diff options
Diffstat (limited to 'src/mesa/drivers/dri/r300')
| -rw-r--r-- | src/mesa/drivers/dri/r300/r300_context.h | 11 | ||||
| -rw-r--r-- | src/mesa/drivers/dri/r300/r300_emit.c | 13 | ||||
| -rw-r--r-- | src/mesa/drivers/dri/r300/r300_reg.h | 23 | ||||
| -rw-r--r-- | src/mesa/drivers/dri/r300/r300_state.c | 91 | ||||
| -rw-r--r-- | src/mesa/drivers/dri/r300/r300_vertprog.c | 1637 | ||||
| -rw-r--r-- | src/mesa/drivers/dri/r300/r300_vertprog.h | 6 |
6 files changed, 1020 insertions, 761 deletions
diff --git a/src/mesa/drivers/dri/r300/r300_context.h b/src/mesa/drivers/dri/r300/r300_context.h index 6615bc79fb..0349bac9a2 100644 --- a/src/mesa/drivers/dri/r300/r300_context.h +++ b/src/mesa/drivers/dri/r300/r300_context.h @@ -568,19 +568,10 @@ struct r300_vertex_shader_fragment { union { GLuint d[VSF_MAX_FRAGMENT_LENGTH]; float f[VSF_MAX_FRAGMENT_LENGTH]; - VERTEX_SHADER_INSTRUCTION i[VSF_MAX_FRAGMENT_LENGTH / 4]; + struct r300_vertprog_instruction i[VSF_MAX_FRAGMENT_LENGTH / 4]; } body; }; -#define VSF_DEST_PROGRAM 0x0 -#define VSF_DEST_MATRIX0 0x200 -#define VSF_DEST_MATRIX1 0x204 -#define VSF_DEST_MATRIX2 0x208 -#define VSF_DEST_VECTOR0 0x20c -#define VSF_DEST_VECTOR1 0x20d -#define VSF_DEST_UNKNOWN1 0x400 -#define VSF_DEST_UNKNOWN2 0x406 - struct r300_vertex_shader_state { struct r300_vertex_shader_fragment program; }; diff --git a/src/mesa/drivers/dri/r300/r300_emit.c b/src/mesa/drivers/dri/r300/r300_emit.c index 229439dfa8..424bf44e59 100644 --- a/src/mesa/drivers/dri/r300/r300_emit.c +++ b/src/mesa/drivers/dri/r300/r300_emit.c @@ -299,13 +299,14 @@ GLuint r300VAPOutputCntl0(GLcontext * ctx, GLuint OutputsWritten) if (OutputsWritten & (1 << VERT_RESULT_COL1)) ret |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_1_PRESENT; -#if 0 - if (OutputsWritten & (1 << VERT_RESULT_BFC0)) - ret |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_2_PRESENT; - - if (OutputsWritten & (1 << VERT_RESULT_BFC1)) - ret |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_3_PRESENT; + if (OutputsWritten & (1 << VERT_RESULT_BFC0) + || OutputsWritten & (1 << VERT_RESULT_BFC1)) + ret |= + R300_VAP_OUTPUT_VTX_FMT_0__COLOR_1_PRESENT | + R300_VAP_OUTPUT_VTX_FMT_0__COLOR_2_PRESENT | + R300_VAP_OUTPUT_VTX_FMT_0__COLOR_3_PRESENT; +#if 0 if (OutputsWritten & (1 << VERT_RESULT_FOGC)) ; #endif diff --git a/src/mesa/drivers/dri/r300/r300_reg.h b/src/mesa/drivers/dri/r300/r300_reg.h index e59919be49..1baa74c526 100644 --- a/src/mesa/drivers/dri/r300/r300_reg.h +++ b/src/mesa/drivers/dri/r300/r300_reg.h @@ -282,9 +282,32 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. */ #define R300_VAP_PVS_UPLOAD_ADDRESS 0x2200 # define R300_PVS_UPLOAD_PROGRAM 0x00000000 +/* gap */ # define R300_PVS_UPLOAD_PARAMETERS 0x00000200 +/* gap */ +# define R300_PVS_UPLOAD_CLIP_PLANE0 0x00000400 +# define R300_PVS_UPLOAD_CLIP_PLANE1 0x00000401 +# define R300_PVS_UPLOAD_CLIP_PLANE2 0x00000402 +# define R300_PVS_UPLOAD_CLIP_PLANE3 0x00000403 +# define R300_PVS_UPLOAD_CLIP_PLANE4 0x00000404 +# define R300_PVS_UPLOAD_CLIP_PLANE5 0x00000405 # define R300_PVS_UPLOAD_POINTSIZE 0x00000406 +/* + * These are obsolete defines form r300_context.h, but they might give some + * clues when investigating the addresses further... + */ +#if 0 +#define VSF_DEST_PROGRAM 0x0 +#define VSF_DEST_MATRIX0 0x200 +#define VSF_DEST_MATRIX1 0x204 +#define VSF_DEST_MATRIX2 0x208 +#define VSF_DEST_VECTOR0 0x20c +#define VSF_DEST_VECTOR1 0x20d +#define VSF_DEST_UNKNOWN1 0x400 +#define VSF_DEST_UNKNOWN2 0x406 +#endif + /* gap */ #define R300_VAP_PVS_UPLOAD_DATA 0x2208 diff --git a/src/mesa/drivers/dri/r300/r300_state.c b/src/mesa/drivers/dri/r300/r300_state.c index b5cf21d644..6789efd428 100644 --- a/src/mesa/drivers/dri/r300/r300_state.c +++ b/src/mesa/drivers/dri/r300/r300_state.c @@ -82,6 +82,8 @@ static void r300BlendColor(GLcontext * ctx, const GLfloat cf[4]) rmesa->hw.blend_color.cmd[1] = PACK_COLOR_8888(color[3], color[0], color[1], color[2]); + rmesa->hw.blend_color.cmd[2] = 0; + rmesa->hw.blend_color.cmd[3] = 0; } /** @@ -317,20 +319,34 @@ static void r300UpdateCulling(GLcontext * ctx) r300ContextPtr r300 = R300_CONTEXT(ctx); uint32_t val = 0; - R300_STATECHANGE(r300, cul); if (ctx->Polygon.CullFlag) { - if (ctx->Polygon.CullFaceMode == GL_FRONT_AND_BACK) - val = R300_CULL_FRONT | R300_CULL_BACK; - else if (ctx->Polygon.CullFaceMode == GL_FRONT) + switch (ctx->Polygon.CullFaceMode) { + case GL_FRONT: val = R300_CULL_FRONT; - else + break; + case GL_BACK: val = R300_CULL_BACK; + break; + case GL_FRONT_AND_BACK: + val = R300_CULL_FRONT | R300_CULL_BACK; + break; + default: + break; + } + } - if (ctx->Polygon.FrontFace == GL_CW) - val |= R300_FRONT_FACE_CW; - else - val |= R300_FRONT_FACE_CCW; + switch (ctx->Polygon.FrontFace) { + case GL_CW: + val |= R300_FRONT_FACE_CW; + break; + case GL_CCW: + val |= R300_FRONT_FACE_CCW; + break; + default: + break; } + + R300_STATECHANGE(r300, cul); r300->hw.cul.cmd[R300_CUL_CULL] = val; } @@ -344,6 +360,20 @@ static void r300SetEarlyZState(GLcontext * ctx) r300ContextPtr r300 = R300_CONTEXT(ctx); R300_STATECHANGE(r300, zstencil_format); + switch (ctx->Visual.depthBits) { + case 16: + r300->hw.zstencil_format.cmd[1] = R300_DEPTH_FORMAT_16BIT_INT_Z; + break; + case 24: + r300->hw.zstencil_format.cmd[1] = R300_DEPTH_FORMAT_24BIT_INT_Z; + break; + default: + fprintf(stderr, "Error: Unsupported depth %d... exiting\n", ctx->Visual.depthBits); + _mesa_exit(-1); + } + + // r300->hw.zstencil_format.cmd[1] |= R300_DEPTH_FORMAT_UNK32; + if (ctx->Color.AlphaEnabled && ctx->Color.AlphaFunc != GL_ALWAYS) /* disable early Z */ r300->hw.zstencil_format.cmd[2] = R300_EARLY_Z_DISABLE; @@ -355,6 +385,9 @@ static void r300SetEarlyZState(GLcontext * ctx) /* disable early Z */ r300->hw.zstencil_format.cmd[2] = R300_EARLY_Z_DISABLE; } + + r300->hw.zstencil_format.cmd[3] = 0x00000003; + r300->hw.zstencil_format.cmd[4] = 0x00000000; } static void r300SetAlphaState(GLcontext * ctx) @@ -403,6 +436,7 @@ static void r300SetAlphaState(GLcontext * ctx) R300_STATECHANGE(r300, at); r300->hw.at.cmd[R300_AT_ALPHA_TEST] = pp_misc; + r300->hw.at.cmd[R300_AT_UNKNOWN] = 0; r300SetEarlyZState(ctx); } @@ -513,6 +547,9 @@ static void r300UpdatePolygonMode(GLcontext * ctx) R300_STATECHANGE(r300, polygon_mode); r300->hw.polygon_mode.cmd[1] = hw_mode; } + + r300->hw.polygon_mode.cmd[2] = 0x00000001; + r300->hw.polygon_mode.cmd[3] = 0x00000000; } /** @@ -762,6 +799,7 @@ static void r300ShadeModel(GLcontext * ctx, GLenum mode) r300ContextPtr rmesa = R300_CONTEXT(ctx); R300_STATECHANGE(rmesa, shade); + rmesa->hw.shade.cmd[1] = 0x00000002; switch (mode) { case GL_FLAT: rmesa->hw.shade.cmd[2] = R300_RE_SHADE_MODEL_FLAT; @@ -772,6 +810,8 @@ static void r300ShadeModel(GLcontext * ctx, GLenum mode) default: return; } + rmesa->hw.shade.cmd[3] = 0x00000000; + rmesa->hw.shade.cmd[4] = 0x00000000; } static void r300StencilFuncSeparate(GLcontext * ctx, GLenum face, @@ -1526,7 +1566,7 @@ static void r300SetupDefaultVertexProgram(r300ContextPtr rmesa) for (i = VERT_ATTRIB_POS; i < VERT_ATTRIB_MAX; i++) { if (rmesa->state.sw_tcl_inputs[i] != -1) { - prog->program.body.i[program_end].op = EASY_VSF_OP(MUL, o_reg++, ALL, RESULT); + prog->program.body.i[program_end].opcode = EASY_VSF_OP(MUL, o_reg++, ALL, RESULT); prog->program.body.i[program_end].src[0] = VSF_REG(rmesa->state.sw_tcl_inputs[i]); prog->program.body.i[program_end].src[1] = VSF_ATTR_UNITY(rmesa->state.sw_tcl_inputs[i]); prog->program.body.i[program_end].src[2] = VSF_UNITY(rmesa->state.sw_tcl_inputs[i]); @@ -1536,7 +1576,7 @@ static void r300SetupDefaultVertexProgram(r300ContextPtr rmesa) prog->program.length = program_end * 4; - r300SetupVertexProgramFragment(rmesa, VSF_DEST_PROGRAM, &(prog->program)); + r300SetupVertexProgramFragment(rmesa, R300_PVS_UPLOAD_PROGRAM, &(prog->program)); inst_count = (prog->program.length / 4) - 1; R300_STATECHANGE(rmesa, pvs); @@ -1570,7 +1610,7 @@ static void r300SetupRealVertexProgram(r300ContextPtr rmesa) bump_vpu_count(rmesa->hw.vpp.cmd, param_count); param_count /= 4; - r300SetupVertexProgramFragment(rmesa, VSF_DEST_PROGRAM, &(prog->program)); + r300SetupVertexProgramFragment(rmesa, R300_PVS_UPLOAD_PROGRAM, &(prog->program)); inst_count = (prog->program.length / 4) - 1; R300_STATECHANGE(rmesa, pvs); @@ -1848,15 +1888,10 @@ static void r300ResetHwState(r300ContextPtr r300) r300->hw.unk4260.cmd[2] = r300PackFloat32(0.0); r300->hw.unk4260.cmd[3] = r300PackFloat32(1.0); - r300->hw.shade.cmd[1] = 0x00000002; r300ShadeModel(ctx, ctx->Light.ShadeModel); - r300->hw.shade.cmd[3] = 0x00000000; - r300->hw.shade.cmd[4] = 0x00000000; r300PolygonMode(ctx, GL_FRONT, ctx->Polygon.FrontMode); r300PolygonMode(ctx, GL_BACK, ctx->Polygon.BackMode); - r300->hw.polygon_mode.cmd[2] = 0x00000001; - r300->hw.polygon_mode.cmd[3] = 0x00000000; r300->hw.zbias_cntl.cmd[1] = 0x00000000; r300PolygonOffset(ctx, ctx->Polygon.OffsetFactor, @@ -1887,14 +1922,11 @@ static void r300ResetHwState(r300ContextPtr r300) r300Fogfv(ctx, GL_FOG_COLOR, ctx->Fog.Color); r300Fogfv(ctx, GL_FOG_COORDINATE_SOURCE_EXT, NULL); - r300->hw.at.cmd[R300_AT_UNKNOWN] = 0; r300->hw.unk4BD8.cmd[1] = 0; r300->hw.unk4E00.cmd[1] = 0; r300BlendColor(ctx, ctx->Color.BlendColor); - r300->hw.blend_color.cmd[2] = 0; - r300->hw.blend_color.cmd[3] = 0; /* Again, r300ClearBuffer uses this */ r300->hw.cb.cmd[R300_CB_OFFSET] = @@ -1925,25 +1957,6 @@ static void r300ResetHwState(r300ContextPtr r300) r300->hw.unk4EA0.cmd[1] = 0x00000000; r300->hw.unk4EA0.cmd[2] = 0xffffffff; - switch (ctx->Visual.depthBits) { - case 16: - r300->hw.zstencil_format.cmd[1] = R300_DEPTH_FORMAT_16BIT_INT_Z; - break; - case 24: - r300->hw.zstencil_format.cmd[1] = R300_DEPTH_FORMAT_24BIT_INT_Z; - break; - default: - fprintf(stderr, "Error: Unsupported depth %d... exiting\n", - ctx->Visual.depthBits); - _mesa_exit(-1); - - } - /* z compress? */ - //r300->hw.zstencil_format.cmd[1] |= R300_DEPTH_FORMAT_UNK32; - - r300->hw.zstencil_format.cmd[3] = 0x00000003; - r300->hw.zstencil_format.cmd[4] = 0x00000000; - r300->hw.zb.cmd[R300_ZB_OFFSET] = r300->radeon.radeonScreen->depthOffset + r300->radeon.radeonScreen->fbLocation; diff --git a/src/mesa/drivers/dri/r300/r300_vertprog.c b/src/mesa/drivers/dri/r300/r300_vertprog.c index 16dddf6557..0fb6110494 100644 --- a/src/mesa/drivers/dri/r300/r300_vertprog.c +++ b/src/mesa/drivers/dri/r300/r300_vertprog.c @@ -29,6 +29,8 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. * \file * * \author Aapo Tahkola <aet@rasterburn.org> + * + * \author Oliver McFadden <z3ro.geek@gmail.com> */ #include "glheader.h" @@ -55,54 +57,58 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. #error Cannot change these! #endif -#define SCALAR_FLAG (1<<31) -#define FLAG_MASK (1<<31) -#define OP_MASK (0xf) /* we are unlikely to have more than 15 */ -#define OPN(operator, ip) {#operator, OPCODE_##operator, ip} - -static struct { - char *name; - int opcode; - unsigned long ip; /* number of input operands and flags */ -} op_names[] = { - /* *INDENT-OFF* */ - OPN(ABS, 1), - OPN(ADD, 2), - OPN(ARL, 1 | SCALAR_FLAG), - OPN(DP3, 2), - OPN(DP4, 2), - OPN(DPH, 2), - OPN(DST, 2), - OPN(EX2, 1 | SCALAR_FLAG), - OPN(EXP, 1 | SCALAR_FLAG), - OPN(FLR, 1), - OPN(FRC, 1), - OPN(LG2, 1 | SCALAR_FLAG), - OPN(LIT, 1), - OPN(LOG, 1 | SCALAR_FLAG), - OPN(MAD, 3), - OPN(MAX, 2), - OPN(MIN, 2), - OPN(MOV, 1), - OPN(MUL, 2), - OPN(POW, 2 | SCALAR_FLAG), - OPN(RCP, 1 | SCALAR_FLAG), - OPN(RSQ, 1 | SCALAR_FLAG), - OPN(SGE, 2), - OPN(SLT, 2), - OPN(SUB, 2), - OPN(SWZ, 1), - OPN(XPD, 2), - OPN(RCC, 0), //extra - OPN(PRINT, 0), - OPN(END, 0) - /* *INDENT-ON* */ -}; - -#undef OPN +/* TODO: Get rid of t_src_class call */ +#define CMP_SRCS(a, b) ((a.RelAddr != b.RelAddr) || (a.Index != b.Index && \ + ((t_src_class(a.File) == VSF_IN_CLASS_PARAM && \ + t_src_class(b.File) == VSF_IN_CLASS_PARAM) || \ + (t_src_class(a.File) == VSF_IN_CLASS_ATTR && \ + t_src_class(b.File) == VSF_IN_CLASS_ATTR)))) \ + +#define ZERO_SRC_0 (MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), \ + SWIZZLE_ZERO, SWIZZLE_ZERO, \ + SWIZZLE_ZERO, SWIZZLE_ZERO, \ + t_src_class(src[0].File), VSF_FLAG_NONE) | (src[0].RelAddr << 4)) + +#define ZERO_SRC_1 (MAKE_VSF_SOURCE(t_src_index(vp, &src[1]), \ + SWIZZLE_ZERO, SWIZZLE_ZERO, \ + SWIZZLE_ZERO, SWIZZLE_ZERO, \ + t_src_class(src[1].File), VSF_FLAG_NONE) | (src[1].RelAddr << 4)) + +#define ZERO_SRC_2 (MAKE_VSF_SOURCE(t_src_index(vp, &src[2]), \ + SWIZZLE_ZERO, SWIZZLE_ZERO, \ + SWIZZLE_ZERO, SWIZZLE_ZERO, \ + t_src_class(src[2].File), VSF_FLAG_NONE) | (src[2].RelAddr << 4)) + +#define ONE_SRC_0 (MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), \ + SWIZZLE_ONE, SWIZZLE_ONE, \ + SWIZZLE_ONE, SWIZZLE_ONE, \ + t_src_class(src[0].File), VSF_FLAG_NONE) | (src[0].RelAddr << 4)) + +#define ONE_SRC_1 (MAKE_VSF_SOURCE(t_src_index(vp, &src[1]), \ + SWIZZLE_ONE, SWIZZLE_ONE, \ + SWIZZLE_ONE, SWIZZLE_ONE, \ + t_src_class(src[1].File), VSF_FLAG_NONE) | (src[1].RelAddr << 4)) + +#define ONE_SRC_2 (MAKE_VSF_SOURCE(t_src_index(vp, &src[2]), \ + SWIZZLE_ONE, SWIZZLE_ONE, \ + SWIZZLE_ONE, SWIZZLE_ONE, \ + t_src_class(src[2].File), VSF_FLAG_NONE) | (src[2].RelAddr << 4)) + +/* DP4 version seems to trigger some hw peculiarity */ +//#define PREFER_DP4 + +#define FREE_TEMPS() \ + do { \ + if(u_temp_i < vp->num_temporaries) { \ + WARN_ONCE("Ran out of temps, num temps %d, us %d\n", vp->num_temporaries, u_temp_i); \ + vp->native = GL_FALSE; \ + } \ + u_temp_i=VSF_MAX_FRAGMENT_TEMPS-1; \ + } while (0) int r300VertexProgUpdateParams(GLcontext * ctx, - struct r300_vertex_program_cont *vp, float *dst) + struct r300_vertex_program_cont *vp, + float *dst) { int pi; struct gl_vertex_program *mesa_vp = &vp->mesa_program; @@ -234,8 +240,8 @@ static void vp_dump_inputs(struct r300_vertex_program *vp, char *caller) int i; if (vp == NULL) { - fprintf(stderr, "vp null in call to %s from %s\n", __FUNCTION__, - caller); + fprintf(stderr, "vp null in call to %s from %s\n", + __FUNCTION__, caller); return; } @@ -276,6 +282,8 @@ static unsigned long t_src_index(struct r300_vertex_program *vp, } } +/* these two functions should probably be merged... */ + static unsigned long t_src(struct r300_vertex_program *vp, struct prog_src_register *src) { @@ -294,7 +302,9 @@ static unsigned long t_src(struct r300_vertex_program *vp, static unsigned long t_src_scalar(struct r300_vertex_program *vp, struct prog_src_register *src) { - + /* src->NegateBase uses the NEGATE_ flags from program_instruction.h, + * which equal our VSF_FLAGS_ values, so it's safe to just pass it here. + */ return MAKE_VSF_SOURCE(t_src_index(vp, src), t_swizzle(GET_SWZ(src->Swizzle, 0)), t_swizzle(GET_SWZ(src->Swizzle, 0)), @@ -306,128 +316,727 @@ static unsigned long t_src_scalar(struct r300_vertex_program *vp, (src->RelAddr << 4); } -static unsigned long t_opcode(enum prog_opcode opcode) +static GLboolean valid_dst(struct r300_vertex_program *vp, + struct prog_dst_register *dst) { + if (dst->File == PROGRAM_OUTPUT && vp->outputs[dst->Index] == -1) { + return GL_FALSE; + } else if (dst->File == PROGRAM_ADDRESS) { + assert(dst->Index == 0); + } + + return GL_TRUE; +} - switch (opcode) { - /* *INDENT-OFF* */ - case OPCODE_ARL: return R300_VPI_OUT_OP_ARL; - case OPCODE_DST: return R300_VPI_OUT_OP_DST; - case OPCODE_EX2: return R300_VPI_OUT_OP_EX2; - case OPCODE_EXP: return R300_VPI_OUT_OP_EXP; - case OPCODE_FRC: return R300_VPI_OUT_OP_FRC; - case OPCODE_LG2: return R300_VPI_OUT_OP_LG2; - case OPCODE_LOG: return R300_VPI_OUT_OP_LOG; - case OPCODE_MAX: return R300_VPI_OUT_OP_MAX; - case OPCODE_MIN: return R300_VPI_OUT_OP_MIN; - case OPCODE_MUL: return R300_VPI_OUT_OP_MUL; - case OPCODE_RCP: return R300_VPI_OUT_OP_RCP; - case OPCODE_RSQ: return R300_VPI_OUT_OP_RSQ; - case OPCODE_SGE: return R300_VPI_OUT_OP_SGE; - case OPCODE_SLT: return R300_VPI_OUT_OP_SLT; - case OPCODE_DP4: return R300_VPI_OUT_OP_DOT; - /* *INDENT-ON* */ +/* + * Instruction Inputs Output Description + * ----------- ------ ------ -------------------------------- + * ABS v v absolute value + * ADD v,v v add + * ARL s a address register load + * DP3 v,v ssss 3-component dot product + * DP4 v,v ssss 4-component dot product + * DPH v,v ssss homogeneous dot product + * DST v,v v distance vector + * EX2 s ssss exponential base 2 + * EXP s v exponential base 2 (approximate) + * FLR v v floor + * FRC v v fraction + * LG2 s ssss logarithm base 2 + * LIT v v compute light coefficients + * LOG s v logarithm base 2 (approximate) + * MAD v,v,v v multiply and add + * MAX v,v v maximum + * MIN v,v v minimum + * MOV v v move + * MUL v,v v multiply + * POW s,s ssss exponentiate + * RCP s ssss reciprocal + * RSQ s ssss reciprocal square root + * SGE v,v v set on greater than or equal + * SLT v,v v set on less than + * SUB v,v v subtract + * SWZ v v extended swizzle + * XPD v,v v cross product + * + * Table X.5: Summary of vertex program instructions. "v" indicates a + * floating-point vector input or output, "s" indicates a floating-point + * scalar input, "ssss" indicates a scalar output replicated across a + * 4-component result vector, and "a" indicates a single address register + * component. + */ - default: - fprintf(stderr, "%s: Should not be called with opcode %d!", - __FUNCTION__, opcode); - } - _mesa_exit(-1); - return 0; +static void t_opcode_abs(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) +{ + //MAX RESULT 1.X Y Z W PARAM 0{} {X Y Z W} PARAM 0{X Y Z W } {X Y Z W} neg Xneg Yneg Zneg W + + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_MAX, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + + o_inst->src[0] = t_src(vp, &src[0]); + o_inst->src[1] = + MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), + t_swizzle(GET_SWZ(src[0].Swizzle, 0)), + t_swizzle(GET_SWZ(src[0].Swizzle, 1)), + t_swizzle(GET_SWZ(src[0].Swizzle, 2)), + t_swizzle(GET_SWZ(src[0].Swizzle, 3)), + t_src_class(src[0].File), + (!src[0]. + NegateBase) ? VSF_FLAG_ALL : VSF_FLAG_NONE) | + (src[0].RelAddr << 4); + o_inst->src[2] = 0; } -static unsigned long op_operands(enum prog_opcode opcode) +static void t_opcode_add(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) { - int i; + unsigned long hw_op; - /* Can we trust mesas opcodes to be in order ? */ - for (i = 0; i < sizeof(op_names) / sizeof(*op_names); i++) - if (op_names[i].opcode == opcode) - return op_names[i].ip; +#if 1 + hw_op = (src[0].File == PROGRAM_TEMPORARY + && src[1].File == + PROGRAM_TEMPORARY) ? R300_VPI_OUT_OP_MAD_2 : + R300_VPI_OUT_OP_MAD; + + o_inst->opcode = + MAKE_VSF_OP(hw_op, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + o_inst->src[0] = ONE_SRC_0; + o_inst->src[1] = t_src(vp, &src[0]); + o_inst->src[2] = t_src(vp, &src[1]); +#else + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_ADD, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + o_inst->src[0] = t_src(vp, &src[0]); + o_inst->src[1] = t_src(vp, &src[1]); + o_inst->src[2] = ZERO_SRC_1; - fprintf(stderr, "op %d not found in op_names\n", opcode); - _mesa_exit(-1); - return 0; +#endif } -static GLboolean valid_dst(struct r300_vertex_program *vp, - struct prog_dst_register *dst) +static void t_opcode_arl(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) { - if (dst->File == PROGRAM_OUTPUT && vp->outputs[dst->Index] == -1) { - return GL_FALSE; - } else if (dst->File == PROGRAM_ADDRESS) { - assert(dst->Index == 0); - } + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_ARL, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + + o_inst->src[0] = t_src(vp, &src[0]); + o_inst->src[1] = ZERO_SRC_0; + o_inst->src[2] = ZERO_SRC_0; +} - return GL_TRUE; +static void t_opcode_dp3(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) +{ + //DOT RESULT 1.X Y Z W PARAM 0{} {X Y Z ZERO} PARAM 0{} {X Y Z ZERO} + + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_DOT, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + + o_inst->src[0] = + MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), + t_swizzle(GET_SWZ(src[0].Swizzle, 0)), + t_swizzle(GET_SWZ(src[0].Swizzle, 1)), + t_swizzle(GET_SWZ(src[0].Swizzle, 2)), + SWIZZLE_ZERO, t_src_class(src[0].File), + src[0]. + NegateBase ? VSF_FLAG_XYZ : VSF_FLAG_NONE) | + (src[0].RelAddr << 4); + + o_inst->src[1] = + MAKE_VSF_SOURCE(t_src_index(vp, &src[1]), + t_swizzle(GET_SWZ(src[1].Swizzle, 0)), + t_swizzle(GET_SWZ(src[1].Swizzle, 1)), + t_swizzle(GET_SWZ(src[1].Swizzle, 2)), + SWIZZLE_ZERO, t_src_class(src[1].File), + src[1]. + NegateBase ? VSF_FLAG_XYZ : VSF_FLAG_NONE) | + (src[1].RelAddr << 4); + + o_inst->src[2] = ZERO_SRC_1; } -/* TODO: Get rid of t_src_class call */ -#define CMP_SRCS(a, b) ((a.RelAddr != b.RelAddr) || (a.Index != b.Index && \ - ((t_src_class(a.File) == VSF_IN_CLASS_PARAM && \ - t_src_class(b.File) == VSF_IN_CLASS_PARAM) || \ - (t_src_class(a.File) == VSF_IN_CLASS_ATTR && \ - t_src_class(b.File) == VSF_IN_CLASS_ATTR)))) \ +static void t_opcode_dp4(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) +{ + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_DOT, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + + o_inst->src[0] = t_src(vp, &src[0]); + o_inst->src[1] = t_src(vp, &src[1]); + o_inst->src[2] = ZERO_SRC_1; +} -#define ZERO_SRC_0 (MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), \ - SWIZZLE_ZERO, SWIZZLE_ZERO, \ - SWIZZLE_ZERO, SWIZZLE_ZERO, \ - t_src_class(src[0].File), VSF_FLAG_NONE) | (src[0].RelAddr << 4)) +static void t_opcode_dph(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) +{ + //DOT RESULT 1.X Y Z W PARAM 0{} {X Y Z ONE} PARAM 0{} {X Y Z W} + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_DOT, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + + o_inst->src[0] = + MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), + t_swizzle(GET_SWZ(src[0].Swizzle, 0)), + t_swizzle(GET_SWZ(src[0].Swizzle, 1)), + t_swizzle(GET_SWZ(src[0].Swizzle, 2)), + VSF_IN_COMPONENT_ONE, t_src_class(src[0].File), + src[0]. + NegateBase ? VSF_FLAG_XYZ : VSF_FLAG_NONE) | + (src[0].RelAddr << 4); + o_inst->src[1] = t_src(vp, &src[1]); + o_inst->src[2] = ZERO_SRC_1; +} -#define ZERO_SRC_1 (MAKE_VSF_SOURCE(t_src_index(vp, &src[1]), \ - SWIZZLE_ZERO, SWIZZLE_ZERO, \ - SWIZZLE_ZERO, SWIZZLE_ZERO, \ - t_src_class(src[1].File), VSF_FLAG_NONE) | (src[1].RelAddr << 4)) +static void t_opcode_dst(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) +{ + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_DST, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + + o_inst->src[0] = t_src(vp, &src[0]); + o_inst->src[1] = t_src(vp, &src[1]); + o_inst->src[2] = ZERO_SRC_1; +} -#define ZERO_SRC_2 (MAKE_VSF_SOURCE(t_src_index(vp, &src[2]), \ - SWIZZLE_ZERO, SWIZZLE_ZERO, \ - SWIZZLE_ZERO, SWIZZLE_ZERO, \ - t_src_class(src[2].File), VSF_FLAG_NONE) | (src[2].RelAddr << 4)) +static void t_opcode_ex2(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) +{ + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_EX2, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + + o_inst->src[0] = t_src_scalar(vp, &src[0]); + o_inst->src[1] = ZERO_SRC_0; + o_inst->src[2] = ZERO_SRC_0; +} -#define ONE_SRC_0 (MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), \ - SWIZZLE_ONE, SWIZZLE_ONE, \ - SWIZZLE_ONE, SWIZZLE_ONE, \ - t_src_class(src[0].File), VSF_FLAG_NONE) | (src[0].RelAddr << 4)) +static void t_opcode_exp(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) +{ + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_EXP, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + + o_inst->src[0] = t_src_scalar(vp, &src[0]); + o_inst->src[1] = ZERO_SRC_0; + o_inst->src[2] = ZERO_SRC_0; +} -#define ONE_SRC_1 (MAKE_VSF_SOURCE(t_src_index(vp, &src[1]), \ - SWIZZLE_ONE, SWIZZLE_ONE, \ - SWIZZLE_ONE, SWIZZLE_ONE, \ - t_src_class(src[1].File), VSF_FLAG_NONE) | (src[1].RelAddr << 4)) +static struct r300_vertprog_instruction *t_opcode_flr(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3], + int *u_temp_i) +{ + /* FRC TMP 0.X Y Z W PARAM 0{} {X Y Z W} + ADD RESULT 1.X Y Z W PARAM 0{} {X Y Z W} TMP 0{X Y Z W } {X Y Z W} neg Xneg Yneg Zneg W */ + + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_FRC, *u_temp_i, + t_dst_mask(vpi->DstReg.WriteMask), + VSF_OUT_CLASS_TMP); + + o_inst->src[0] = t_src(vp, &src[0]); + o_inst->src[1] = ZERO_SRC_0; + o_inst->src[2] = ZERO_SRC_0; + o_inst++; + + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_ADD, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + + o_inst->src[0] = t_src(vp, &src[0]); + o_inst->src[1] = + MAKE_VSF_SOURCE(*u_temp_i, VSF_IN_COMPONENT_X, + VSF_IN_COMPONENT_Y, VSF_IN_COMPONENT_Z, + VSF_IN_COMPONENT_W, VSF_IN_CLASS_TMP, + /* Not 100% sure about this */ + (!src[0]. + NegateBase) ? VSF_FLAG_ALL : VSF_FLAG_NONE + /*VSF_FLAG_ALL */ ); + + o_inst->src[2] = ZERO_SRC_0; + (*u_temp_i)--; + + return o_inst; +} -#define ONE_SRC_2 (MAKE_VSF_SOURCE(t_src_index(vp, &src[2]), \ - SWIZZLE_ONE, SWIZZLE_ONE, \ - SWIZZLE_ONE, SWIZZLE_ONE, \ - t_src_class(src[2].File), VSF_FLAG_NONE) | (src[2].RelAddr << 4)) +static void t_opcode_frc(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) +{ + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_FRC, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + + o_inst->src[0] = t_src(vp, &src[0]); + o_inst->src[1] = ZERO_SRC_0; + o_inst->src[2] = ZERO_SRC_0; +} -/* DP4 version seems to trigger some hw peculiarity */ -//#define PREFER_DP4 +static void t_opcode_lg2(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) +{ + // LG2 RESULT 1.X Y Z W PARAM 0{} {X X X X} + + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_LG2, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + + o_inst->src[0] = + MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), + t_swizzle(GET_SWZ(src[0].Swizzle, 0)), + t_swizzle(GET_SWZ(src[0].Swizzle, 0)), + t_swizzle(GET_SWZ(src[0].Swizzle, 0)), + t_swizzle(GET_SWZ(src[0].Swizzle, 0)), + t_src_class(src[0].File), + src[0]. + NegateBase ? VSF_FLAG_ALL : VSF_FLAG_NONE) | + (src[0].RelAddr << 4); + o_inst->src[1] = ZERO_SRC_0; + o_inst->src[2] = ZERO_SRC_0; +} -#define FREE_TEMPS() \ - do { \ - if(u_temp_i < vp->num_temporaries) { \ - WARN_ONCE("Ran out of temps, num temps %d, us %d\n", vp->num_temporaries, u_temp_i); \ - vp->native = GL_FALSE; \ - } \ - u_temp_i=VSF_MAX_FRAGMENT_TEMPS-1; \ - } while (0) +static void t_opcode_lit(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) +{ + //LIT TMP 1.Y Z TMP 1{} {X W Z Y} TMP 1{} {Y W Z X} TMP 1{} {Y X Z W} + + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_LIT, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + /* NOTE: Users swizzling might not work. */ + o_inst->src[0] = MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), t_swizzle(GET_SWZ(src[0].Swizzle, 0)), // x + t_swizzle(GET_SWZ(src[0].Swizzle, 3)), // w + VSF_IN_COMPONENT_ZERO, // z + t_swizzle(GET_SWZ(src[0].Swizzle, 1)), // y + t_src_class(src[0].File), + src[0]. + NegateBase ? VSF_FLAG_ALL : + VSF_FLAG_NONE) | (src[0]. + RelAddr << 4); + o_inst->src[1] = MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), t_swizzle(GET_SWZ(src[0].Swizzle, 1)), // y + t_swizzle(GET_SWZ(src[0].Swizzle, 3)), // w + VSF_IN_COMPONENT_ZERO, // z + t_swizzle(GET_SWZ(src[0].Swizzle, 0)), // x + t_src_class(src[0].File), + src[0]. + NegateBase ? VSF_FLAG_ALL : + VSF_FLAG_NONE) | (src[0]. + RelAddr << 4); + o_inst->src[2] = MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), t_swizzle(GET_SWZ(src[0].Swizzle, 1)), // y + t_swizzle(GET_SWZ(src[0].Swizzle, 0)), // x + VSF_IN_COMPONENT_ZERO, // z + t_swizzle(GET_SWZ(src[0].Swizzle, 3)), // w + t_src_class(src[0].File), + src[0]. + NegateBase ? VSF_FLAG_ALL : + VSF_FLAG_NONE) | (src[0]. + RelAddr << 4); +} -static void r300TranslateVertexShader(struct r300_vertex_program *vp, - struct prog_instruction *vpi) +static void t_opcode_log(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) +{ + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_LOG, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + + o_inst->src[0] = t_src_scalar(vp, &src[0]); + o_inst->src[1] = ZERO_SRC_0; + o_inst->src[2] = ZERO_SRC_0; +} + +static void t_opcode_mad(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) { - int i, cur_reg = 0; - VERTEX_SHADER_INSTRUCTION *o_inst; - unsigned long operands; - int are_srcs_scalar; unsigned long hw_op; - /* Initial value should be last tmp reg that hw supports. - Strangely enough r300 doesnt mind even though these would be out of range. - Smart enough to realize that it doesnt need it? */ - int u_temp_i = VSF_MAX_FRAGMENT_TEMPS - 1; - struct prog_src_register src[3]; - vp->pos_end = 0; /* Not supported yet */ - vp->program.length = 0; - /*vp->num_temporaries=mesa_vp->Base.NumTemporaries; */ + hw_op = (src[0].File == PROGRAM_TEMPORARY + && src[1].File == PROGRAM_TEMPORARY + && src[2].File == + PROGRAM_TEMPORARY) ? R300_VPI_OUT_OP_MAD_2 : + R300_VPI_OUT_OP_MAD; + + o_inst->opcode = + MAKE_VSF_OP(hw_op, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + o_inst->src[0] = t_src(vp, &src[0]); + o_inst->src[1] = t_src(vp, &src[1]); + o_inst->src[2] = t_src(vp, &src[2]); +} + +static void t_opcode_max(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) +{ + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_MAX, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + + o_inst->src[0] = t_src(vp, &src[0]); + o_inst->src[1] = t_src(vp, &src[1]); + o_inst->src[2] = ZERO_SRC_1; +} + +static void t_opcode_min(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) +{ + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_MIN, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + + o_inst->src[0] = t_src(vp, &src[0]); + o_inst->src[1] = t_src(vp, &src[1]); + o_inst->src[2] = ZERO_SRC_1; +} + +static void t_opcode_mov(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) +{ + //ADD RESULT 1.X Y Z W PARAM 0{} {X Y Z W} PARAM 0{} {ZERO ZERO ZERO ZERO} + +#if 1 + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_ADD, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + o_inst->src[0] = t_src(vp, &src[0]); + o_inst->src[1] = ZERO_SRC_0; + o_inst->src[2] = ZERO_SRC_0; +#else + hw_op = + (src[0].File == + PROGRAM_TEMPORARY) ? R300_VPI_OUT_OP_MAD_2 : + R300_VPI_OUT_OP_MAD; + + o_inst->opcode = + MAKE_VSF_OP(hw_op, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + o_inst->src[0] = t_src(vp, &src[0]); + o_inst->src[1] = ONE_SRC_0; + o_inst->src[2] = ZERO_SRC_0; +#endif +} + +static void t_opcode_mul(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) +{ + unsigned long hw_op; + + // HW mul can take third arg but appears to have some other limitations. + + hw_op = (src[0].File == PROGRAM_TEMPORARY + && src[1].File == + PROGRAM_TEMPORARY) ? R300_VPI_OUT_OP_MAD_2 : + R300_VPI_OUT_OP_MAD; + + o_inst->opcode = + MAKE_VSF_OP(hw_op, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + o_inst->src[0] = t_src(vp, &src[0]); + o_inst->src[1] = t_src(vp, &src[1]); + + o_inst->src[2] = ZERO_SRC_1; +} + +static void t_opcode_pow(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) +{ + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_POW, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + o_inst->src[0] = t_src_scalar(vp, &src[0]); + o_inst->src[1] = ZERO_SRC_0; + o_inst->src[2] = t_src_scalar(vp, &src[1]); +} + +static void t_opcode_rcp(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) +{ + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_RCP, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + + o_inst->src[0] = t_src_scalar(vp, &src[0]); + o_inst->src[1] = ZERO_SRC_0; + o_inst->src[2] = ZERO_SRC_0; +} + +static void t_opcode_rsq(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) +{ + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_RSQ, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + + o_inst->src[0] = t_src_scalar(vp, &src[0]); + o_inst->src[1] = ZERO_SRC_0; + o_inst->src[2] = ZERO_SRC_0; +} + +static void t_opcode_sge(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) +{ + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_SGE, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + + o_inst->src[0] = t_src(vp, &src[0]); + o_inst->src[1] = t_src(vp, &src[1]); + o_inst->src[2] = ZERO_SRC_1; +} + +static void t_opcode_slt(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) +{ + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_SLT, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + + o_inst->src[0] = t_src(vp, &src[0]); + o_inst->src[1] = t_src(vp, &src[1]); + o_inst->src[2] = ZERO_SRC_1; +} + +static void t_opcode_sub(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) +{ + unsigned long hw_op; + + //ADD RESULT 1.X Y Z W TMP 0{} {X Y Z W} PARAM 1{X Y Z W } {X Y Z W} neg Xneg Yneg Zneg W + +#if 1 + hw_op = (src[0].File == PROGRAM_TEMPORARY + && src[1].File == + PROGRAM_TEMPORARY) ? R300_VPI_OUT_OP_MAD_2 : + R300_VPI_OUT_OP_MAD; + + o_inst->opcode = + MAKE_VSF_OP(hw_op, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + o_inst->src[0] = t_src(vp, &src[0]); + o_inst->src[1] = ONE_SRC_0; + o_inst->src[2] = + MAKE_VSF_SOURCE(t_src_index(vp, &src[1]), + t_swizzle(GET_SWZ(src[1].Swizzle, 0)), + t_swizzle(GET_SWZ(src[1].Swizzle, 1)), + t_swizzle(GET_SWZ(src[1].Swizzle, 2)), + t_swizzle(GET_SWZ(src[1].Swizzle, 3)), + t_src_class(src[1].File), + (!src[1]. + NegateBase) ? VSF_FLAG_ALL : VSF_FLAG_NONE) | + (src[1].RelAddr << 4); +#else + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_ADD, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + + o_inst->src[0] = t_src(vp, &src[0]); + o_inst->src[1] = + MAKE_VSF_SOURCE(t_src_index(vp, &src[1]), + t_swizzle(GET_SWZ(src[1].Swizzle, 0)), + t_swizzle(GET_SWZ(src[1].Swizzle, 1)), + t_swizzle(GET_SWZ(src[1].Swizzle, 2)), + t_swizzle(GET_SWZ(src[1].Swizzle, 3)), + t_src_class(src[1].File), + (!src[1]. + NegateBase) ? VSF_FLAG_ALL : VSF_FLAG_NONE) | + (src[1].RelAddr << 4); + o_inst->src[2] = 0; +#endif +} + +static void t_opcode_swz(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3]) +{ + //ADD RESULT 1.X Y Z W PARAM 0{} {X Y Z W} PARAM 0{} {ZERO ZERO ZERO ZERO} + +#if 1 + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_ADD, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + o_inst->src[0] = t_src(vp, &src[0]); + o_inst->src[1] = ZERO_SRC_0; + o_inst->src[2] = ZERO_SRC_0; +#else + hw_op = + (src[0].File == + PROGRAM_TEMPORARY) ? R300_VPI_OUT_OP_MAD_2 : + R300_VPI_OUT_OP_MAD; + + o_inst->opcode = + MAKE_VSF_OP(hw_op, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + o_inst->src[0] = t_src(vp, &src[0]); + o_inst->src[1] = ONE_SRC_0; + o_inst->src[2] = ZERO_SRC_0; +#endif +} + +static struct r300_vertprog_instruction *t_opcode_xpd(struct r300_vertex_program *vp, + struct prog_instruction *vpi, + struct r300_vertprog_instruction *o_inst, + struct prog_src_register src[3], + int *u_temp_i) +{ + /* mul r0, r1.yzxw, r2.zxyw + mad r0, -r2.yzxw, r1.zxyw, r0 + NOTE: might need MAD_2 + */ + + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_MAD, *u_temp_i, + t_dst_mask(vpi->DstReg.WriteMask), + VSF_OUT_CLASS_TMP); + + o_inst->src[0] = MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), t_swizzle(GET_SWZ(src[0].Swizzle, 1)), // y + t_swizzle(GET_SWZ(src[0].Swizzle, 2)), // z + t_swizzle(GET_SWZ(src[0].Swizzle, 0)), // x + t_swizzle(GET_SWZ(src[0].Swizzle, 3)), // w + t_src_class(src[0].File), + src[0]. + NegateBase ? VSF_FLAG_ALL : + VSF_FLAG_NONE) | (src[0]. + RelAddr << 4); + + o_inst->src[1] = MAKE_VSF_SOURCE(t_src_index(vp, &src[1]), t_swizzle(GET_SWZ(src[1].Swizzle, 2)), // z + t_swizzle(GET_SWZ(src[1].Swizzle, 0)), // x + t_swizzle(GET_SWZ(src[1].Swizzle, 1)), // y + t_swizzle(GET_SWZ(src[1].Swizzle, 3)), // w + t_src_class(src[1].File), + src[1]. + NegateBase ? VSF_FLAG_ALL : + VSF_FLAG_NONE) | (src[1]. + RelAddr << 4); + + o_inst->src[2] = ZERO_SRC_1; + o_inst++; + (*u_temp_i)--; + + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_MAD, t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + + o_inst->src[0] = MAKE_VSF_SOURCE(t_src_index(vp, &src[1]), t_swizzle(GET_SWZ(src[1].Swizzle, 1)), // y + t_swizzle(GET_SWZ(src[1].Swizzle, 2)), // z + t_swizzle(GET_SWZ(src[1].Swizzle, 0)), // x + t_swizzle(GET_SWZ(src[1].Swizzle, 3)), // w + t_src_class(src[1].File), + (!src[1]. + NegateBase) ? VSF_FLAG_ALL : + VSF_FLAG_NONE) | (src[1]. + RelAddr << 4); + + o_inst->src[1] = MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), t_swizzle(GET_SWZ(src[0].Swizzle, 2)), // z + t_swizzle(GET_SWZ(src[0].Swizzle, 0)), // x + t_swizzle(GET_SWZ(src[0].Swizzle, 1)), // y + t_swizzle(GET_SWZ(src[0].Swizzle, 3)), // w + t_src_class(src[0].File), + src[0]. + NegateBase ? VSF_FLAG_ALL : + VSF_FLAG_NONE) | (src[0]. + RelAddr << 4); + + o_inst->src[2] = + MAKE_VSF_SOURCE(*u_temp_i + 1, VSF_IN_COMPONENT_X, + VSF_IN_COMPONENT_Y, VSF_IN_COMPONENT_Z, + VSF_IN_COMPONENT_W, VSF_IN_CLASS_TMP, + VSF_FLAG_NONE); + + return o_inst; +} + +static void t_inputs_outputs(struct r300_vertex_program *vp) +{ + int i; + int cur_reg = 0; for (i = 0; i < VERT_ATTRIB_MAX; i++) vp->inputs[i] = -1; @@ -437,39 +1046,71 @@ static void r300TranslateVertexShader(struct r300_vertex_program *vp, assert(vp->key.OutputsWritten & (1 << VERT_RESULT_HPOS)); - /* Assign outputs */ - if (vp->key.OutputsWritten & (1 << VERT_RESULT_HPOS)) + if (vp->key.OutputsWritten & (1 << VERT_RESULT_HPOS)) { vp->outputs[VERT_RESULT_HPOS] = cur_reg++; + } - if (vp->key.OutputsWritten & (1 << VERT_RESULT_PSIZ)) + if (vp->key.OutputsWritten & (1 << VERT_RESULT_PSIZ)) { vp->outputs[VERT_RESULT_PSIZ] = cur_reg++; + } - if (vp->key.OutputsWritten & (1 << VERT_RESULT_COL0)) + if (vp->key.OutputsWritten & (1 << VERT_RESULT_COL0)) { vp->outputs[VERT_RESULT_COL0] = cur_reg++; + } - if (vp->key.OutputsWritten & (1 << VERT_RESULT_COL1)) - vp->outputs[VERT_RESULT_COL1] = cur_reg++; - -#if 0 /* Not supported yet */ - if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC0)) - vp->outputs[VERT_RESULT_BFC0] = cur_reg++; + if (vp->key.OutputsWritten & (1 << VERT_RESULT_COL1)) { + vp->outputs[VERT_RESULT_COL1] = + vp->outputs[VERT_RESULT_COL0] + 1; + cur_reg = vp->outputs[VERT_RESULT_COL1] + 1; + } - if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC1)) - vp->outputs[VERT_RESULT_BFC1] = cur_reg++; + if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC0)) { + vp->outputs[VERT_RESULT_BFC0] = + vp->outputs[VERT_RESULT_COL0] + 2; + cur_reg = vp->outputs[VERT_RESULT_BFC0] + 1; + } - if (vp->key.OutputsWritten & (1 << VERT_RESULT_FOGC)) + if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC1)) { + vp->outputs[VERT_RESULT_BFC1] = + vp->outputs[VERT_RESULT_COL0] + 3; + cur_reg = vp->outputs[VERT_RESULT_BFC1] + 1; + } +#if 0 + if (vp->key.OutputsWritten & (1 << VERT_RESULT_FOGC)) { vp->outputs[VERT_RESULT_FOGC] = cur_reg++; + } #endif - for (i = VERT_RESULT_TEX0; i <= VERT_RESULT_TEX7; i++) - if (vp->key.OutputsWritten & (1 << i)) + for (i = VERT_RESULT_TEX0; i <= VERT_RESULT_TEX7; i++) { + if (vp->key.OutputsWritten & (1 << i)) { vp->outputs[i] = cur_reg++; + } + } +} + +static void r300TranslateVertexShader(struct r300_vertex_program *vp, + struct prog_instruction *vpi) +{ + int i; + struct r300_vertprog_instruction *o_inst; + unsigned long num_operands; + /* Initial value should be last tmp reg that hw supports. + Strangely enough r300 doesnt mind even though these would be out of range. + Smart enough to realize that it doesnt need it? */ + int u_temp_i = VSF_MAX_FRAGMENT_TEMPS - 1; + struct prog_src_register src[3]; + vp->pos_end = 0; /* Not supported yet */ + vp->program.length = 0; + /*vp->num_temporaries=mesa_vp->Base.NumTemporaries; */ vp->translated = GL_TRUE; vp->native = GL_TRUE; - o_inst = vp->program.body.i; - for (; vpi->Opcode != OPCODE_END; vpi++, o_inst++) { + t_inputs_outputs(vp); + + for (o_inst = vp->program.body.i; vpi->Opcode != OPCODE_END; + vpi++, o_inst++) { + FREE_TEMPS(); if (!valid_dst(vp, &vpi->DstReg)) { @@ -478,29 +1119,30 @@ static void r300TranslateVertexShader(struct r300_vertex_program *vp, vpi->DstReg.Index = u_temp_i; } - operands = op_operands(vpi->Opcode); - are_srcs_scalar = operands & SCALAR_FLAG; - operands &= OP_MASK; + num_operands = _mesa_num_inst_src_regs(vpi->Opcode); - for (i = 0; i < operands; i++) + /* copy the sources (src) from mesa into a local variable... is this needed? */ + for (i = 0; i < num_operands; i++) { src[i] = vpi->SrcReg[i]; + } - if (operands == 3) { /* TODO: scalars */ + if (num_operands == 3) { /* TODO: scalars */ if (CMP_SRCS(src[1], src[2]) || CMP_SRCS(src[0], src[2])) { - o_inst->op = - MAKE_VSF_OP(R300_VPI_OUT_OP_ADD, u_temp_i, - VSF_FLAG_ALL, + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_ADD, + u_temp_i, VSF_FLAG_ALL, VSF_OUT_CLASS_TMP); o_inst->src[0] = - MAKE_VSF_SOURCE(t_src_index(vp, &src[2]), + MAKE_VSF_SOURCE(t_src_index + (vp, &src[2]), SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_W, - t_src_class(src[2].File), - VSF_FLAG_NONE) | (src[2]. - RelAddr << - 4); + t_src_class(src[2]. + File), + VSF_FLAG_NONE) | + (src[2].RelAddr << 4); o_inst->src[1] = ZERO_SRC_2; o_inst->src[2] = ZERO_SRC_2; @@ -511,24 +1153,24 @@ static void r300TranslateVertexShader(struct r300_vertex_program *vp, src[2].RelAddr = 0; u_temp_i--; } - } - if (operands >= 2) { + if (num_operands >= 2) { if (CMP_SRCS(src[1], src[0])) { - o_inst->op = - MAKE_VSF_OP(R300_VPI_OUT_OP_ADD, u_temp_i, - VSF_FLAG_ALL, + o_inst->opcode = + MAKE_VSF_OP(R300_VPI_OUT_OP_ADD, + u_temp_i, VSF_FLAG_ALL, VSF_OUT_CLASS_TMP); o_inst->src[0] = - MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), + MAKE_VSF_SOURCE(t_src_index + (vp, &src[0]), SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_W, - t_src_class(src[0].File), - VSF_FLAG_NONE) | (src[0]. - RelAddr << - 4); + t_src_class(src[0]. + File), + VSF_FLAG_NONE) | + (src[0].RelAddr << 4); o_inst->src[1] = ZERO_SRC_0; o_inst->src[2] = ZERO_SRC_0; @@ -541,517 +1183,101 @@ static void r300TranslateVertexShader(struct r300_vertex_program *vp, } } - /* These ops need special handling. */ switch (vpi->Opcode) { - case OPCODE_POW: - o_inst->op = - MAKE_VSF_OP(R300_VPI_OUT_OP_POW, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - o_inst->src[0] = t_src_scalar(vp, &src[0]); - o_inst->src[1] = ZERO_SRC_0; - o_inst->src[2] = t_src_scalar(vp, &src[1]); - goto next; - - case OPCODE_MOV: //ADD RESULT 1.X Y Z W PARAM 0{} {X Y Z W} PARAM 0{} {ZERO ZERO ZERO ZERO} - case OPCODE_SWZ: -#if 1 - o_inst->op = - MAKE_VSF_OP(R300_VPI_OUT_OP_ADD, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - o_inst->src[0] = t_src(vp, &src[0]); - o_inst->src[1] = ZERO_SRC_0; - o_inst->src[2] = ZERO_SRC_0; -#else - hw_op = - (src[0].File == - PROGRAM_TEMPORARY) ? R300_VPI_OUT_OP_MAD_2 : - R300_VPI_OUT_OP_MAD; - - o_inst->op = - MAKE_VSF_OP(hw_op, t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - o_inst->src[0] = t_src(vp, &src[0]); - o_inst->src[1] = ONE_SRC_0; - o_inst->src[2] = ZERO_SRC_0; -#endif - - goto next; - + case OPCODE_ABS: + t_opcode_abs(vp, vpi, o_inst, src); + break; case OPCODE_ADD: -#if 1 - hw_op = (src[0].File == PROGRAM_TEMPORARY && - src[1].File == - PROGRAM_TEMPORARY) ? R300_VPI_OUT_OP_MAD_2 : - R300_VPI_OUT_OP_MAD; - - o_inst->op = - MAKE_VSF_OP(hw_op, t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - o_inst->src[0] = ONE_SRC_0; - o_inst->src[1] = t_src(vp, &src[0]); - o_inst->src[2] = t_src(vp, &src[1]); -#else - o_inst->op = - MAKE_VSF_OP(R300_VPI_OUT_OP_ADD, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - o_inst->src[0] = t_src(vp, &src[0]); - o_inst->src[1] = t_src(vp, &src[1]); - o_inst->src[2] = ZERO_SRC_1; - -#endif - goto next; - - case OPCODE_MAD: - hw_op = (src[0].File == PROGRAM_TEMPORARY && - src[1].File == PROGRAM_TEMPORARY && - src[2].File == - PROGRAM_TEMPORARY) ? R300_VPI_OUT_OP_MAD_2 : - R300_VPI_OUT_OP_MAD; - - o_inst->op = - MAKE_VSF_OP(hw_op, t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - o_inst->src[0] = t_src(vp, &src[0]); - o_inst->src[1] = t_src(vp, &src[1]); - o_inst->src[2] = t_src(vp, &src[2]); - goto