            break;
 
        case WINED3DSPR_CONSTINT:
            sprintf(register_name, "I%u", reg->idx[0].offset);
            break;
 
        case WINED3DSPR_MISCTYPE:
            if (!reg->idx[0].offset)
                sprintf(register_name, "vpos");
            else if (reg->idx[0].offset == 1)
                sprintf(register_name, "fragment.facing.x");
            else
                FIXME("Unknown MISCTYPE register index %u.\n", reg->idx[0].offset);
            break;
 
        default:
            FIXME("Unhandled register type %#x[%u].\n", reg->type, reg->idx[0].offset);
            sprintf(register_name, "unrecognized_register[%u]", reg->idx[0].offset);
            break;
    }
}
 
static void shader_arb_get_dst_param(const struct wined3d_shader_instruction *ins,
        const struct wined3d_shader_dst_param *wined3d_dst, char *str)
{
    char register_name[255];
    char write_mask[6];
    BOOL is_color;
 
    shader_arb_get_register_name(ins, &wined3d_dst->reg, register_name, &is_color);
    strcpy(str, register_name);
 
    shader_arb_get_write_mask(ins, wined3d_dst, write_mask);
    strcat(str, write_mask);
}
 
static const char *shader_arb_get_fixup_swizzle(enum fixup_channel_source channel_source)
{
    switch(channel_source)
    {
        case CHANNEL_SOURCE_ZERO: return "0";
        case CHANNEL_SOURCE_ONE: return "1";
        case CHANNEL_SOURCE_X: return "x";
        case CHANNEL_SOURCE_Y: return "y";
        case CHANNEL_SOURCE_Z: return "z";
        case CHANNEL_SOURCE_W: return "w";
        default:
            FIXME("Unhandled channel source %#x\n", channel_source);
            return "undefined";
    }
}
 
struct color_fixup_masks
{
    DWORD source;
    DWORD sign;
};
 
static struct color_fixup_masks calc_color_correction(struct color_fixup_desc fixup, DWORD dst_mask)
{
    struct color_fixup_masks masks = {0, 0};
 
    if (is_complex_fixup(fixup))
    {
        enum complex_fixup complex_fixup = get_complex_fixup(fixup);
        FIXME("Complex fixup (%#x) not supported\n", complex_fixup);
        return masks;
    }
 
    if (fixup.x_source != CHANNEL_SOURCE_X)
        masks.source |= WINED3DSP_WRITEMASK_0;
    if (fixup.y_source != CHANNEL_SOURCE_Y)
        masks.source |= WINED3DSP_WRITEMASK_1;
    if (fixup.z_source != CHANNEL_SOURCE_Z)
        masks.source |= WINED3DSP_WRITEMASK_2;
    if (fixup.w_source != CHANNEL_SOURCE_W)
        masks.source |= WINED3DSP_WRITEMASK_3;
    masks.source &= dst_mask;
 
    if (fixup.x_sign_fixup)
        masks.sign |= WINED3DSP_WRITEMASK_0;
    if (fixup.y_sign_fixup)
        masks.sign |= WINED3DSP_WRITEMASK_1;
    if (fixup.z_sign_fixup)
        masks.sign |= WINED3DSP_WRITEMASK_2;
    if (fixup.w_sign_fixup)
        masks.sign |= WINED3DSP_WRITEMASK_3;
    masks.sign &= dst_mask;
 
    return masks;
}
 
static void gen_color_correction(struct wined3d_string_buffer *buffer, const char *dst,
        const char *src, const char *one, const char *two,
        struct color_fixup_desc fixup, struct color_fixup_masks masks)
{
    const char *sign_fixup_src = dst;
 
    if (masks.source)
    {
        if (masks.sign)
            sign_fixup_src = "TA";
 
        shader_addline(buffer, "SWZ %s, %s, %s, %s, %s, %s;\n", sign_fixup_src, src,
                shader_arb_get_fixup_swizzle(fixup.x_source), shader_arb_get_fixup_swizzle(fixup.y_source),
                shader_arb_get_fixup_swizzle(fixup.z_source), shader_arb_get_fixup_swizzle(fixup.w_source));
    }
    else if (masks.sign)
    {
        sign_fixup_src = src;
    }
 
    if (masks.sign)
    {
        char reg_mask[6];
        char *ptr = reg_mask;
 
        if (masks.sign != WINED3DSP_WRITEMASK_ALL)
        {
            *ptr++ = '.';
            if (masks.sign & WINED3DSP_WRITEMASK_0)
                *ptr++ = 'x';
            if (masks.sign & WINED3DSP_WRITEMASK_1)
                *ptr++ = 'y';
            if (masks.sign & WINED3DSP_WRITEMASK_2)
                *ptr++ = 'z';
            if (masks.sign & WINED3DSP_WRITEMASK_3)
                *ptr++ = 'w';
        }
        *ptr = '\0';
 
        shader_addline(buffer, "MAD %s%s, %s, %s, -%s;\n", dst, reg_mask, sign_fixup_src, two, one);
    }
}
 
static const char *shader_arb_get_modifier(const struct wined3d_shader_instruction *ins)
{
    DWORD mod;
    struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
    if (!ins->dst_count) return "";
 
    mod = ins->dst[0].modifiers;
 
    /* Silently ignore PARTIALPRECISION if it's not supported */
    if(priv->target_version == ARB) mod &= ~WINED3DSPDM_PARTIALPRECISION;
 
    if(mod & WINED3DSPDM_MSAMPCENTROID)
    {
        FIXME("Unhandled modifier WINED3DSPDM_MSAMPCENTROID\n");
        mod &= ~WINED3DSPDM_MSAMPCENTROID;
    }
 
    switch(mod)
    {
        case WINED3DSPDM_SATURATE | WINED3DSPDM_PARTIALPRECISION:
            return "H_SAT";
 
        case WINED3DSPDM_SATURATE:
            return "_SAT";
 
        case WINED3DSPDM_PARTIALPRECISION:
            return "H";
 
        case 0:
            return "";
 
        default:
            FIXME("Unknown modifiers 0x%08x\n", mod);
            return "";
    }
}
 
#define TEX_PROJ        0x1
#define TEX_BIAS        0x2
#define TEX_LOD         0x4
#define TEX_DERIV       0x10
 
static void shader_hw_sample(const struct wined3d_shader_instruction *ins, DWORD sampler_idx,
        const char *dst_str, const char *coord_reg, WORD flags, const char *dsx, const char *dsy)
{
    enum wined3d_shader_resource_type resource_type = ins->ctx->reg_maps->resource_info[sampler_idx].type;
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    const char *tex_type;
    BOOL np2_fixup = FALSE;
    struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
    const char *mod;
    BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
    const struct wined3d_shader *shader;
    const struct wined3d_device *device;
    const struct wined3d_gl_info *gl_info;
    const char *tex_dst = dst_str;
    struct color_fixup_masks masks;
 
    /* D3D vertex shader sampler IDs are vertex samplers(0-3), not global d3d samplers */
    if(!pshader) sampler_idx += MAX_FRAGMENT_SAMPLERS;
 
    switch (resource_type)
    {
        case WINED3D_SHADER_RESOURCE_TEXTURE_1D:
            tex_type = "1D";
            break;
 
        case WINED3D_SHADER_RESOURCE_TEXTURE_2D:
            shader = ins->ctx->shader;
            device = shader->device;
            gl_info = &device->adapter->gl_info;
 
            if (pshader && priv->cur_ps_args->super.np2_fixup & (1u << sampler_idx)
                    && gl_info->supported[ARB_TEXTURE_RECTANGLE])
                tex_type = "RECT";
            else
                tex_type = "2D";
            if (shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type))
            {
                if (priv->cur_np2fixup_info->super.active & (1u << sampler_idx))
                {
                    if (flags) FIXME("Only ordinary sampling from NP2 textures is supported.\n");
                    else np2_fixup = TRUE;
                }
            }
            break;
 
        case WINED3D_SHADER_RESOURCE_TEXTURE_3D:
            tex_type = "3D";
            break;
 
        case WINED3D_SHADER_RESOURCE_TEXTURE_CUBE:
            tex_type = "CUBE";
            break;
 
        default:
            ERR("Unexpected resource type %#x.\n", resource_type);
            tex_type = "";
    }
 
    /* TEX, TXL, TXD and TXP do not support the "H" modifier,
     * so don't use shader_arb_get_modifier
     */
    if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) mod = "_SAT";
    else mod = "";
 
    /* Fragment samplers always have indentity mapping */
    if(sampler_idx >= MAX_FRAGMENT_SAMPLERS)
    {
        sampler_idx = priv->cur_vs_args->vertex.samplers[sampler_idx - MAX_FRAGMENT_SAMPLERS];
    }
 
    if (pshader)
    {
        masks = calc_color_correction(priv->cur_ps_args->super.color_fixup[sampler_idx],
                ins->dst[0].write_mask);
 
        if (masks.source || masks.sign)
            tex_dst = "TA";
    }
 
    if (flags & TEX_DERIV)
    {
        if(flags & TEX_PROJ) FIXME("Projected texture sampling with custom derivatives\n");
        if(flags & TEX_BIAS) FIXME("Biased texture sampling with custom derivatives\n");
        shader_addline(buffer, "TXD%s %s, %s, %s, %s, texture[%u], %s;\n", mod, tex_dst, coord_reg,
                       dsx, dsy, sampler_idx, tex_type);
    }
    else if(flags & TEX_LOD)
    {
        if(flags & TEX_PROJ) FIXME("Projected texture sampling with explicit lod\n");
        if(flags & TEX_BIAS) FIXME("Biased texture sampling with explicit lod\n");
        shader_addline(buffer, "TXL%s %s, %s, texture[%u], %s;\n", mod, tex_dst, coord_reg,
                       sampler_idx, tex_type);
    }
    else if (flags & TEX_BIAS)
    {
        /* Shouldn't be possible, but let's check for it */
        if(flags & TEX_PROJ) FIXME("Biased and Projected texture sampling\n");
        /* TXB takes the 4th component of the source vector automatically, as d3d. Nothing more to do */
        shader_addline(buffer, "TXB%s %s, %s, texture[%u], %s;\n", mod, tex_dst, coord_reg, sampler_idx, tex_type);
    }
    else if (flags & TEX_PROJ)
    {
        shader_addline(buffer, "TXP%s %s, %s, texture[%u], %s;\n", mod, tex_dst, coord_reg, sampler_idx, tex_type);
    }
    else
    {
        if (np2_fixup)
        {
            const unsigned char idx = priv->cur_np2fixup_info->super.idx[sampler_idx];
            shader_addline(buffer, "MUL TA, np2fixup[%u].%s, %s;\n", idx >> 1,
                           (idx % 2) ? "zwxy" : "xyzw", coord_reg);
 
            shader_addline(buffer, "TEX%s %s, TA, texture[%u], %s;\n", mod, tex_dst, sampler_idx, tex_type);
        }
        else
            shader_addline(buffer, "TEX%s %s, %s, texture[%u], %s;\n", mod, tex_dst, coord_reg, sampler_idx, tex_type);
    }
 
    if (pshader)
    {
        gen_color_correction(buffer, dst_str, tex_dst,
                arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_ONE),
                arb_get_helper_value(WINED3D_SHADER_TYPE_PIXEL, ARB_TWO),
                priv->cur_ps_args->super.color_fixup[sampler_idx], masks);
    }
}
 
static void shader_arb_get_src_param(const struct wined3d_shader_instruction *ins,
        const struct wined3d_shader_src_param *src, unsigned int tmpreg, char *outregstr)
{
    /* Generate a line that does the input modifier computation and return the input register to use */
    BOOL is_color = FALSE, insert_line;
    char regstr[256];
    char swzstr[20];
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
    const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
    const char *two = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_TWO);
 
    /* Assume a new line will be added */
    insert_line = TRUE;
 
