glsl_shader.c

这是一个开放源代码的与WINNT/WIN2K/WIN2003兼容的操作系统

    tmpArg.src[0]      = arg->src[0];
    tmpArg.src_addr[0] = arg->src_addr[0];
    tmpArg.src_addr[1] = arg->src_addr[1];
    tmpArg.reg_maps = arg->reg_maps; 
    
    switch(arg->opcode->opcode) {
        case WINED3DSIO_M4x4:
            nComponents = 4;
            tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP4);
            break;
        case WINED3DSIO_M4x3:
            nComponents = 3;
            tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP4);
            break;
        case WINED3DSIO_M3x4:
            nComponents = 4;
            tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP3);
            break;
        case WINED3DSIO_M3x3:
            nComponents = 3;
            tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP3);
            break;
        case WINED3DSIO_M3x2:
            nComponents = 2;
            tmpArg.opcode = shader_get_opcode(arg->shader, WINED3DSIO_DP3);
            break;
        default:
            break;
    }

    for (i = 0; i < nComponents; i++) {
        tmpArg.dst = ((arg->dst) & ~WINED3DSP_WRITEMASK_ALL)|(WINED3DSP_WRITEMASK_0<<i);
        tmpArg.src[1]      = arg->src[1]+i;
        shader_glsl_dot(&tmpArg);
    }
}

/**
    The LRP instruction performs a component-wise linear interpolation 
    between the second and third operands using the first operand as the
    blend factor.  Equation:  (dst = src2 + src0 * (src1 - src2))
    This is equivalent to mix(src2, src1, src0);
*/
void shader_glsl_lrp(SHADER_OPCODE_ARG* arg) {
    glsl_src_param_t src0_param;
    glsl_src_param_t src1_param;
    glsl_src_param_t src2_param;
    DWORD write_mask;

    write_mask = shader_glsl_append_dst(arg->buffer, arg);

    shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], write_mask, &src0_param);
    shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], write_mask, &src1_param);
    shader_glsl_add_src_param(arg, arg->src[2], arg->src_addr[2], write_mask, &src2_param);

    shader_addline(arg->buffer, "mix(%s, %s, %s));\n",
            src2_param.param_str, src1_param.param_str, src0_param.param_str);
}

/** Process the WINED3DSIO_LIT instruction in GLSL:
 * dst.x = dst.w = 1.0
 * dst.y = (src0.x > 0) ? src0.x
 * dst.z = (src0.x > 0) ? ((src0.y > 0) ? pow(src0.y, src.w) : 0) : 0
 *                                        where src.w is clamped at +- 128
 */
void shader_glsl_lit(SHADER_OPCODE_ARG* arg) {
    glsl_src_param_t src0_param;
    glsl_src_param_t src1_param;
    glsl_src_param_t src3_param;
    char dst_mask[6];

    shader_glsl_append_dst(arg->buffer, arg);
    shader_glsl_get_write_mask(arg->dst, dst_mask);

    shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
    shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_1, &src1_param);
    shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_3, &src3_param);

    shader_addline(arg->buffer, "vec4(1.0, (%s > 0.0 ? %s : 0.0), (%s > 0.0 ? ((%s > 0.0) ? pow(%s, clamp(%s, -128.0, 128.0)) : 0.0) : 0.0), 1.0)%s);\n",
        src0_param.param_str, src0_param.param_str, src0_param.param_str, src1_param.param_str, src1_param.param_str, src3_param.param_str, dst_mask);
}

/** Process the WINED3DSIO_DST instruction in GLSL:
 * dst.x = 1.0
 * dst.y = src0.x * src0.y
 * dst.z = src0.z
 * dst.w = src1.w
 */
void shader_glsl_dst(SHADER_OPCODE_ARG* arg) {
    glsl_src_param_t src0y_param;
    glsl_src_param_t src0z_param;
    glsl_src_param_t src1y_param;
    glsl_src_param_t src1w_param;
    char dst_mask[6];

    shader_glsl_append_dst(arg->buffer, arg);
    shader_glsl_get_write_mask(arg->dst, dst_mask);

    shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_1, &src0y_param);
    shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_2, &src0z_param);
    shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_1, &src1y_param);
    shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_3, &src1w_param);

    shader_addline(arg->buffer, "vec4(1.0, %s * %s, %s, %s))%s;\n",
            src0y_param.param_str, src1y_param.param_str, src0z_param.param_str, src1w_param.param_str, dst_mask);
}