    /* Get register name */
    shader_arb_get_register_name(ins, &src->reg, regstr, &is_color);
    shader_arb_get_swizzle(src, is_color, swzstr);
 
    switch (src->modifiers)
    {
    case WINED3DSPSM_NONE:
        sprintf(outregstr, "%s%s", regstr, swzstr);
        insert_line = FALSE;
        break;
    case WINED3DSPSM_NEG:
        sprintf(outregstr, "-%s%s", regstr, swzstr);
        insert_line = FALSE;
        break;
    case WINED3DSPSM_BIAS:
        shader_addline(buffer, "ADD T%c, %s, -coefdiv.x;\n", 'A' + tmpreg, regstr);
        break;
    case WINED3DSPSM_BIASNEG:
        shader_addline(buffer, "ADD T%c, -%s, coefdiv.x;\n", 'A' + tmpreg, regstr);
        break;
    case WINED3DSPSM_SIGN:
        shader_addline(buffer, "MAD T%c, %s, %s, -%s;\n", 'A' + tmpreg, regstr, two, one);
        break;
    case WINED3DSPSM_SIGNNEG:
        shader_addline(buffer, "MAD T%c, %s, -%s, %s;\n", 'A' + tmpreg, regstr, two, one);
        break;
    case WINED3DSPSM_COMP:
        shader_addline(buffer, "SUB T%c, %s, %s;\n", 'A' + tmpreg, one, regstr);
        break;
    case WINED3DSPSM_X2:
        shader_addline(buffer, "ADD T%c, %s, %s;\n", 'A' + tmpreg, regstr, regstr);
        break;
    case WINED3DSPSM_X2NEG:
        shader_addline(buffer, "ADD T%c, -%s, -%s;\n", 'A' + tmpreg, regstr, regstr);
        break;
    case WINED3DSPSM_DZ:
        shader_addline(buffer, "RCP T%c, %s.z;\n", 'A' + tmpreg, regstr);
        shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
        break;
    case WINED3DSPSM_DW:
        shader_addline(buffer, "RCP T%c, %s.w;\n", 'A' + tmpreg, regstr);
        shader_addline(buffer, "MUL T%c, %s, T%c;\n", 'A' + tmpreg, regstr, 'A' + tmpreg);
        break;
    case WINED3DSPSM_ABS:
        if(ctx->target_version >= NV2) {
            sprintf(outregstr, "|%s%s|", regstr, swzstr);
            insert_line = FALSE;
        } else {
            shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
        }
        break;
    case WINED3DSPSM_ABSNEG:
        if(ctx->target_version >= NV2) {
            sprintf(outregstr, "-|%s%s|", regstr, swzstr);
        } else {
            shader_addline(buffer, "ABS T%c, %s;\n", 'A' + tmpreg, regstr);
            sprintf(outregstr, "-T%c%s", 'A' + tmpreg, swzstr);
        }
        insert_line = FALSE;
        break;
    default:
        sprintf(outregstr, "%s%s", regstr, swzstr);
        insert_line = FALSE;
    }
 
    /* Return modified or original register, with swizzle */
    if (insert_line)
        sprintf(outregstr, "T%c%s", 'A' + tmpreg, swzstr);
}
 
static void pshader_hw_bem(const struct wined3d_shader_instruction *ins)
{
    const struct wined3d_shader_dst_param *dst = &ins->dst[0];
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    DWORD sampler_code = dst->reg.idx[0].offset;
    char dst_name[50];
    char src_name[2][50];
 
    shader_arb_get_dst_param(ins, dst, dst_name);
 
    /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
     *
     * Keep in mind that src_name[1] can be "TB" and src_name[0] can be "TA" because modifiers like _x2 are valid
     * with bem. So delay loading the first parameter until after the perturbation calculation which needs two
     * temps is done.
     */
    shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
    shader_addline(buffer, "SWZ TA, bumpenvmat%d, x, z, 0, 0;\n", sampler_code);
    shader_addline(buffer, "DP3 TC.r, TA, %s;\n", src_name[1]);
    shader_addline(buffer, "SWZ TA, bumpenvmat%d, y, w, 0, 0;\n", sampler_code);
    shader_addline(buffer, "DP3 TC.g, TA, %s;\n", src_name[1]);
 
    shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
    shader_addline(buffer, "ADD %s, %s, TC;\n", dst_name, src_name[0]);
}
 
static DWORD negate_modifiers(DWORD mod, char *extra_char)
{
    *extra_char = ' ';
    switch(mod)
    {
        case WINED3DSPSM_NONE:      return WINED3DSPSM_NEG;
        case WINED3DSPSM_NEG:       return WINED3DSPSM_NONE;
        case WINED3DSPSM_BIAS:      return WINED3DSPSM_BIASNEG;
        case WINED3DSPSM_BIASNEG:   return WINED3DSPSM_BIAS;
        case WINED3DSPSM_SIGN:      return WINED3DSPSM_SIGNNEG;
        case WINED3DSPSM_SIGNNEG:   return WINED3DSPSM_SIGN;
        case WINED3DSPSM_COMP:      *extra_char = '-'; return WINED3DSPSM_COMP;
        case WINED3DSPSM_X2:        return WINED3DSPSM_X2NEG;
        case WINED3DSPSM_X2NEG:     return WINED3DSPSM_X2;
        case WINED3DSPSM_DZ:        *extra_char = '-'; return WINED3DSPSM_DZ;
        case WINED3DSPSM_DW:        *extra_char = '-'; return WINED3DSPSM_DW;
        case WINED3DSPSM_ABS:       return WINED3DSPSM_ABSNEG;
        case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
    }
    FIXME("Unknown modifier %u\n", mod);
    return mod;
}
 
static void pshader_hw_cnd(const struct wined3d_shader_instruction *ins)
{
    const struct wined3d_shader_dst_param *dst = &ins->dst[0];
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    char dst_name[50];
    char src_name[3][50];
    DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
            ins->ctx->reg_maps->shader_version.minor);
 
    shader_arb_get_dst_param(ins, dst, dst_name);
    shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
 
    if (shader_version <= WINED3D_SHADER_VERSION(1, 3) && ins->coissue
            && ins->dst->write_mask != WINED3DSP_WRITEMASK_3)
    {
        shader_addline(buffer, "MOV%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[1]);
    }
    else
    {
        struct wined3d_shader_src_param src0_copy = ins->src[0];
        char extra_neg;
 
        /* src0 may have a negate srcmod set, so we can't blindly add "-" to the name */
        src0_copy.modifiers = negate_modifiers(src0_copy.modifiers, &extra_neg);
 
        shader_arb_get_src_param(ins, &src0_copy, 0, src_name[0]);
        shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
        shader_addline(buffer, "ADD TA, %c%s, coefdiv.x;\n", extra_neg, src_name[0]);
        shader_addline(buffer, "CMP%s %s, TA, %s, %s;\n", shader_arb_get_modifier(ins),
                dst_name, src_name[1], src_name[2]);
    }
}
 
static void pshader_hw_cmp(const struct wined3d_shader_instruction *ins)
{
    const struct wined3d_shader_dst_param *dst = &ins->dst[0];
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    char dst_name[50];
    char src_name[3][50];
 
    shader_arb_get_dst_param(ins, dst, dst_name);
 
    /* Generate input register names (with modifiers) */
    shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
    shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
    shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
 
    shader_addline(buffer, "CMP%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
            dst_name, src_name[0], src_name[2], src_name[1]);
}
 
static void pshader_hw_dp2add(const struct wined3d_shader_instruction *ins)
{
    const struct wined3d_shader_dst_param *dst = &ins->dst[0];
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    char dst_name[50];
    char src_name[3][50];
    struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
 
    shader_arb_get_dst_param(ins, dst, dst_name);
    shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
    shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
 
    if(ctx->target_version >= NV3)
    {
        /* GL_NV_fragment_program2 has a 1:1 matching instruction */
        shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
        shader_addline(buffer, "DP2A%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
                       dst_name, src_name[0], src_name[1], src_name[2]);
    }
    else if(ctx->target_version >= NV2)
    {
        /* dst.x = src2.?, src0.x, src1.x + src0.y * src1.y
         * dst.y = src2.?, src0.x, src1.z + src0.y * src1.w
         * dst.z = src2.?, src0.x, src1.x + src0.y * src1.y
         * dst.z = src2.?, src0.x, src1.z + src0.y * src1.w
         *
         * Make sure that src1.zw = src1.xy, then we get a classic dp2add
         *
         * .xyxy and other swizzles that we could get with this are not valid in
         * plain ARBfp, but luckily the NV extension grammar lifts this limitation.
         */
        struct wined3d_shader_src_param tmp_param = ins->src[1];
        DWORD swizzle = tmp_param.swizzle & 0xf; /* Selects .xy */
        tmp_param.swizzle = swizzle | (swizzle << 4); /* Creates .xyxy */
 
        shader_arb_get_src_param(ins, &tmp_param, 1, src_name[1]);
 
        shader_addline(buffer, "X2D%s %s, %s, %s, %s;\n", shader_arb_get_modifier(ins),
                       dst_name, src_name[2], src_name[0], src_name[1]);
    }
    else
    {
        shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
        /* Emulate a DP2 with a DP3 and 0.0. Don't use the dest as temp register, it could be src[1] or src[2]
        * src_name[0] can be TA, but TA is a private temp for modifiers, so it is save to overwrite
        */
        shader_addline(buffer, "MOV TA, %s;\n", src_name[0]);
        shader_addline(buffer, "MOV TA.z, 0.0;\n");
        shader_addline(buffer, "DP3 TA, TA, %s;\n", src_name[1]);
        shader_addline(buffer, "ADD%s %s, TA, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name[2]);
    }
}
 
/* Map the opcode 1-to-1 to the GL code */
static void shader_hw_map2gl(const struct wined3d_shader_instruction *ins)
{
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    const char *instruction;
    char arguments[256], dst_str[50];
    unsigned int i;
    const struct wined3d_shader_dst_param *dst = &ins->dst[0];
 
    switch (ins->handler_idx)
    {
        case WINED3DSIH_ABS: instruction = "ABS"; break;
        case WINED3DSIH_ADD: instruction = "ADD"; break;
        case WINED3DSIH_CRS: instruction = "XPD"; break;
        case WINED3DSIH_DP3: instruction = "DP3"; break;
        case WINED3DSIH_DP4: instruction = "DP4"; break;
        case WINED3DSIH_DST: instruction = "DST"; break;
        case WINED3DSIH_FRC: instruction = "FRC"; break;
        case WINED3DSIH_LIT: instruction = "LIT"; break;
        case WINED3DSIH_LRP: instruction = "LRP"; break;
        case WINED3DSIH_MAD: instruction = "MAD"; break;
        case WINED3DSIH_MAX: instruction = "MAX"; break;
        case WINED3DSIH_MIN: instruction = "MIN"; break;
        case WINED3DSIH_MOV: instruction = "MOV"; break;
        case WINED3DSIH_MUL: instruction = "MUL"; break;
        case WINED3DSIH_SGE: instruction = "SGE"; break;
        case WINED3DSIH_SLT: instruction = "SLT"; break;
        case WINED3DSIH_SUB: instruction = "SUB"; break;
        case WINED3DSIH_MOVA:instruction = "ARR"; break;
        case WINED3DSIH_DSX: instruction = "DDX"; break;
        default: instruction = "";
            FIXME("Unhandled opcode %s.\n", debug_d3dshaderinstructionhandler(ins->handler_idx));
            break;
    }
 
    /* Note that shader_arb_add_dst_param() adds spaces. */
    arguments[0] = '\0';
    shader_arb_get_dst_param(ins, dst, dst_str);
    for (i = 0; i < ins->src_count; ++i)
    {
        char operand[100];
        strcat(arguments, ", ");
        shader_arb_get_src_param(ins, &ins->src[i], i, operand);
        strcat(arguments, operand);
    }
    shader_addline(buffer, "%s%s %s%s;\n", instruction, shader_arb_get_modifier(ins), dst_str, arguments);
}
 
static void shader_hw_nop(const struct wined3d_shader_instruction *ins) {}
 
static DWORD shader_arb_select_component(DWORD swizzle, DWORD component)
{
    return ((swizzle >> 2 * component) & 0x3) * 0x55;
}
 
static void shader_hw_mov(const struct wined3d_shader_instruction *ins)
{
    const struct wined3d_shader *shader = ins->ctx->shader;
    const struct wined3d_shader_reg_maps *reg_maps = ins->ctx->reg_maps;
    BOOL pshader = shader_is_pshader_version(reg_maps->shader_version.type);
    struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
    const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
    const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
    const char *two = arb_get_helper_value(reg_maps->shader_version.type, ARB_TWO);
 
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    char src0_param[256];
 
    if (ins->handler_idx == WINED3DSIH_MOVA)
    {
        const struct arb_vshader_private *shader_data = shader->backend_data;
        char write_mask[6];
        const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
 
        if(ctx->target_version >= NV2) {
            shader_hw_map2gl(ins);
            return;
        }
        shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
        shader_arb_get_write_mask(ins, &ins->dst[0], write_mask);
 
        /* This implements the mova formula used in GLSL. The first two instructions
         * prepare the sign() part. Note that it is fine to have my_sign(0.0) = 1.0
         * in this case:
         * mova A0.x, 0.0
         *
         * A0.x = arl(floor(abs(0.0) + 0.5) * 1.0) = floor(0.5) = 0.0 since arl does a floor
         *
         * The ARL is performed when A0 is used - the requested component is read from A0_SHADOW into
         * A0.x. We can use the overwritten component of A0_shadow as temporary storage for the sign.
         */
        shader_addline(buffer, "SGE A0_SHADOW%s, %s, %s;\n", write_mask, src0_param, zero);
        shader_addline(buffer, "MAD A0_SHADOW%s, A0_SHADOW, %s, -%s;\n", write_mask, two, one);
 
        shader_addline(buffer, "ABS TA%s, %s;\n", write_mask, src0_param);
        shader_addline(buffer, "ADD TA%s, TA, rel_addr_const.x;\n", write_mask);
        shader_addline(buffer, "FLR TA%s, TA;\n", write_mask);
        if (shader_data->rel_offset)
        {
            shader_addline(buffer, "ADD TA%s, TA, %s;\n", write_mask, offset);
        }
        shader_addline(buffer, "MUL A0_SHADOW%s, TA, A0_SHADOW;\n", write_mask);
 
        ((struct shader_arb_ctx_priv *)ins->ctx->backend_data)->addr_reg[0] = '\0';
    }
    else if (reg_maps->shader_version.major == 1
          && !shader_is_pshader_version(reg_maps->shader_version.type)
          && ins->dst[0].reg.type == WINED3DSPR_ADDR)
    {
        const struct arb_vshader_private *shader_data = shader->backend_data;
        src0_param[0] = '\0';
 
        if (shader_data->rel_offset && ctx->target_version == ARB)
        {
            const char *offset = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_VS_REL_OFFSET);
            shader_arb_get_src_param(ins, &ins->src[0], 0, src0_param);
            shader_addline(buffer, "ADD TA.x, %s, %s;\n", src0_param, offset);
            shader_addline(buffer, "ARL A0.x, TA.x;\n");
        }
        else
        {
            /* Apple's ARB_vertex_program implementation does not accept an ARL source argument
             * with more than one component. Thus replicate the first source argument over all
             * 4 components. For example, .xyzw -> .x (or better: .xxxx), .zwxy -> .z, etc) */
            struct wined3d_shader_src_param tmp_src = ins->src[0];
            tmp_src.swizzle = shader_arb_select_component(tmp_src.swizzle, 0);
            shader_arb_get_src_param(ins, &tmp_src, 0, src0_param);
            shader_addline(buffer, "ARL A0.x, %s;\n", src0_param);
        }
    }
    else if (ins->dst[0].reg.type == WINED3DSPR_COLOROUT && !ins->dst[0].reg.idx[0].offset && pshader)
    {
        if (ctx->ps_post_process && shader->u.ps.color0_mov)
        {
            shader_addline(buffer, "#mov handled in srgb write or fog code\n");
            return;
        }
        shader_hw_map2gl(ins);
    }
    else
    {
        shader_hw_map2gl(ins);
    }
}
 
static void pshader_hw_texkill(const struct wined3d_shader_instruction *ins)
{
    const struct wined3d_shader_dst_param *dst = &ins->dst[0];
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    char reg_dest[40];
 
    /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
     * but >= 2.0 honors it (undocumented, but tested by the d3d9 testsuite)
     */
    shader_arb_get_dst_param(ins, dst, reg_dest);
 
    if (ins->ctx->reg_maps->shader_version.major >= 2)
    {
        const char *kilsrc = "TA";
        BOOL is_color;
 
        shader_arb_get_register_name(ins, &dst->reg, reg_dest, &is_color);
        if(dst->write_mask == WINED3DSP_WRITEMASK_ALL)
        {
            kilsrc = reg_dest;
        }
        else
        {
            /* Sigh. KIL doesn't support swizzles/writemasks. KIL passes a writemask, but ".xy" for example
             * is not valid as a swizzle in ARB (needs ".xyyy"). Use SWZ to load the register properly, and set
             * masked out components to 0(won't kill)
             */
            char x = '0', y = '0', z = '0', w = '0';
            if(dst->write_mask & WINED3DSP_WRITEMASK_0) x = 'x';
            if(dst->write_mask & WINED3DSP_WRITEMASK_1) y = 'y';
            if(dst->write_mask & WINED3DSP_WRITEMASK_2) z = 'z';
            if(dst->write_mask & WINED3DSP_WRITEMASK_3) w = 'w';
            shader_addline(buffer, "SWZ TA, %s, %c, %c, %c, %c;\n", reg_dest, x, y, z, w);
        }
        shader_addline(buffer, "KIL %s;\n", kilsrc);
    }
    else
    {
        /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
         * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
         *
         * ps_1_3 shaders use the texcoord incarnation of the Tx register. ps_1_4 shaders can use the same,
         * or pass in any temporary register(in shader phase 2)
         */
        if (ins->ctx->reg_maps->shader_version.minor <= 3)
            sprintf(reg_dest, "fragment.texcoord[%u]", dst->reg.idx[0].offset);
        else
            shader_arb_get_dst_param(ins, dst, reg_dest);
        shader_addline(buffer, "SWZ TA, %s, x, y, z, 1;\n", reg_dest);
        shader_addline(buffer, "KIL TA;\n");
    }
}
 
static void pshader_hw_tex(const struct wined3d_shader_instruction *ins)
{
    struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
    const struct wined3d_shader_dst_param *dst = &ins->dst[0];
    DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
            ins->ctx->reg_maps->shader_version.minor);
    struct wined3d_shader_src_param src;
 
    char reg_dest[40];
    char reg_coord[40];
    DWORD reg_sampler_code;
    WORD myflags = 0;
    BOOL swizzle_coord = FALSE;
 
    /* All versions have a destination register */
    shader_arb_get_dst_param(ins, dst, reg_dest);
 
    /* 1.0-1.4: Use destination register number as texture code.
       2.0+: Use provided sampler number as texture code. */
    if (shader_version < WINED3D_SHADER_VERSION(2,0))
        reg_sampler_code = dst->reg.idx[0].offset;
    else
        reg_sampler_code = ins->src[1].reg.idx[0].offset;
 
    /* 1.0-1.3: Use the texcoord varying.
       1.4+: Use provided coordinate source register. */
    if (shader_version < WINED3D_SHADER_VERSION(1,4))
        sprintf(reg_coord, "fragment.texcoord[%u]", reg_sampler_code);
    else {
        /* TEX is the only instruction that can handle DW and DZ natively */
        src = ins->src[0];
        if(src.modifiers == WINED3DSPSM_DW) src.modifiers = WINED3DSPSM_NONE;
        if(src.modifiers == WINED3DSPSM_DZ) src.modifiers = WINED3DSPSM_NONE;
        shader_arb_get_src_param(ins, &src, 0, reg_coord);
    }
 
    /* projection flag:
     * 1.1, 1.2, 1.3: Use WINED3D_TSS_TEXTURETRANSFORMFLAGS
     * 1.4: Use WINED3DSPSM_DZ or WINED3DSPSM_DW on src[0]
     * 2.0+: Use WINED3DSI_TEXLD_PROJECT on the opcode
     */
    if (shader_version < WINED3D_SHADER_VERSION(1,4))
    {
        DWORD flags = 0;
        if (reg_sampler_code < MAX_TEXTURES)
            flags = priv->cur_ps_args->super.tex_transform >> reg_sampler_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
        if (flags & WINED3D_PSARGS_PROJECTED)
        {
            myflags |= TEX_PROJ;
            if ((flags & ~WINED3D_PSARGS_PROJECTED) == WINED3D_TTFF_COUNT3)
                swizzle_coord = TRUE;
        }
    }
    else if (shader_version < WINED3D_SHADER_VERSION(2,0))
    {
        enum wined3d_shader_src_modifier src_mod = ins->src[0].modifiers;
        if (src_mod == WINED3DSPSM_DZ)
        {
            swizzle_coord = TRUE;
            myflags |= TEX_PROJ;
        } else if(src_mod == WINED3DSPSM_DW) {
            myflags |= TEX_PROJ;
        }
    } else {
        if (ins->flags & WINED3DSI_TEXLD_PROJECT) myflags |= TEX_PROJ;
        if (ins->flags & WINED3DSI_TEXLD_BIAS) myflags |= TEX_BIAS;
    }
 
    if (swizzle_coord)
    {
        /* TXP cannot handle DZ natively, so move the z coordinate to .w.
         * reg_coord is a read-only varying register, so we need a temp reg */
        shader_addline(ins->ctx->buffer, "SWZ TA, %s, x, y, z, z;\n", reg_coord);
        strcpy(reg_coord, "TA");
    }
 
    shader_hw_sample(ins, reg_sampler_code, reg_dest, reg_coord, myflags, NULL, NULL);
}
 
static void pshader_hw_texcoord(const struct wined3d_shader_instruction *ins)
{
    const struct wined3d_shader_dst_param *dst = &ins->dst[0];
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    DWORD shader_version = WINED3D_SHADER_VERSION(ins->ctx->reg_maps->shader_version.major,
            ins->ctx->reg_maps->shader_version.minor);
    char dst_str[50];
 
    if (shader_version < WINED3D_SHADER_VERSION(1,4))
    {
        DWORD reg = dst->reg.idx[0].offset;
 
        shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
        shader_addline(buffer, "MOV_SAT %s, fragment.texcoord[%u];\n", dst_str, reg);
    } else {
        char reg_src[40];
 
        shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src);
        shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
        shader_addline(buffer, "MOV %s, %s;\n", dst_str, reg_src);
    }
}
 
static void pshader_hw_texreg2ar(const struct wined3d_shader_instruction *ins)
{
     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
     DWORD flags = 0;
 
     DWORD reg1 = ins->dst[0].reg.idx[0].offset;
     char dst_str[50];
     char src_str[50];
 
     /* Note that texreg2ar treats Tx as a temporary register, not as a varying */
     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
     /* Move .x first in case src_str is "TA" */
     shader_addline(buffer, "MOV TA.y, %s.x;\n", src_str);
     shader_addline(buffer, "MOV TA.x, %s.w;\n", src_str);
     if (reg1 < MAX_TEXTURES)
     {
         struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
         flags = priv->cur_ps_args->super.tex_transform >> reg1 * WINED3D_PSARGS_TEXTRANSFORM_SHIFT;
     }
     shader_hw_sample(ins, reg1, dst_str, "TA", flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
}
 
static void pshader_hw_texreg2gb(const struct wined3d_shader_instruction *ins)
{
     struct wined3d_string_buffer *buffer = ins->ctx->buffer;
 
     DWORD reg1 = ins->dst[0].reg.idx[0].offset;
     char dst_str[50];
     char src_str[50];
 
     /* Note that texreg2gb treats Tx as a temporary register, not as a varying */
     shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
     shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
     shader_addline(buffer, "MOV TA.x, %s.y;\n", src_str);
     shader_addline(buffer, "MOV TA.y, %s.z;\n", src_str);
     shader_hw_sample(ins, reg1, dst_str, "TA", 0, NULL, NULL);
}
 
static void pshader_hw_texreg2rgb(const struct wined3d_shader_instruction *ins)
{
    DWORD reg1 = ins->dst[0].reg.idx[0].offset;
    char dst_str[50];
    char src_str[50];
 
    /* Note that texreg2rg treats Tx as a temporary register, not as a varying */
    shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
    shader_arb_get_src_param(ins, &ins->src[0], 0, src_str);
    shader_hw_sample(ins, reg1, dst_str, src_str, 0, NULL, NULL);
}
 
static void pshader_hw_texbem(const struct wined3d_shader_instruction *ins)
{
    struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
    const struct wined3d_shader_dst_param *dst = &ins->dst[0];
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    char reg_coord[40], dst_reg[50], src_reg[50];
    DWORD reg_dest_code;
 
    /* All versions have a destination register. The Tx where the texture coordinates come
     * from is the varying incarnation of the texture register
     */
    reg_dest_code = dst->reg.idx[0].offset;
    shader_arb_get_dst_param(ins, &ins->dst[0], dst_reg);
    shader_arb_get_src_param(ins, &ins->src[0], 0, src_reg);
    sprintf(reg_coord, "fragment.texcoord[%u]", reg_dest_code);
 
    /* Sampling the perturbation map in Tsrc was done already, including the signedness correction if needed
     * The Tx in which the perturbation map is stored is the tempreg incarnation of the texture register
     *
     * GL_NV_fragment_program_option could handle this in one instruction via X2D:
     * X2D TA.xy, fragment.texcoord, T%u, bumpenvmat%u.xzyw
     *
     * However, the NV extensions are never enabled for <= 2.0 shaders because of the performance penalty that
     * comes with it, and texbem is an 1.x only instruction. No 1.x instruction forces us to enable the NV
     * extension.
     */
    shader_addline(buffer, "SWZ TB, bumpenvmat%d, x, z, 0, 0;\n", reg_dest_code);
    shader_addline(buffer, "DP3 TA.x, TB, %s;\n", src_reg);
    shader_addline(buffer, "SWZ TB, bumpenvmat%d, y, w, 0, 0;\n", reg_dest_code);
    shader_addline(buffer, "DP3 TA.y, TB, %s;\n", src_reg);
 