/** Process the WINED3DSIO_SINCOS instruction in GLSL:
 * VS 2.0 requires that specific cosine and sine constants be passed to this instruction so the hardware
 * can handle it.  But, these functions are built-in for GLSL, so we can just ignore the last 2 params.
 * 
 * dst.x = cos(src0.?)
 * dst.y = sin(src0.?)
 * dst.z = dst.z
 * dst.w = dst.w
 */
void shader_glsl_sincos(SHADER_OPCODE_ARG* arg) {
    glsl_src_param_t src0_param;
    DWORD write_mask;

    write_mask = shader_glsl_append_dst(arg->buffer, arg);
    shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);

    switch (write_mask) {
        case WINED3DSP_WRITEMASK_0:
            shader_addline(arg->buffer, "cos(%s));\n", src0_param.param_str);
            break;

        case WINED3DSP_WRITEMASK_1:
            shader_addline(arg->buffer, "sin(%s));\n", src0_param.param_str);
            break;

        case (WINED3DSP_WRITEMASK_0 | WINED3DSP_WRITEMASK_1):
            shader_addline(arg->buffer, "vec2(cos(%s), sin(%s)));\n", src0_param.param_str, src0_param.param_str);
            break;

        default:
            ERR("Write mask should be .x, .y or .xy\n");
            break;
    }
}

/** Process the WINED3DSIO_LOOP instruction in GLSL:
 * Start a for() loop where src1.y is the initial value of aL,
 *  increment aL by src1.z for a total of src1.x iterations.
 *  Need to use a temporary variable for this operation.
 */
/* FIXME: I don't think nested loops will work correctly this way. */
void shader_glsl_loop(SHADER_OPCODE_ARG* arg) {
    glsl_src_param_t src1_param;

    shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_ALL, &src1_param);
  
    shader_addline(arg->buffer, "for (tmpInt = 0, aL = %s.y; tmpInt < %s.x; tmpInt++, aL += %s.z) {\n",
            src1_param.reg_name, src1_param.reg_name, src1_param.reg_name);
}

void shader_glsl_end(SHADER_OPCODE_ARG* arg) {
    shader_addline(arg->buffer, "}\n");
}

void shader_glsl_rep(SHADER_OPCODE_ARG* arg) {
    glsl_src_param_t src0_param;

    shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
    shader_addline(arg->buffer, "for (tmpInt = 0; tmpInt < %s; tmpInt++) {\n", src0_param.param_str);
}

void shader_glsl_if(SHADER_OPCODE_ARG* arg) {
    glsl_src_param_t src0_param;

    shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
    shader_addline(arg->buffer, "if (%s) {\n", src0_param.param_str);
}

void shader_glsl_ifc(SHADER_OPCODE_ARG* arg) {
    glsl_src_param_t src0_param;
    glsl_src_param_t src1_param;

    shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
    shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);

    shader_addline(arg->buffer, "if (%s %s %s) {\n",
            src0_param.param_str, shader_get_comp_op(arg->opcode_token), src1_param.param_str);
}

void shader_glsl_else(SHADER_OPCODE_ARG* arg) {
    shader_addline(arg->buffer, "} else {\n");
}

void shader_glsl_break(SHADER_OPCODE_ARG* arg) {
    shader_addline(arg->buffer, "break;\n");
}

/* FIXME: According to MSDN the compare is done per component. */
void shader_glsl_breakc(SHADER_OPCODE_ARG* arg) {
    glsl_src_param_t src0_param;
    glsl_src_param_t src1_param;

    shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], WINED3DSP_WRITEMASK_0, &src0_param);
    shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);

    shader_addline(arg->buffer, "if (%s %s %s) break;\n",
            src0_param.param_str, shader_get_comp_op(arg->opcode_token), src1_param.param_str);
}

void shader_glsl_label(SHADER_OPCODE_ARG* arg) {

    DWORD snum = (arg->src[0]) & WINED3DSP_REGNUM_MASK;
    shader_addline(arg->buffer, "}\n");
    shader_addline(arg->buffer, "void subroutine%lu () {\n",  snum);
}

void shader_glsl_call(SHADER_OPCODE_ARG* arg) {
    DWORD snum = (arg->src[0]) & WINED3DSP_REGNUM_MASK;
    shader_addline(arg->buffer, "subroutine%lu();\n", snum);
}

void shader_glsl_callnz(SHADER_OPCODE_ARG* arg) {
    glsl_src_param_t src1_param;

    DWORD snum = (arg->src[0]) & WINED3DSP_REGNUM_MASK;
    shader_glsl_add_src_param(arg, arg->src[1], arg->src_addr[1], WINED3DSP_WRITEMASK_0, &src1_param);
    shader_addline(arg->buffer, "if (%s) subroutine%lu();\n", src1_param.param_str, snum);
}