    /* with projective textures, texbem only divides the static texture coord, not the displacement,
     * so we can't let the GL handle this.
     */
    if ((priv->cur_ps_args->super.tex_transform >> reg_dest_code * WINED3D_PSARGS_TEXTRANSFORM_SHIFT)
            & WINED3D_PSARGS_PROJECTED)
    {
        shader_addline(buffer, "RCP TB.w, %s.w;\n", reg_coord);
        shader_addline(buffer, "MUL TB.xy, %s, TB.w;\n", reg_coord);
        shader_addline(buffer, "ADD TA.xy, TA, TB;\n");
    } else {
        shader_addline(buffer, "ADD TA.xy, TA, %s;\n", reg_coord);
    }
 
    shader_hw_sample(ins, reg_dest_code, dst_reg, "TA", 0, NULL, NULL);
 
    if (ins->handler_idx == WINED3DSIH_TEXBEML)
    {
        /* No src swizzles are allowed, so this is ok */
        shader_addline(buffer, "MAD TA, %s.z, luminance%d.x, luminance%d.y;\n",
                       src_reg, reg_dest_code, reg_dest_code);
        shader_addline(buffer, "MUL %s, %s, TA;\n", dst_reg, dst_reg);
    }
}
 
static void pshader_hw_texm3x2pad(const struct wined3d_shader_instruction *ins)
{
    DWORD reg = ins->dst[0].reg.idx[0].offset;
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    char src0_name[50], dst_name[50];
    BOOL is_color;
    struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
 
    shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
    /* The next instruction will be a texm3x2tex or texm3x2depth that writes to the uninitialized
     * T<reg+1> register. Use this register to store the calculated vector
     */
    tmp_reg.idx[0].offset = reg + 1;
    shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
    shader_addline(buffer, "DP3 %s.x, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
}
 
static void pshader_hw_texm3x2tex(const struct wined3d_shader_instruction *ins)
{
    struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
    DWORD flags;
    DWORD reg = ins->dst[0].reg.idx[0].offset;
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    char dst_str[50];
    char src0_name[50];
    char dst_reg[50];
    BOOL is_color;
 
    /* We know that we're writing to the uninitialized T<reg> register, so use it for temporary storage */
    shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
 
    shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
    shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
    shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
    flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
    shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
}
 
static void pshader_hw_texm3x3pad(const struct wined3d_shader_instruction *ins)
{
    struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
    DWORD reg = ins->dst[0].reg.idx[0].offset;
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    char src0_name[50], dst_name[50];
    struct wined3d_shader_register tmp_reg = ins->dst[0].reg;
    BOOL is_color;
 
    /* There are always 2 texm3x3pad instructions followed by one texm3x3[tex,vspec, ...] instruction, with
     * incrementing ins->dst[0].register_idx numbers. So the pad instruction already knows the final destination
     * register, and this register is uninitialized(otherwise the assembler complains that it is 'redeclared')
     */
    tmp_reg.idx[0].offset = reg + 2 - tex_mx->current_row;
    shader_arb_get_register_name(ins, &tmp_reg, dst_name, &is_color);
 
    shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
    shader_addline(buffer, "DP3 %s.%c, fragment.texcoord[%u], %s;\n",
                   dst_name, 'x' + tex_mx->current_row, reg, src0_name);
    tex_mx->texcoord_w[tex_mx->current_row++] = reg;
}
 
static void pshader_hw_texm3x3tex(const struct wined3d_shader_instruction *ins)
{
    struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
    struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
    DWORD flags;
    DWORD reg = ins->dst[0].reg.idx[0].offset;
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    char dst_str[50];
    char src0_name[50], dst_name[50];
    BOOL is_color;
 
    shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
    shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
    shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, reg, src0_name);
 
    /* Sample the texture using the calculated coordinates */
    shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
    flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
    shader_hw_sample(ins, reg, dst_str, dst_name, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
    tex_mx->current_row = 0;
}
 
static void pshader_hw_texm3x3vspec(const struct wined3d_shader_instruction *ins)
{
    struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
    struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
    DWORD flags;
    DWORD reg = ins->dst[0].reg.idx[0].offset;
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    char dst_str[50];
    char src0_name[50];
    char dst_reg[50];
    BOOL is_color;
 
    /* Get the dst reg without writemask strings. We know this register is uninitialized, so we can use all
     * components for temporary data storage
     */
    shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
    shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
    shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
 
    /* Construct the eye-ray vector from w coordinates */
    shader_addline(buffer, "MOV TB.x, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[0]);
    shader_addline(buffer, "MOV TB.y, fragment.texcoord[%u].w;\n", tex_mx->texcoord_w[1]);
    shader_addline(buffer, "MOV TB.z, fragment.texcoord[%u].w;\n", reg);
 
    /* Calculate reflection vector
     */
    shader_addline(buffer, "DP3 %s.w, %s, TB;\n", dst_reg, dst_reg);
    /* The .w is ignored when sampling, so I can use TB.w to calculate dot(N, N) */
    shader_addline(buffer, "DP3 TB.w, %s, %s;\n", dst_reg, dst_reg);
    shader_addline(buffer, "RCP TB.w, TB.w;\n");
    shader_addline(buffer, "MUL %s.w, %s.w, TB.w;\n", dst_reg, dst_reg);
    shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
    shader_addline(buffer, "MAD %s, coefmul.x, %s, -TB;\n", dst_reg, dst_reg);
 
    /* Sample the texture using the calculated coordinates */
    shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
    flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
    shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
    tex_mx->current_row = 0;
}
 
static void pshader_hw_texm3x3spec(const struct wined3d_shader_instruction *ins)
{
    struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
    struct wined3d_shader_tex_mx *tex_mx = ins->ctx->tex_mx;
    DWORD flags;
    DWORD reg = ins->dst[0].reg.idx[0].offset;
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    char dst_str[50];
    char src0_name[50];
    char src1_name[50];
    char dst_reg[50];
    BOOL is_color;
 
    shader_arb_get_src_param(ins, &ins->src[0], 0, src0_name);
    shader_arb_get_src_param(ins, &ins->src[0], 1, src1_name);
    shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_reg, &is_color);
    /* Note: dst_reg.xy is input here, generated by two texm3x3pad instructions */
    shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_reg, reg, src0_name);
 
    /* Calculate reflection vector.
     *
     *                   dot(N, E)
     * dst_reg.xyz = 2 * --------- * N - E
     *                   dot(N, N)
     *
     * Which normalizes the normal vector
     */
    shader_addline(buffer, "DP3 %s.w, %s, %s;\n", dst_reg, dst_reg, src1_name);
    shader_addline(buffer, "DP3 TC.w, %s, %s;\n", dst_reg, dst_reg);
    shader_addline(buffer, "RCP TC.w, TC.w;\n");
    shader_addline(buffer, "MUL %s.w, %s.w, TC.w;\n", dst_reg, dst_reg);
    shader_addline(buffer, "MUL %s, %s.w, %s;\n", dst_reg, dst_reg, dst_reg);
    shader_addline(buffer, "MAD %s, coefmul.x, %s, -%s;\n", dst_reg, dst_reg, src1_name);
 
    /* Sample the texture using the calculated coordinates */
    shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
    flags = reg < MAX_TEXTURES ? priv->cur_ps_args->super.tex_transform >> reg * WINED3D_PSARGS_TEXTRANSFORM_SHIFT : 0;
    shader_hw_sample(ins, reg, dst_str, dst_reg, flags & WINED3D_PSARGS_PROJECTED ? TEX_PROJ : 0, NULL, NULL);
    tex_mx->current_row = 0;
}
 
static void pshader_hw_texdepth(const struct wined3d_shader_instruction *ins)
{
    const struct wined3d_shader_dst_param *dst = &ins->dst[0];
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    char dst_name[50];
    const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
    const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
 
    /* texdepth has an implicit destination, the fragment depth value. It's only parameter,
     * which is essentially an input, is the destination register because it is the first
     * parameter. According to the msdn, this must be register r5, but let's keep it more flexible
     * here(writemasks/swizzles are not valid on texdepth)
     */
    shader_arb_get_dst_param(ins, dst, dst_name);
 
    /* According to the msdn, the source register(must be r5) is unusable after
     * the texdepth instruction, so we're free to modify it
     */
    shader_addline(buffer, "MIN %s.y, %s.y, %s;\n", dst_name, dst_name, one);
 
    /* How to deal with the special case dst_name.g == 0? if r != 0, then
     * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
     * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
     */
    shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
    shader_addline(buffer, "MUL TA.x, %s.x, %s.y;\n", dst_name, dst_name);
    shader_addline(buffer, "MIN TA.x, TA.x, %s;\n", one);
    shader_addline(buffer, "MAX result.depth, TA.x, %s;\n", zero);
}
 
static void pshader_hw_texdp3tex(const struct wined3d_shader_instruction *ins)
{
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    DWORD sampler_idx = ins->dst[0].reg.idx[0].offset;
    char src0[50];
    char dst_str[50];
 
    shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
    shader_addline(buffer, "MOV TB, 0.0;\n");
    shader_addline(buffer, "DP3 TB.x, fragment.texcoord[%u], %s;\n", sampler_idx, src0);
 
    shader_arb_get_dst_param(ins, &ins->dst[0], dst_str);
    shader_hw_sample(ins, sampler_idx, dst_str, "TB", 0 /* Only one coord, can't be projected */, NULL, NULL);
}
 
static void pshader_hw_texdp3(const struct wined3d_shader_instruction *ins)
{
    const struct wined3d_shader_dst_param *dst = &ins->dst[0];
    char src0[50];
    char dst_str[50];
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
 
    /* Handle output register */
    shader_arb_get_dst_param(ins, dst, dst_str);
    shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
    shader_addline(buffer, "DP3 %s, fragment.texcoord[%u], %s;\n", dst_str, dst->reg.idx[0].offset, src0);
}
 
static void pshader_hw_texm3x3(const struct wined3d_shader_instruction *ins)
{
    const struct wined3d_shader_dst_param *dst = &ins->dst[0];
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    char dst_str[50], dst_name[50];
    char src0[50];
    BOOL is_color;
 
    shader_arb_get_dst_param(ins, dst, dst_str);
    shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
    shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
    shader_addline(buffer, "DP3 %s.z, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx[0].offset, src0);
    shader_addline(buffer, "MOV %s, %s;\n", dst_str, dst_name);
}
 
static void pshader_hw_texm3x2depth(const struct wined3d_shader_instruction *ins)
{
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    const struct wined3d_shader_dst_param *dst = &ins->dst[0];
    char src0[50], dst_name[50];
    BOOL is_color;
    const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
    const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
 
    shader_arb_get_src_param(ins, &ins->src[0], 0, src0);
    shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
    shader_addline(buffer, "DP3 %s.y, fragment.texcoord[%u], %s;\n", dst_name, dst->reg.idx[0].offset, src0);
 
    /* How to deal with the special case dst_name.g == 0? if r != 0, then
     * the r * (1 / 0) will give infinity, which is clamped to 1.0, the correct
     * result. But if r = 0.0, then 0 * inf = 0, which is incorrect.
     */
    shader_addline(buffer, "RCP %s.y, %s.y;\n", dst_name, dst_name);
    shader_addline(buffer, "MUL %s.x, %s.x, %s.y;\n", dst_name, dst_name, dst_name);
    shader_addline(buffer, "MIN %s.x, %s.x, %s;\n", dst_name, dst_name, one);
    shader_addline(buffer, "MAX result.depth, %s.x, %s;\n", dst_name, zero);
}
 
static void shader_hw_mnxn(const struct wined3d_shader_instruction *ins)
{
    int i;
    int nComponents = 0;
    struct wined3d_shader_dst_param tmp_dst = {{0}};
    struct wined3d_shader_src_param tmp_src[2] = {{{0}}};
    struct wined3d_shader_instruction tmp_ins;
 
    memset(&tmp_ins, 0, sizeof(tmp_ins));
 
    /* Set constants for the temporary argument */
    tmp_ins.ctx = ins->ctx;
    tmp_ins.dst_count = 1;
    tmp_ins.dst = &tmp_dst;
    tmp_ins.src_count = 2;
    tmp_ins.src = tmp_src;
 
    switch(ins->handler_idx)
    {
        case WINED3DSIH_M4x4:
            nComponents = 4;
            tmp_ins.handler_idx = WINED3DSIH_DP4;
            break;
        case WINED3DSIH_M4x3:
            nComponents = 3;
            tmp_ins.handler_idx = WINED3DSIH_DP4;
            break;
        case WINED3DSIH_M3x4:
            nComponents = 4;
            tmp_ins.handler_idx = WINED3DSIH_DP3;
            break;
        case WINED3DSIH_M3x3:
            nComponents = 3;
            tmp_ins.handler_idx = WINED3DSIH_DP3;
            break;
        case WINED3DSIH_M3x2:
            nComponents = 2;
            tmp_ins.handler_idx = WINED3DSIH_DP3;
            break;
        default:
            FIXME("Unhandled opcode %s.\n", debug_d3dshaderinstructionhandler(ins->handler_idx));
            break;
    }
 
    tmp_dst = ins->dst[0];
    tmp_src[0] = ins->src[0];
    tmp_src[1] = ins->src[1];
    for (i = 0; i < nComponents; ++i)
    {
        tmp_dst.write_mask = WINED3DSP_WRITEMASK_0 << i;
        shader_hw_map2gl(&tmp_ins);
        ++tmp_src[1].reg.idx[0].offset;
    }
}
 
static DWORD abs_modifier(DWORD mod, BOOL *need_abs)
{
    *need_abs = FALSE;
 
    switch(mod)
    {
        case WINED3DSPSM_NONE:      return WINED3DSPSM_ABS;
        case WINED3DSPSM_NEG:       return WINED3DSPSM_ABS;
        case WINED3DSPSM_BIAS:      *need_abs = TRUE; return WINED3DSPSM_BIAS;
        case WINED3DSPSM_BIASNEG:   *need_abs = TRUE; return WINED3DSPSM_BIASNEG;
        case WINED3DSPSM_SIGN:      *need_abs = TRUE; return WINED3DSPSM_SIGN;
        case WINED3DSPSM_SIGNNEG:   *need_abs = TRUE; return WINED3DSPSM_SIGNNEG;
        case WINED3DSPSM_COMP:      *need_abs = TRUE; return WINED3DSPSM_COMP;
        case WINED3DSPSM_X2:        *need_abs = TRUE; return WINED3DSPSM_X2;
        case WINED3DSPSM_X2NEG:     *need_abs = TRUE; return WINED3DSPSM_X2NEG;
        case WINED3DSPSM_DZ:        *need_abs = TRUE; return WINED3DSPSM_DZ;
        case WINED3DSPSM_DW:        *need_abs = TRUE; return WINED3DSPSM_DW;
        case WINED3DSPSM_ABS:       return WINED3DSPSM_ABS;
        case WINED3DSPSM_ABSNEG:    return WINED3DSPSM_ABS;
    }
    FIXME("Unknown modifier %u\n", mod);
    return mod;
}
 
static void shader_hw_scalar_op(const struct wined3d_shader_instruction *ins)
{
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    const char *instruction;
    struct wined3d_shader_src_param src0_copy = ins->src[0];
    BOOL need_abs = FALSE;
 
    char dst[50];
    char src[50];
 
    switch(ins->handler_idx)
    {
        case WINED3DSIH_RSQ:  instruction = "RSQ"; break;
        case WINED3DSIH_RCP:  instruction = "RCP"; break;
        case WINED3DSIH_EXPP:
            if (ins->ctx->reg_maps->shader_version.major < 2)
            {
                instruction = "EXP";
                break;
            }
            /* Drop through. */
        case WINED3DSIH_EXP:
            instruction = "EX2";
            break;
        case WINED3DSIH_LOG:
        case WINED3DSIH_LOGP:
            /* The precision requirements suggest that LOGP matches ARBvp's LOG
             * instruction, but notice that the output of those instructions is
             * different. */
            src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
            instruction = "LG2";
            break;
        default: instruction = "";
            FIXME("Unhandled opcode %s.\n", debug_d3dshaderinstructionhandler(ins->handler_idx));
            break;
    }
 
    /* Dx sdk says .x is used if no swizzle is given, but our test shows that
     * .w is used. */
    src0_copy.swizzle = shader_arb_select_component(src0_copy.swizzle, 3);
 
    shader_arb_get_dst_param(ins, &ins->dst[0], dst); /* Destination */
    shader_arb_get_src_param(ins, &src0_copy, 0, src);
 
    if(need_abs)
    {
        shader_addline(buffer, "ABS TA.w, %s;\n", src);
        shader_addline(buffer, "%s%s %s, TA.w;\n", instruction, shader_arb_get_modifier(ins), dst);
    }
    else
    {
        shader_addline(buffer, "%s%s %s, %s;\n", instruction, shader_arb_get_modifier(ins), dst, src);
    }
 
}
 
static void shader_hw_nrm(const struct wined3d_shader_instruction *ins)
{
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    char dst_name[50];
    char src_name[50];
    struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
    BOOL pshader = shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type);
    const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
 
    shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
    shader_arb_get_src_param(ins, &ins->src[0], 1 /* Use TB */, src_name);
 
    /* In D3D, NRM of a vector with length zero returns zero. Catch this situation, as
     * otherwise NRM or RSQ would return NaN */
    if(pshader && priv->target_version >= NV3)
    {
        /* GL_NV_fragment_program2's NRM needs protection against length zero vectors too
         *
         * TODO: Find out if DP3+NRM+MOV is really faster than DP3+RSQ+MUL
         */
        shader_addline(buffer, "DP3C TA, %s, %s;\n", src_name, src_name);
        shader_addline(buffer, "NRM%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
        shader_addline(buffer, "MOV %s (EQ), %s;\n", dst_name, zero);
    }
    else if(priv->target_version >= NV2)
    {
        shader_addline(buffer, "DP3C TA.x, %s, %s;\n", src_name, src_name);
        shader_addline(buffer, "RSQ TA.x (NE), TA.x;\n");
        shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
                       src_name);
    }
    else
    {
        const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
 
        shader_addline(buffer, "DP3 TA.x, %s, %s;\n", src_name, src_name);
        /* Pass any non-zero value to RSQ if the input vector has a length of zero. The
         * RSQ result doesn't matter, as long as multiplying it by 0 returns 0.
         */
        shader_addline(buffer, "SGE TA.y, -TA.x, %s;\n", zero);
        shader_addline(buffer, "MAD TA.x, %s, TA.y, TA.x;\n", one);
 
        shader_addline(buffer, "RSQ TA.x, TA.x;\n");
        /* dst.w = src[0].w * 1 / (src.x^2 + src.y^2 + src.z^2)^(1/2) according to msdn*/
        shader_addline(buffer, "MUL%s %s, %s, TA.x;\n", shader_arb_get_modifier(ins), dst_name,
                    src_name);
    }
}
 
static void shader_hw_lrp(const struct wined3d_shader_instruction *ins)
{
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    char dst_name[50];
    char src_name[3][50];
 
    /* ARB_fragment_program has a convenient LRP instruction */
    if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
        shader_hw_map2gl(ins);
        return;
    }
 
    shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
    shader_arb_get_src_param(ins, &ins->src[0], 0, src_name[0]);
    shader_arb_get_src_param(ins, &ins->src[1], 1, src_name[1]);
    shader_arb_get_src_param(ins, &ins->src[2], 2, src_name[2]);
 
    shader_addline(buffer, "SUB TA, %s, %s;\n", src_name[1], src_name[2]);
    shader_addline(buffer, "MAD%s %s, %s, TA, %s;\n", shader_arb_get_modifier(ins),
                   dst_name, src_name[0], src_name[2]);
}
 
static void shader_hw_sincos(const struct wined3d_shader_instruction *ins)
{
    /* This instruction exists in ARB, but the d3d instruction takes two extra parameters which
     * must contain fixed constants. So we need a separate function to filter those constants and
     * can't use map2gl
     */
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
    const struct wined3d_shader_dst_param *dst = &ins->dst[0];
    char dst_name[50];
    char src_name0[50], src_name1[50], src_name2[50];
    BOOL is_color;
 
    shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
    if(shader_is_pshader_version(ins->ctx->reg_maps->shader_version.type)) {
        shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
        /* No modifiers are supported on SCS */
        shader_addline(buffer, "SCS %s, %s;\n", dst_name, src_name0);
 
        if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE)
        {
            shader_arb_get_register_name(ins, &dst->reg, src_name0, &is_color);
            shader_addline(buffer, "MOV_SAT %s, %s;\n", dst_name, src_name0);
        }
    } else if(priv->target_version >= NV2) {
        shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
 
        /* Sincos writemask must be .x, .y or .xy */
        if(dst->write_mask & WINED3DSP_WRITEMASK_0)
            shader_addline(buffer, "COS%s %s.x, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
        if(dst->write_mask & WINED3DSP_WRITEMASK_1)
            shader_addline(buffer, "SIN%s %s.y, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name0);
    } else {
        /* Approximate sine and cosine with a taylor series, as per math textbook. The application passes 8
         * helper constants(D3DSINCOSCONST1 and D3DSINCOSCONST2) in src1 and src2.
         *
         * sin(x) = x - x^3/3! + x^5/5! - x^7/7! + ...
         * cos(x) = 1 - x^2/2! + x^4/4! - x^6/6! + ...
         *
         * The constants we get are:
         *
         *  +1   +1,     -1     -1     +1      +1      -1       -1
         *      ---- ,  ---- , ---- , ----- , ----- , ----- , ------
         *      1!*2    2!*4   3!*8   4!*16   5!*32   6!*64   7!*128
         *
         * If used with x^2, x^3, x^4 etc they calculate sin(x/2) and cos(x/2):
         *
         * (x/2)^2 = x^2 / 4
         * (x/2)^3 = x^3 / 8
         * (x/2)^4 = x^4 / 16
         * (x/2)^5 = x^5 / 32
         * etc
         *
         * To get the final result:
         * sin(x) = 2 * sin(x/2) * cos(x/2)
         * cos(x) = cos(x/2)^2 - sin(x/2)^2
         * (from sin(x+y) and cos(x+y) rules)
         *
         * As per MSDN, dst.z is undefined after the operation, and so is
         * dst.x and dst.y if they're masked out by the writemask. Ie
         * sincos dst.y, src1, c0, c1
         * returns the sine in dst.y. dst.x and dst.z are undefined, dst.w is not touched. The assembler
         * vsa.exe also stops with an error if the dest register is the same register as the source
         * register. This means we can use dest.xyz as temporary storage. The assembler vsa.exe output also
         * indicates that sincos consumes 8 instruction slots in vs_2_0(and, strangely, in vs_3_0).
         */
        shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
        shader_arb_get_src_param(ins, &ins->src[2], 2, src_name2);
        shader_arb_get_register_name(ins, &dst->reg, dst_name, &is_color);
 
        shader_addline(buffer, "MUL %s.x, %s, %s;\n", dst_name, src_name0, src_name0);  /* x ^ 2 */
        shader_addline(buffer, "MUL TA.y, %s.x, %s;\n", dst_name, src_name0);           /* x ^ 3 */
        shader_addline(buffer, "MUL %s.y, TA.y, %s;\n", dst_name, src_name0);           /* x ^ 4 */
        shader_addline(buffer, "MUL TA.z, %s.y, %s;\n", dst_name, src_name0);           /* x ^ 5 */
        shader_addline(buffer, "MUL %s.z, TA.z, %s;\n", dst_name, src_name0);           /* x ^ 6 */
        shader_addline(buffer, "MUL TA.w, %s.z, %s;\n", dst_name, src_name0);           /* x ^ 7 */
 
        /* sin(x/2)
         *
         * Unfortunately we don't get the constants in a DP4-capable form. Is there a way to
         * properly merge that with MULs in the code above?
         * The swizzles .yz and xw however fit into the .yzxw swizzle added to ps_2_0. Maybe
         * we can merge the sine and cosine MAD rows to calculate them together.
         */
        shader_addline(buffer, "MUL TA.x, %s, %s.w;\n", src_name0, src_name2); /* x^1, +1/(1!*2) */
        shader_addline(buffer, "MAD TA.x, TA.y, %s.x, TA.x;\n", src_name2); /* -1/(3!*8) */
        shader_addline(buffer, "MAD TA.x, TA.z, %s.w, TA.x;\n", src_name1); /* +1/(5!*32) */
        shader_addline(buffer, "MAD TA.x, TA.w, %s.x, TA.x;\n", src_name1); /* -1/(7!*128) */
 
        /* cos(x/2) */
        shader_addline(buffer, "MAD TA.y, %s.x, %s.y, %s.z;\n", dst_name, src_name2, src_name2); /* -1/(2!*4), +1.0 */
        shader_addline(buffer, "MAD TA.y, %s.y, %s.z, TA.y;\n", dst_name, src_name1); /* +1/(4!*16) */
        shader_addline(buffer, "MAD TA.y, %s.z, %s.y, TA.y;\n", dst_name, src_name1); /* -1/(6!*64) */
 
        if(dst->write_mask & WINED3DSP_WRITEMASK_0) {
            /* cos x */
            shader_addline(buffer, "MUL TA.z, TA.y, TA.y;\n");
            shader_addline(buffer, "MAD %s.x, -TA.x, TA.x, TA.z;\n", dst_name);
        }
        if(dst->write_mask & WINED3DSP_WRITEMASK_1) {
            /* sin x */
            shader_addline(buffer, "MUL %s.y, TA.x, TA.y;\n", dst_name);
            shader_addline(buffer, "ADD %s.y, %s.y, %s.y;\n", dst_name, dst_name, dst_name);
        }
    }
}
 
static void shader_hw_sgn(const struct wined3d_shader_instruction *ins)
{
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    char dst_name[50];
    char src_name[50];
    struct shader_arb_ctx_priv *ctx = ins->ctx->backend_data;
 
    shader_arb_get_dst_param(ins, &ins->dst[0], dst_name);
    shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
 