/*********************************************
 * Pixel Shader Specific Code begins here
 ********************************************/
void pshader_glsl_tex(SHADER_OPCODE_ARG* arg) {
    IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
    DWORD hex_version = This->baseShader.hex_version;
    char dst_swizzle[6];
    glsl_sample_function_t sample_function;
    DWORD sampler_type;
    DWORD sampler_idx;
    BOOL projected;
    DWORD mask = 0;

    /* All versions have a destination register */
    shader_glsl_append_dst(arg->buffer, arg);

    /* 1.0-1.4: Use destination register as sampler source.
     * 2.0+: Use provided sampler source. */
    if (hex_version < WINED3DPS_VERSION(1,4)) {
        IWineD3DDeviceImpl* deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
        DWORD flags;

        sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;
        flags = deviceImpl->stateBlock->textureState[sampler_idx][WINED3DTSS_TEXTURETRANSFORMFLAGS];

        if (flags & WINED3DTTFF_PROJECTED) {
            projected = TRUE;
            switch (flags & ~WINED3DTTFF_PROJECTED) {
                case WINED3DTTFF_COUNT1: FIXME("WINED3DTTFF_PROJECTED with WINED3DTTFF_COUNT1?\n"); break;
                case WINED3DTTFF_COUNT2: mask = WINED3DSP_WRITEMASK_1; break;
                case WINED3DTTFF_COUNT3: mask = WINED3DSP_WRITEMASK_2; break;
                case WINED3DTTFF_COUNT4:
                case WINED3DTTFF_DISABLE: mask = WINED3DSP_WRITEMASK_3; break;
            }
        } else {
            projected = FALSE;
        }
    } else if (hex_version < WINED3DPS_VERSION(2,0)) {
        DWORD src_mod = arg->src[0] & WINED3DSP_SRCMOD_MASK;
        sampler_idx = arg->dst & WINED3DSP_REGNUM_MASK;

        if (src_mod == WINED3DSPSM_DZ) {
            projected = TRUE;
            mask = WINED3DSP_WRITEMASK_2;
        } else if (src_mod == WINED3DSPSM_DW) {
            projected = TRUE;
            mask = WINED3DSP_WRITEMASK_3;
        } else {
            projected = FALSE;
        }
    } else {
        sampler_idx = arg->src[1] & WINED3DSP_REGNUM_MASK;
        if(arg->opcode_token & WINED3DSI_TEXLD_PROJECT) {
                /* ps 2.0 texldp instruction always divides by the fourth component. */
                projected = TRUE;
                mask = WINED3DSP_WRITEMASK_3;
        } else {
            projected = FALSE;
        }
    }

    sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_MASK;
    shader_glsl_get_sample_function(sampler_type, projected, &sample_function);
    mask |= sample_function.coord_mask;

    if (hex_version < WINED3DPS_VERSION(2,0)) {
        shader_glsl_get_write_mask(arg->dst, dst_swizzle);
    } else {
        shader_glsl_get_swizzle(arg->src[1], FALSE, arg->dst, dst_swizzle);
    }

    /* 1.0-1.3: Use destination register as coordinate source.
       1.4+: Use provided coordinate source register. */
    if (hex_version < WINED3DPS_VERSION(1,4)) {
        char coord_mask[6];
        shader_glsl_get_write_mask(mask, coord_mask);
        shader_addline(arg->buffer, "%s(Psampler%u, T%u%s)%s);\n",
                sample_function.name, sampler_idx, sampler_idx, coord_mask, dst_swizzle);
    } else {
        glsl_src_param_t coord_param;
        shader_glsl_add_src_param(arg, arg->src[0], arg->src_addr[0], mask, &coord_param);
        shader_addline(arg->buffer, "%s(Psampler%u, %s)%s);\n",
                sample_function.name, sampler_idx, coord_param.param_str, dst_swizzle);
    }
}

void shader_glsl_texldl(SHADER_OPCODE_ARG* arg) {
    IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*)arg->shader;
    glsl_sample_function_t sample_function;
    glsl_src_param_t coord_param, lod_param;
    char dst_swizzle[6];
    DWORD sampler_type;
    DWORD sampler_idx;

    shader_glsl_append_dst(arg->buffer, arg);
    shader_glsl_get_swizzle(arg->src[1], FALSE, arg->dst, dst_swizzle);

    sampler_idx = arg->src[1] & WINED3DSP_REGNUM_MASK;
    sampler_type = arg->reg_maps->samplers[sampler_idx] & WINED3DSP_TEXTURETYPE_M