    /* SGN is only valid in vertex shaders */
    if(ctx->target_version >= NV2) {
        shader_addline(buffer, "SSG%s %s, %s;\n", shader_arb_get_modifier(ins), dst_name, src_name);
        return;
    }
 
    /* If SRC > 0.0, -SRC < SRC = TRUE, otherwise false.
     * if SRC < 0.0,  SRC < -SRC = TRUE. If neither is true, src = 0.0
     */
    if(ins->dst[0].modifiers & WINED3DSPDM_SATURATE) {
        shader_addline(buffer, "SLT %s, -%s, %s;\n", dst_name, src_name, src_name);
    } else {
        /* src contains TA? Write to the dest first. This won't overwrite our destination.
         * Then use TA, and calculate the final result
         *
         * Not reading from TA? Store the first result in TA to avoid overwriting the
         * destination if src reg = dst reg
         */
        if(strstr(src_name, "TA"))
        {
            shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
            shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
            shader_addline(buffer, "ADD %s, %s, -TA;\n", dst_name, dst_name);
        }
        else
        {
            shader_addline(buffer, "SLT TA, -%s, %s;\n", src_name, src_name);
            shader_addline(buffer, "SLT %s,  %s, -%s;\n", dst_name, src_name, src_name);
            shader_addline(buffer, "ADD %s, TA, -%s;\n", dst_name, dst_name);
        }
    }
}
 
static void shader_hw_dsy(const struct wined3d_shader_instruction *ins)
{
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    char src[50];
    char dst[50];
    char dst_name[50];
    BOOL is_color;
 
    shader_arb_get_dst_param(ins, &ins->dst[0], dst);
    shader_arb_get_src_param(ins, &ins->src[0], 0, src);
    shader_arb_get_register_name(ins, &ins->dst[0].reg, dst_name, &is_color);
 
    shader_addline(buffer, "DDY %s, %s;\n", dst, src);
    shader_addline(buffer, "MUL%s %s, %s, ycorrection.y;\n", shader_arb_get_modifier(ins), dst, dst_name);
}
 
static void shader_hw_pow(const struct wined3d_shader_instruction *ins)
{
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    char src0[50], src1[50], dst[50];
    struct wined3d_shader_src_param src0_copy = ins->src[0];
    BOOL need_abs = FALSE;
    struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
    const char *one = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ONE);
 
    /* POW operates on the absolute value of the input */
    src0_copy.modifiers = abs_modifier(src0_copy.modifiers, &need_abs);
 
    shader_arb_get_dst_param(ins, &ins->dst[0], dst);
    shader_arb_get_src_param(ins, &src0_copy, 0, src0);
    shader_arb_get_src_param(ins, &ins->src[1], 1, src1);
 
    if (need_abs)
        shader_addline(buffer, "ABS TA.x, %s;\n", src0);
    else
        shader_addline(buffer, "MOV TA.x, %s;\n", src0);
 
    if (priv->target_version >= NV2)
    {
        shader_addline(buffer, "MOVC TA.y, %s;\n", src1);
        shader_addline(buffer, "POW%s %s, TA.x, TA.y;\n", shader_arb_get_modifier(ins), dst);
        shader_addline(buffer, "MOV %s (EQ.y), %s;\n", dst, one);
    }
    else
    {
        const char *zero = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_ZERO);
        const char *flt_eps = arb_get_helper_value(ins->ctx->reg_maps->shader_version.type, ARB_EPS);
 
        shader_addline(buffer, "ABS TA.y, %s;\n", src1);
        shader_addline(buffer, "SGE TA.y, -TA.y, %s;\n", zero);
        /* Possibly add flt_eps to avoid getting float special values */
        shader_addline(buffer, "MAD TA.z, TA.y, %s, %s;\n", flt_eps, src1);
        shader_addline(buffer, "POW%s TA.x, TA.x, TA.z;\n", shader_arb_get_modifier(ins));
        shader_addline(buffer, "MAD TA.x, -TA.x, TA.y, TA.x;\n");
        shader_addline(buffer, "MAD %s, TA.y, %s, TA.x;\n", dst, one);
    }
}
 
static void shader_hw_loop(const struct wined3d_shader_instruction *ins)
{
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    char src_name[50];
    BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
 
    /* src0 is aL */
    shader_arb_get_src_param(ins, &ins->src[1], 0, src_name);
 
    if(vshader)
    {
        struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
        struct list *e = list_head(&priv->control_frames);
        struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
 
        if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
        /* The constant loader makes sure to load -1 into iX.w */
        shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
        shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
        shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
    }
    else
    {
        shader_addline(buffer, "LOOP %s;\n", src_name);
    }
}
 
static void shader_hw_rep(const struct wined3d_shader_instruction *ins)
{
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    char src_name[50];
    BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
 
    shader_arb_get_src_param(ins, &ins->src[0], 0, src_name);
 
    /* The constant loader makes sure to load -1 into iX.w */
    if(vshader)
    {
        struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
        struct list *e = list_head(&priv->control_frames);
        struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
 
        if(priv->loop_depth > 1) shader_addline(buffer, "PUSHA aL;\n");
 
        shader_addline(buffer, "ARLC aL, %s.xywz;\n", src_name);
        shader_addline(buffer, "BRA loop_%u_end (LE.x);\n", control_frame->no.loop);
        shader_addline(buffer, "loop_%u_start:\n", control_frame->no.loop);
    }
    else
    {
        shader_addline(buffer, "REP %s;\n", src_name);
    }
}
 
static void shader_hw_endloop(const struct wined3d_shader_instruction *ins)
{
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
 
    if(vshader)
    {
        struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
        struct list *e = list_head(&priv->control_frames);
        struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
 
        shader_addline(buffer, "ARAC aL.xy, aL;\n");
        shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
        shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
 
        if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
    }
    else
    {
        shader_addline(buffer, "ENDLOOP;\n");
    }
}
 
static void shader_hw_endrep(const struct wined3d_shader_instruction *ins)
{
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
 
    if(vshader)
    {
        struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
        struct list *e = list_head(&priv->control_frames);
        struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
 
        shader_addline(buffer, "ARAC aL.xy, aL;\n");
        shader_addline(buffer, "BRA loop_%u_start (GT.x);\n", control_frame->no.loop);
        shader_addline(buffer, "loop_%u_end:\n", control_frame->no.loop);
 
        if(priv->loop_depth > 1) shader_addline(buffer, "POPA aL;\n");
    }
    else
    {
        shader_addline(buffer, "ENDREP;\n");
    }
}
 
static const struct control_frame *find_last_loop(const struct shader_arb_ctx_priv *priv)
{
    struct control_frame *control_frame;
 
    LIST_FOR_EACH_ENTRY(control_frame, &priv->control_frames, struct control_frame, entry)
    {
        if(control_frame->type == LOOP || control_frame->type == REP) return control_frame;
    }
    ERR("Could not find loop for break\n");
    return NULL;
}
 
static void shader_hw_break(const struct wined3d_shader_instruction *ins)
{
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
    BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
 
    if(vshader)
    {
        shader_addline(buffer, "BRA loop_%u_end;\n", control_frame->no.loop);
    }
    else
    {
        shader_addline(buffer, "BRK;\n");
    }
}
 
static const char *get_compare(enum wined3d_shader_rel_op op)
{
    switch (op)
    {
        case WINED3D_SHADER_REL_OP_GT: return "GT";
        case WINED3D_SHADER_REL_OP_EQ: return "EQ";
        case WINED3D_SHADER_REL_OP_GE: return "GE";
        case WINED3D_SHADER_REL_OP_LT: return "LT";
        case WINED3D_SHADER_REL_OP_NE: return "NE";
        case WINED3D_SHADER_REL_OP_LE: return "LE";
        default:
            FIXME("Unrecognized operator %#x.\n", op);
            return "(\?\?)";
    }
}
 
static enum wined3d_shader_rel_op invert_compare(enum wined3d_shader_rel_op op)
{
    switch (op)
    {
        case WINED3D_SHADER_REL_OP_GT: return WINED3D_SHADER_REL_OP_LE;
        case WINED3D_SHADER_REL_OP_EQ: return WINED3D_SHADER_REL_OP_NE;
        case WINED3D_SHADER_REL_OP_GE: return WINED3D_SHADER_REL_OP_LT;
        case WINED3D_SHADER_REL_OP_LT: return WINED3D_SHADER_REL_OP_GE;
        case WINED3D_SHADER_REL_OP_NE: return WINED3D_SHADER_REL_OP_EQ;
        case WINED3D_SHADER_REL_OP_LE: return WINED3D_SHADER_REL_OP_GT;
        default:
            FIXME("Unrecognized operator %#x.\n", op);
            return -1;
    }
}
 
static void shader_hw_breakc(const struct wined3d_shader_instruction *ins)
{
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
    const struct control_frame *control_frame = find_last_loop(ins->ctx->backend_data);
    char src_name0[50];
    char src_name1[50];
    const char *comp = get_compare(ins->flags);
 
    shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
    shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
 
    if(vshader)
    {
        /* SUBC CC, src0, src1" works only in pixel shaders, so use TA to throw
         * away the subtraction result
         */
        shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
        shader_addline(buffer, "BRA loop_%u_end (%s.x);\n", control_frame->no.loop, comp);
    }
    else
    {
        shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
        shader_addline(buffer, "BRK (%s.x);\n", comp);
    }
}
 
static void shader_hw_ifc(const struct wined3d_shader_instruction *ins)
{
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
    struct list *e = list_head(&priv->control_frames);
    struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
    const char *comp;
    char src_name0[50];
    char src_name1[50];
    BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
 
    shader_arb_get_src_param(ins, &ins->src[0], 0, src_name0);
    shader_arb_get_src_param(ins, &ins->src[1], 1, src_name1);
 
    if(vshader)
    {
        /* Invert the flag. We jump to the else label if the condition is NOT true */
        comp = get_compare(invert_compare(ins->flags));
        shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
        shader_addline(buffer, "BRA ifc_%u_else (%s.x);\n", control_frame->no.ifc, comp);
    }
    else
    {
        comp = get_compare(ins->flags);
        shader_addline(buffer, "SUBC TA, %s, %s;\n", src_name0, src_name1);
        shader_addline(buffer, "IF %s.x;\n", comp);
    }
}
 
static void shader_hw_else(const struct wined3d_shader_instruction *ins)
{
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
    struct list *e = list_head(&priv->control_frames);
    struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
    BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
 
    if(vshader)
    {
        shader_addline(buffer, "BRA ifc_%u_endif;\n", control_frame->no.ifc);
        shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
        control_frame->had_else = TRUE;
    }
    else
    {
        shader_addline(buffer, "ELSE;\n");
    }
}
 
static void shader_hw_endif(const struct wined3d_shader_instruction *ins)
{
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
    struct list *e = list_head(&priv->control_frames);
    struct control_frame *control_frame = LIST_ENTRY(e, struct control_frame, entry);
    BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
 
    if(vshader)
    {
        if(control_frame->had_else)
        {
            shader_addline(buffer, "ifc_%u_endif:\n", control_frame->no.ifc);
        }
        else
        {
            shader_addline(buffer, "#No else branch. else is endif\n");
            shader_addline(buffer, "ifc_%u_else:\n", control_frame->no.ifc);
        }
    }
    else
    {
        shader_addline(buffer, "ENDIF;\n");
    }
}
 
static void shader_hw_texldd(const struct wined3d_shader_instruction *ins)
{
    DWORD sampler_idx = ins->src[1].reg.idx[0].offset;
    char reg_dest[40];
    char reg_src[3][40];
    WORD flags = TEX_DERIV;
 
    shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
    shader_arb_get_src_param(ins, &ins->src[0], 0, reg_src[0]);
    shader_arb_get_src_param(ins, &ins->src[2], 1, reg_src[1]);
    shader_arb_get_src_param(ins, &ins->src[3], 2, reg_src[2]);
 
    if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
    if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
 
    shader_hw_sample(ins, sampler_idx, reg_dest, reg_src[0], flags, reg_src[1], reg_src[2]);
}
 
static void shader_hw_texldl(const struct wined3d_shader_instruction *ins)
{
    DWORD sampler_idx = ins->src[1].reg.idx[0].offset;
    char reg_dest[40];
    char reg_coord[40];
    WORD flags = TEX_LOD;
 
    shader_arb_get_dst_param(ins, &ins->dst[0], reg_dest);
    shader_arb_get_src_param(ins, &ins->src[0], 0, reg_coord);
 
    if (ins->flags & WINED3DSI_TEXLD_PROJECT) flags |= TEX_PROJ;
    if (ins->flags & WINED3DSI_TEXLD_BIAS) flags |= TEX_BIAS;
 
    shader_hw_sample(ins, sampler_idx, reg_dest, reg_coord, flags, NULL, NULL);
}
 
static void shader_hw_label(const struct wined3d_shader_instruction *ins)
{
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
 
    priv->in_main_func = FALSE;
    /* Call instructions activate the NV extensions, not labels and rets. If there is an uncalled
     * subroutine, don't generate a label that will make GL complain
     */
    if(priv->target_version == ARB) return;
 
    shader_addline(buffer, "l%u:\n", ins->src[0].reg.idx[0].offset);
}
 
static void vshader_add_footer(struct shader_arb_ctx_priv *priv_ctx,
        const struct arb_vshader_private *shader_data, const struct arb_vs_compile_args *args,
        const struct wined3d_shader_reg_maps *reg_maps, const struct wined3d_gl_info *gl_info,
        struct wined3d_string_buffer *buffer)
{
    unsigned int i;
 
    /* The D3DRS_FOGTABLEMODE render state defines if the shader-generated fog coord is used
     * or if the fragment depth is used. If the fragment depth is used(FOGTABLEMODE != NONE),
     * the fog frag coord is thrown away. If the fog frag coord is used, but not written by
     * the shader, it is set to 0.0(fully fogged, since start = 1.0, end = 0.0)
     */
    if (args->super.fog_src == VS_FOG_Z)
    {
        shader_addline(buffer, "MOV result.fogcoord, TMP_OUT.z;\n");
    }
    else
    {
        if (!reg_maps->fog)
        {
            /* posFixup.x is always 1.0, so we can safely use it */
            shader_addline(buffer, "ADD result.fogcoord, posFixup.x, -posFixup.x;\n");
        }
        else
        {
            /* Clamp fogcoord */
            const char *zero = arb_get_helper_value(reg_maps->shader_version.type, ARB_ZERO);
            const char *one = arb_get_helper_value(reg_maps->shader_version.type, ARB_ONE);
 
            shader_addline(buffer, "MIN TMP_FOGCOORD.x, TMP_FOGCOORD.x, %s;\n", one);
            shader_addline(buffer, "MAX result.fogcoord.x, TMP_FOGCOORD.x, %s;\n", zero);
        }
    }
 
    /* Clipplanes are always stored without y inversion */
    if (use_nv_clip(gl_info) && priv_ctx->target_version >= NV2)
    {
        if (args->super.clip_enabled)
        {
            for (i = 0; i < priv_ctx->vs_clipplanes; i++)
            {
                shader_addline(buffer, "DP4 result.clip[%u].x, TMP_OUT, state.clip[%u].plane;\n", i, i);
            }
        }
    }
    else if (args->clip.boolclip.clip_texcoord)
    {
        static const char component[4] = {'x', 'y', 'z', 'w'};
        unsigned int cur_clip = 0;
        const char *zero = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_ZERO);
 
        for (i = 0; i < gl_info->limits.user_clip_distances; ++i)
        {
            if (args->clip.boolclip.clipplane_mask & (1u << i))
            {
                shader_addline(buffer, "DP4 TA.%c, TMP_OUT, state.clip[%u].plane;\n",
                               component[cur_clip++], i);
            }
        }
        switch (cur_clip)
        {
            case 0:
                shader_addline(buffer, "MOV TA, %s;\n", zero);
                break;
            case 1:
                shader_addline(buffer, "MOV TA.yzw, %s;\n", zero);
                break;
            case 2:
                shader_addline(buffer, "MOV TA.zw, %s;\n", zero);
                break;
            case 3:
                shader_addline(buffer, "MOV TA.w, %s;\n", zero);
                break;
        }
        shader_addline(buffer, "MOV result.texcoord[%u], TA;\n",
                       args->clip.boolclip.clip_texcoord - 1);
    }
 
    /* Write the final position.
     *
     * OpenGL coordinates specify the center of the pixel while d3d coords specify
     * the corner. The offsets are stored in z and w in posFixup. posFixup.y contains
     * 1.0 or -1.0 to turn the rendering upside down for offscreen rendering. PosFixup.x
     * contains 1.0 to allow a mad, but arb vs swizzles are too restricted for that.
     */
    if (!gl_info->supported[ARB_CLIP_CONTROL])
    {
        shader_addline(buffer, "MUL TA, posFixup, TMP_OUT.w;\n");
        shader_addline(buffer, "ADD TMP_OUT.x, TMP_OUT.x, TA.z;\n");
        shader_addline(buffer, "MAD TMP_OUT.y, TMP_OUT.y, posFixup.y, TA.w;\n");
 
        /* Z coord [0;1]->[-1;1] mapping, see comment in
         * get_projection_matrix() in utils.c. */
        if (need_helper_const(shader_data, reg_maps, gl_info))
        {
            const char *two = arb_get_helper_value(WINED3D_SHADER_TYPE_VERTEX, ARB_TWO);
            shader_addline(buffer, "MAD TMP_OUT.z, TMP_OUT.z, %s, -TMP_OUT.w;\n", two);
        }
        else
        {
            shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, TMP_OUT.z;\n");
            shader_addline(buffer, "ADD TMP_OUT.z, TMP_OUT.z, -TMP_OUT.w;\n");
        }
    }
 
    shader_addline(buffer, "MOV result.position, TMP_OUT;\n");
 
    priv_ctx->footer_written = TRUE;
}
 
static void shader_hw_ret(const struct wined3d_shader_instruction *ins)
{
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    struct shader_arb_ctx_priv *priv = ins->ctx->backend_data;
    const struct wined3d_shader *shader = ins->ctx->shader;
    BOOL vshader = shader_is_vshader_version(ins->ctx->reg_maps->shader_version.type);
 
    if(priv->target_version == ARB) return;
 
    if(vshader)
    {
        if (priv->in_main_func) vshader_add_footer(priv, shader->backend_data,
                priv->cur_vs_args, ins->ctx->reg_maps, ins->ctx->gl_info, buffer);
    }
 
    shader_addline(buffer, "RET;\n");
}
 
static void shader_hw_call(const struct wined3d_shader_instruction *ins)
{
    struct wined3d_string_buffer *buffer = ins->ctx->buffer;
    shader_addline(buffer, "CAL l%u;\n", ins->src[0].reg.idx[0].offset);
}
 
static BOOL shader_arb_compile(const struct wined3d_gl_info *gl_info, GLenum target, const char *src)
{
    const char *ptr, *line;
    GLint native, pos;
 
    if (TRACE_ON(d3d_shader))
    {
        ptr = src;
        while ((line = get_line(&ptr))) TRACE_(d3d_shader)("    %.*s", (int)(ptr - line), line);
    }
 
    GL_EXTCALL(glProgramStringARB(target, GL_PROGRAM_FORMAT_ASCII_ARB, strlen(src), src));
    checkGLcall("glProgramStringARB()");
 
    if (FIXME_ON(d3d_shader))
    {
        gl_info->gl_ops.gl.p_glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &pos);
        if (pos != -1)
        {
            FIXME_(d3d_shader)("Program error at position %d: %s\n\n", pos,
                    debugstr_a((const char *)gl_info->gl_ops.gl.p_glGetString(GL_PROGRAM_ERROR_STRING_ARB)));
            ptr = src;
            while ((line = get_line(&ptr))) FIXME_(d3d_shader)("    %.*s", (int)(ptr - line), line);
            FIXME_(d3d_shader)("\n");
 
            return FALSE;
        }
    }
 
    if (WARN_ON(d3d_perf))
    {
        GL_EXTCALL(glGetProgramivARB(target, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &native));
        checkGLcall("glGetProgramivARB()");
        if (!native)
            WARN_(d3d_perf)("Program exceeds native resource limits.\n");
    }
 
    return TRUE;
}
 
static void arbfp_add_sRGB_correction(struct wined3d_string_buffer *buffer, const char *fragcolor,
        const char *tmp1, const char *tmp2, const char *tmp3, const char *tmp4, BOOL condcode)
{
    /* Perform sRGB write correction. See GLX_EXT_framebuffer_sRGB */
 
    if(condcode)
    {
        /* Sigh. MOVC CC doesn't work, so use one of the temps as dummy dest */
        shader_addline(buffer, "SUBC %s, %s.x, srgb_consts1.x;\n", tmp1, fragcolor);
        /* Calculate the > 0.0031308 case */
        shader_addline(buffer, "POW %s.x (GE), %s.x, srgb_consts0.x;\n", fragcolor, fragcolor);
        shader_addline(buffer, "POW %s.y (GE), %s.y, srgb_consts0.x;\n", fragcolor, fragcolor);
        shader_addline(buffer, "POW %s.z (GE), %s.z, srgb_consts0.x;\n", fragcolor, fragcolor);
        shader_addline(buffer, "MUL %s.xyz (GE), %s, srgb_consts0.y;\n", fragcolor, fragcolor);
        shader_addline(buffer, "SUB %s.xyz (GE), %s, srgb_consts0.z;\n", fragcolor, fragcolor);
        /* Calculate the < case */
        shader_addline(buffer, "MUL %s.xyz (LT), srgb_consts0.w, %s;\n", fragcolor, fragcolor);
    }
    else
    {
        /* Calculate the > 0.0031308 case */
        shader_addline(buffer, "POW %s.x, %s.x, srgb_consts0.x;\n", tmp1, fragcolor);
        shader_addline(buffer, "POW %s.y, %s.y, srgb_consts0.x;\n", tmp1, fragcolor);
        shader_addline(buffer, "POW %s.z, %s.z, srgb_consts0.x;\n", tmp1, fragcolor);
        shader_addline(buffer, "MUL %s, %s, srgb_consts0.y;\n", tmp1, tmp1);
        shader_addline(buffer, "SUB %s, %s, srgb_consts0.z;\n", tmp1, tmp1);
        /* Calculate the < case */
        shader_addline(buffer, "MUL %s, srgb_consts0.w, %s;\n", tmp2, fragcolor);
        /* Get 1.0 / 0.0 masks for > 0.0031308 and < 0.0031308 */
        shader_addline(buffer, "SLT %s, srgb_consts1.x, %s;\n", tmp3, fragcolor);
        shader_addline(buffer, "SGE %s, srgb_consts1.x, %s;\n", tmp4, fragcolor);
        /* Store the components > 0.0031308 in the destination */
        shader_addline(buffer, "MUL %s.xyz, %s, %s;\n", fragcolor, tmp1, tmp3);
        /* Add the components that are < 0.0031308 */
        shader_addline(buffer, "MAD %s.xyz, %s, %s, %s;\n", fragcolor, tmp2, tmp4, fragcolor);
        /* Move everything into result.color at once. Nvidia hardware cannot handle partial
        * result.color writes(.rgb first, then .a), or handle overwriting already written
        * components. The assembler uses a temporary register in this case, which is usually
        * not allocated from one of our registers that were used earlier.
        */
    }
    /* [0.0;1.0] clamping. Not needed, this is done implicitly */
}
 
static const DWORD *find_loop_control_values(const struct wined3d_shader *shader, DWORD idx)
{
    const struct wined3d_shader_lconst *constant;
 
    LIST_FOR_EACH_ENTRY(constant, &shader->constantsI, struct wined3d_shader_lconst, entry)
    {
        if (constant->idx == idx)
        {
            return constant->value;
        }
    }
    return NULL;
}
 
static void init_ps_input(const struct wined3d_shader *shader,
        const struct arb_ps_compile_args *args, struct shader_arb_ctx_priv *priv)
{
    static const char * const texcoords[8] =
    {
        "fragment.texcoord[0]", "fragment.texcoord[1]", "fragment.texcoord[2]", "fragment.texcoord[3]",
        "fragment.texcoord[4]", "fragment.texcoord[5]", "fragment.texcoord[6]", "fragment.texcoord[7]"
    };
    unsigned int i;
    const struct wined3d_shader_signature_element *input;
    const char *semantic_name;
    DWORD semantic_idx;
 
    switch(args->super.vp_mode)
    {
        case pretransformed:
        case fixedfunction:
            /* The pixelshader has to collect the varyings on its own. In any case properly load
             * color0 and color1. In the case of pretransformed vertices also load texcoords. Set
             * other attribs to 0.0.
             *
             * For fixedfunction this behavior is correct, according to the tests. For pretransformed
             * we'd either need a replacement shader that can load other attribs like BINORMAL, or
             * load the texcoord attrib pointers to match the pixel shader signature
             */
            for (i = 0; i < shader->input_signature.element_count; ++i)
            {
                input = &shader->input_signature.elements[i];
                if (!(semantic_name = input->semantic_name))
                    continue;
                semantic_idx = input->semantic_idx;
 
                if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_COLOR))
                {
                    if (!semantic_idx)
                        priv->ps_input[input->register_idx] = "fragment.color.primary";
                    else if (semantic_idx == 1)
                        priv->ps_input[input->register_idx] = "fragment.color.secondary";
                    else
                        priv->ps_input[input->register_idx] = "0.0";
                }
                else if (args->super.vp_mode == fixedfunction)
                {
                    priv->ps_input[input->register_idx] = "0.0";
                }
                else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_TEXCOORD))
                {
                    if (semantic_idx < 8)
                        priv->ps_input[input->register_idx] = texcoords[semantic_idx];
                    else
                        priv->ps_input[input->register_idx] = "0.0";
                }
                else if (shader_match_semantic(semantic_name, WINED3D_DECL_USAGE_FOG))
                {
                    if (!semantic_idx)
                        priv->ps_input[input->register_idx] = "fragment.fogcoord";
                    else
                        priv->ps_input[input->register_idx] = "0.0";
                }
                else
                {
                    priv->ps_input[input->register_idx] = "0.0";
                }
 
                TRACE("v%u, semantic %s%u is %s\n", input->register_idx,
                        semantic_name, semantic_idx, priv->ps_input[input->register_idx]);
            }
            break;
 
        case vertexshader:
            /* That one is easy. The vertex shaders provide v0-v7 in fragment.texcoord and v8 and v9 in
             * fragment.color
             */
            for(i = 0; i < 8; i++)
            {
                priv->ps_input[i] = texcoords[i];
            }
            priv->ps_input[8] = "fragment.color.primary";
            priv->ps_input[9] = "fragment.color.secondary";
            break;
    }
}
 
static void arbfp_add_linear_fog(struct wined3d_string_buffer *buffer,
        const char *fragcolor, const char *tmp)
{
    shader_addline(buffer, "SUB %s.x, state.fog.params.z, fragment.fogcoord.x;\n", tmp);
    shader_addline(buffer, "MUL_SAT %s.x, %s.x, state.fog.params.w;\n", tmp, tmp);
    shader_addline(buffer, "LRP %s.rgb, %s.x, %s, state.fog.color;\n", fragcolor, tmp, fragcolor);
}
 
/* Context activation is done by the caller. */
static GLuint shader_arb_generate_pshader(const struct wined3d_shader *shader,
        const struct wined3d_gl_info *gl_info, struct wined3d_string_buffer *buffer,
        const struct arb_ps_compile_args *args, struct arb_ps_compiled_shader *compiled)
{
    const struct wined3d_shader_reg_maps *reg_maps = &shader->reg_maps;
    GLuint retval;
    char fragcolor[16];
    DWORD next_local = 0;
    struct shader_arb_ctx_priv priv_ctx;
    BOOL dcl_td = FALSE;
    BOOL want_nv_prog = FALSE;
    struct arb_pshader_private *shader_priv = shader->backend_data;
    DWORD map;
    BOOL custom_linear_fog = FALSE;
 
    char srgbtmp[4][4];
    char ftoa_tmp[17];
    unsigned int i, found = 0;
 
    for (i = 0, map = reg_maps->temporary; map; map >>= 1, ++i)
    {
        if (!(map & 1)
                || (shader->u.ps.color0_mov && i == shader->u.ps.color0_reg)
                || (reg_maps->shader_version.major < 2 && !i))
            continue;
 
        sprintf(srgbtmp[found], "R%u", i);
        ++found;
        if (found == 4) break;
    }
 
    switch(found) {
        case 0:
            sprintf(srgbtmp[0], "TA");
            sprintf(srgbtmp[1], "TB");
            sprintf(srgbtmp[2], "TC");
            sprintf(srgbtmp[3], "TD");
            dcl_td = TRUE;
            break;
        case 1:
            sprintf(srgbtmp[1], "TA");
            sprintf(srgbtmp[2], "TB");
            sprintf(srgbtmp[3], "TC");
            break;
        case 2:
            sprintf(srgbtmp[2], "TA");
            sprintf(srgbtmp[3], "TB");
            break;
        case 3:
            sprintf(srgbtmp[3], "TA");
            break;
        case 4:
            break;
    }
 
    /*  Create the hw ARB shader */
    memset(&priv_ctx, 0, sizeof(priv_ctx));
    priv_ctx.cur_ps_args = args;
    priv_ctx.compiled_fprog = compiled;
    priv_ctx.cur_np2fixup_info = &compiled->np2fixup_info;
    init_ps_input(shader, args, &priv_ctx);
    list_init(&priv_ctx.control_frames);
    priv_ctx.ps_post_process = args->super.srgb_correction;
 
    /* Avoid enabling NV_fragment_program* if we do not need it.
     *
     * Enabling GL_NV_fragment_program_option causes the driver to occupy a temporary register,
     * and it slows down the shader execution noticeably(about 5%). Usually our instruction emulation
     * is faster than what we gain from using higher native instructions. There are some things though
     * that cannot be emulated. In that case enable the extensions.
     * If the extension is enabled, instruction handlers that support both ways will use it.
     *
     * Testing shows no performance difference between OPTION NV_fragment_program2 and NV_fragment_program.
     * So enable the best we can get.
     */
    if(reg_maps->usesdsx || reg_maps->usesdsy || reg_maps->loop_depth > 0 || reg_maps->usestexldd ||
       reg_maps->usestexldl || reg_maps->usesfacing || reg_maps->usesifc || reg_maps->usescall)
    {
        want_nv_prog = TRUE;
    }
 
    shader_addline(buffer, "!!ARBfp1.0\n");
    if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM2])
    {
        shader_addline(buffer, "OPTION NV_fragment_program2;\n");
        priv_ctx.target_version = NV3;
    }
    else if (want_nv_prog && gl_info->supported[NV_FRAGMENT_PROGRAM_OPTION])
    {
        shader_addline(buffer, "OPTION NV_fragment_program;\n");
        priv_ctx.target_version = NV2;
    } else {
        if(want_nv_prog)
        {
            /* This is an error - either we're advertising the wrong shader version, or aren't enforcing some
             * limits properly
             */
            ERR("The shader requires instructions that are not available in plain GL_ARB_fragment_program\n");
            ERR("Try GLSL\n");
        }
        priv_ctx.target_version = ARB;
    }
 
    if (reg_maps->rt_mask > 1)
    {
        shader_addline(buffer, "OPTION ARB_draw_buffers;\n");
    }
 
    if (reg_maps->shader_version.major < 3)
    {
        switch (args->super.fog)
        {
            case WINED3D_FFP_PS_FOG_OFF:
                break;
            case WINED3D_FFP_PS_FOG_LINEAR:
                if (gl_info->quirks & WINED3D_QUIRK_BROKEN_ARB_FOG)
                {
                    custom_linear_fog = TRUE;
                    priv_ctx.ps_post_process = TRUE;
                    break;
                }
                shader_addline(buffer, "OPTION ARB_fog_linear;\n");
                break;
            case WINED3D_FFP_PS_FOG_EXP:
                shader_addline(buffer, "OPTION ARB_fog_exp;\n");
                break;
            case WINED3D_FFP_PS_FOG_EXP2:
                shader_addline(buffer, "OPTION ARB_fog_exp2;\n");
                break;
        }
    }
 
    /* For now always declare the temps. At least the Nvidia assembler optimizes completely
     * unused temps away(but occupies them for the whole shader if they're used once). Always
     * declaring them avoids tricky bookkeeping work
     */
    shader_addline(buffer, "TEMP TA;\n");      /* Used for modifiers */
    shader_addline(buffer, "TEMP TB;\n");      /* Used for modifiers */
    shader_addline(buffer, "TEMP TC;\n");      /* Used for modifiers */
    if(dcl_td) shader_addline(buffer, "TEMP TD;\n"); /* Used for sRGB writing */
    shader_addline(buffer, "PARAM coefdiv = { 0.5, 0.25, 0.125, 0.0625 };\n");
    shader_addline(buffer, "PARAM coefmul = { 2, 4, 8, 16 };\n");
    wined3d_ftoa(eps, ftoa_tmp);
    shader_addline(buffer, "PARAM ps_helper_const = { 0.0, 1.0, %s, 0.0 };\n", ftoa_tmp);
 
    if (reg_maps->shader_version.major < 2)
    {
        strcpy(fragcolor, "R0");
    }
    else
    {
        if (priv_ctx.ps_post_process)
        {
            if (shader->u.ps.color0_mov)
            {
                sprintf(fragcolor, "R%u", shader->u.ps.color0_reg);
            }
            else
            {
                shader_addline(buffer, "TEMP TMP_COLOR;\n");
                strcpy(fragcolor, "TMP_COLOR");
            }
        } else {
            strcpy(fragcolor, "result.color");
        }
    }
 
    if (args->super.srgb_correction)
    {
        shader_addline(buffer, "PARAM srgb_consts0 = ");
        shader_arb_append_imm_vec4(buffer, wined3d_srgb_const0);
        shader_addline(buffer, ";\n");
        shader_addline(buffer, "PARAM srgb_consts1 = ");
        shader_arb_append_imm_vec4(buffer, wined3d_srgb_const1);
        shader_addline(buffer, ";\n");
    }
 
    /* Base Declarations */
    shader_generate_arb_declarations(shader, reg_maps, buffer, gl_info, NULL, &priv_ctx);
 
    for (i = 0, map = reg_maps->bumpmat; map; map >>= 1, ++i)
    {
        unsigned char bump_const;
 
        if (!(map & 1)) continue;
 
        bump_const = compiled->numbumpenvmatconsts;
        compiled->bumpenvmatconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
        compiled->bumpenvmatconst[bump_const].texunit = i;
        compiled->luminanceconst[bump_const].const_num = WINED3D_CONST_NUM_UNUSED;
        compiled->luminanceconst[bump_const].texunit = i;
 
        /* We can fit the constants into the constant limit for sure because texbem, texbeml, bem and beml are only supported
         * in 1.x shaders, and GL_ARB_fragment_program has a constant limit of 24 constants. So in the worst case we're loading
         * 8 shader constants, 8 bump matrices and 8 luminance parameters and are perfectly fine. (No NP2 fixup on bumpmapped
         * textures due to conditional NP2 restrictions)
         *
         * Use local constants to load the bump env parameters, not program.env. This avoids collisions with d3d constants of
         * shaders in newer shader models. Since the bump env parameters have to share their space with NP2 fixup constants,
         * their location is shader dependent anyway and they cannot be loaded globally.
         */
        compiled->bumpenvmatconst[bump_const].const_num = next_local++;
        shader_addline(buffer, "PARAM bumpenvmat%d = program.local[%d];\n",
                       i, compiled->bumpenvmatconst[bump_const].const_num);
        compiled->numbumpenvmatconsts = bump_const + 1;
 
        if (!(reg_maps->luminanceparams & (1u << i)))
            continue;
 
        compiled->luminanceconst[bump_const].const_num = next_local++;
        shader_addline(buffer, "PARAM luminance%d = program.local[%d];\n",
                       i, compiled->luminanceconst[bump_const].const_num);
    }
 
    for (i = 0; i < WINED3D_MAX_CONSTS_I; ++i)
    {
        compiled->int_consts[i] = WINED3D_CONST_NUM_UNUSED;
        if (reg_maps->integer_constants & (1u << i) && priv_ctx.target_version >= NV2)
        {
            const DWORD *control_values = find_loop_control_values(shader, i);
 
            if(control_values)
            {
                shader_addline(buffer, "PARAM I%u = {%u, %u, %u, -1};\n", i,
                                control_values[0], control_values[1], control_values[2]);
            }
            else
            {
                compiled->int_consts[i] = next_local;
                compiled->num_int_consts++;
                shader_addline(buffer, "PARAM I%u = program.local[%u];\n", i, next_local++);
            }
        }
    }
 
    if(reg_maps->vpos || reg_maps->usesdsy)
    {
        compiled->ycorrection = next_local;
        shader_addline(buffer, "PARAM ycorrection = program.local[%u];\n", next_local++);
 
        if(reg_maps->vpos)
        {
            shader_addline(buffer, "TEMP vpos;\n");
            /* ycorrection.x: Backbuffer height(onscreen) or 0(offscreen).
             * ycorrection.y: -1.0(onscreen), 1.0(offscreen)
             * ycorrection.z: 1.0
             * ycorrection.w: 0.0
             */
            shader_addline(buffer, "MAD vpos, fragment.position, ycorrection.zyww, ycorrection.wxww;\n");
            shader_addline(buffer, "FLR vpos.xy, vpos;\n");
        }
    }
    else
    {
        compiled->ycorrection = WINED3D_CONST_NUM_UNUSED;
    }
 
    /* Load constants to fixup NP2 texcoords if there are still free constants left:
     * Constants (texture dimensions) for the NP2 fixup are loaded as local program parameters. This will consume
     * at most 8 (MAX_FRAGMENT_SAMPLERS / 2) parameters, which is highly unlikely, since the application had to
     * use 16 NP2 textures at the same time. In case that we run out of constants the fixup is simply not
     * applied / activated. This will probably result in wrong rendering of the texture, but will save us from
     * shader compilation errors and the subsequent errors when drawing with this shader. */