next; - - case OPCODE_MUL: /* HW mul can take third arg but appears to have some other limitations. */ - hw_op = (src[0].File == PROGRAM_TEMPORARY && - src[1].File == - PROGRAM_TEMPORARY) ? R300_VPI_OUT_OP_MAD_2 : - R300_VPI_OUT_OP_MAD; - - o_inst->op = - MAKE_VSF_OP(hw_op, t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - o_inst->src[0] = t_src(vp, &src[0]); - o_inst->src[1] = t_src(vp, &src[1]); - - o_inst->src[2] = ZERO_SRC_1; - goto next; - - case OPCODE_DP3: //DOT RESULT 1.X Y Z W PARAM 0{} {X Y Z ZERO} PARAM 0{} {X Y Z ZERO} - o_inst->op = - MAKE_VSF_OP(R300_VPI_OUT_OP_DOT, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - - o_inst->src[0] = - MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), - t_swizzle(GET_SWZ - (src[0].Swizzle, 0)), - t_swizzle(GET_SWZ - (src[0].Swizzle, 1)), - t_swizzle(GET_SWZ - (src[0].Swizzle, 2)), - SWIZZLE_ZERO, - t_src_class(src[0].File), - src[0]. - NegateBase ? VSF_FLAG_XYZ : - VSF_FLAG_NONE) | (src[0]. - RelAddr << 4); - - o_inst->src[1] = - MAKE_VSF_SOURCE(t_src_index(vp, &src[1]), - t_swizzle(GET_SWZ - (src[1].Swizzle, 0)), - t_swizzle(GET_SWZ - (src[1].Swizzle, 1)), - t_swizzle(GET_SWZ - (src[1].Swizzle, 2)), - SWIZZLE_ZERO, - t_src_class(src[1].File), - src[1]. - NegateBase ? VSF_FLAG_XYZ : - VSF_FLAG_NONE) | (src[1]. - RelAddr << 4); - - o_inst->src[2] = ZERO_SRC_1; - goto next; - - case OPCODE_SUB: //ADD RESULT 1.X Y Z W TMP 0{} {X Y Z W} PARAM 1{X Y Z W } {X Y Z W} neg Xneg Yneg Zneg W -#if 1 - hw_op = (src[0].File == PROGRAM_TEMPORARY && - src[1].File == - PROGRAM_TEMPORARY) ? R300_VPI_OUT_OP_MAD_2 : - R300_VPI_OUT_OP_MAD; - - o_inst->op = - MAKE_VSF_OP(hw_op, t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - o_inst->src[0] = t_src(vp, &src[0]); - o_inst->src[1] = ONE_SRC_0; - o_inst->src[2] = - MAKE_VSF_SOURCE(t_src_index(vp, &src[1]), - t_swizzle(GET_SWZ - (src[1].Swizzle, 0)), - t_swizzle(GET_SWZ - (src[1].Swizzle, 1)), - t_swizzle(GET_SWZ - (src[1].Swizzle, 2)), - t_swizzle(GET_SWZ - (src[1].Swizzle, 3)), - t_src_class(src[1].File), - (!src[1]. - NegateBase) ? VSF_FLAG_ALL : - VSF_FLAG_NONE) | (src[1]. - RelAddr << 4); -#else - o_inst->op = - MAKE_VSF_OP(R300_VPI_OUT_OP_ADD, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - - o_inst->src[0] = t_src(vp, &src[0]); - o_inst->src[1] = - MAKE_VSF_SOURCE(t_src_index(vp, &src[1]), - t_swizzle(GET_SWZ - (src[1].Swizzle, 0)), - t_swizzle(GET_SWZ - (src[1].Swizzle, 1)), - t_swizzle(GET_SWZ - (src[1].Swizzle, 2)), - t_swizzle(GET_SWZ - (src[1].Swizzle, 3)), - t_src_class(src[1].File), - (!src[1]. - NegateBase) ? VSF_FLAG_ALL : - VSF_FLAG_NONE) | (src[1]. - RelAddr << 4); - o_inst->src[2] = 0; -#endif - goto next; - - case OPCODE_ABS: //MAX RESULT 1.X Y Z W PARAM 0{} {X Y Z W} PARAM 0{X Y Z W } {X Y Z W} neg Xneg Yneg Zneg W - o_inst->op = - MAKE_VSF_OP(R300_VPI_OUT_OP_MAX, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - - o_inst->src[0] = t_src(vp, &src[0]); - o_inst->src[1] = - MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), - t_swizzle(GET_SWZ - (src[0].Swizzle, 0)), - t_swizzle(GET_SWZ - (src[0].Swizzle, 1)), - t_swizzle(GET_SWZ - (src[0].Swizzle, 2)), - t_swizzle(GET_SWZ - (src[0].Swizzle, 3)), - t_src_class(src[0].File), - (!src[0]. - NegateBase) ? VSF_FLAG_ALL : - VSF_FLAG_NONE) | (src[0]. - RelAddr << 4); - o_inst->src[2] = 0; - goto next; - + t_opcode_add(vp, vpi, o_inst, src); + break; + case OPCODE_ARL: + t_opcode_arl(vp, vpi, o_inst, src); + break; + case OPCODE_DP3: + t_opcode_dp3(vp, vpi, o_inst, src); + break; + case OPCODE_DP4: + t_opcode_dp4(vp, vpi, o_inst, src); + break; + case OPCODE_DPH: + t_opcode_dph(vp, vpi, o_inst, src); + break; + case OPCODE_DST: + t_opcode_dst(vp, vpi, o_inst, src); + break; + case OPCODE_EX2: + t_opcode_ex2(vp, vpi, o_inst, src); + break; + case OPCODE_EXP: + t_opcode_exp(vp, vpi, o_inst, src); + break; case OPCODE_FLR: - /* FRC TMP 0.X Y Z W PARAM 0{} {X Y Z W} - ADD RESULT 1.X Y Z W PARAM 0{} {X Y Z W} TMP 0{X Y Z W } {X Y Z W} neg Xneg Yneg Zneg W */ - - o_inst->op = MAKE_VSF_OP(R300_VPI_OUT_OP_FRC, u_temp_i, - t_dst_mask(vpi->DstReg. - WriteMask), - VSF_OUT_CLASS_TMP); - - o_inst->src[0] = t_src(vp, &src[0]); - o_inst->src[1] = ZERO_SRC_0; - o_inst->src[2] = ZERO_SRC_0; - o_inst++; - - o_inst->op = - MAKE_VSF_OP(R300_VPI_OUT_OP_ADD, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - - o_inst->src[0] = t_src(vp, &src[0]); - o_inst->src[1] = MAKE_VSF_SOURCE(u_temp_i, - VSF_IN_COMPONENT_X, - VSF_IN_COMPONENT_Y, - VSF_IN_COMPONENT_Z, - VSF_IN_COMPONENT_W, - VSF_IN_CLASS_TMP, - /* Not 100% sure about this */ - (!src[0]. - NegateBase) ? - VSF_FLAG_ALL : - VSF_FLAG_NONE - /*VSF_FLAG_ALL */ ); - - o_inst->src[2] = ZERO_SRC_0; - u_temp_i--; - goto next; - - case OPCODE_LG2: // LG2 RESULT 1.X Y Z W PARAM 0{} {X X X X} - o_inst->op = - MAKE_VSF_OP(R300_VPI_OUT_OP_LG2, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - - o_inst->src[0] = - MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), - t_swizzle(GET_SWZ - (src[0].Swizzle, 0)), - t_swizzle(GET_SWZ - (src[0].Swizzle, 0)), - t_swizzle(GET_SWZ - (src[0].Swizzle, 0)), - t_swizzle(GET_SWZ - (src[0].Swizzle, 0)), - t_src_class(src[0].File), - src[0]. - NegateBase ? VSF_FLAG_ALL : - VSF_FLAG_NONE) | (src[0]. - RelAddr << 4); - o_inst->src[1] = ZERO_SRC_0; - o_inst->src[2] = ZERO_SRC_0; - goto next; - - case OPCODE_LIT: //LIT TMP 1.Y Z TMP 1{} {X W Z Y} TMP 1{} {Y W Z X} TMP 1{} {Y X Z W} - o_inst->op = - MAKE_VSF_OP(R300_VPI_OUT_OP_LIT, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - /* NOTE: Users swizzling might not work. */ - o_inst->src[0] = MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), t_swizzle(GET_SWZ(src[0].Swizzle, 0)), // x - t_swizzle(GET_SWZ(src[0].Swizzle, 3)), // w - VSF_IN_COMPONENT_ZERO, // z - t_swizzle(GET_SWZ(src[0].Swizzle, 1)), // y - t_src_class(src[0]. - File), - src[0]. - NegateBase ? - VSF_FLAG_ALL : - VSF_FLAG_NONE) | - (src[0].RelAddr << 4); - o_inst->src[1] = MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), t_swizzle(GET_SWZ(src[0].Swizzle, 1)), // y - t_swizzle(GET_SWZ(src[0].Swizzle, 3)), // w - VSF_IN_COMPONENT_ZERO, // z - t_swizzle(GET_SWZ(src[0].Swizzle, 0)), // x - t_src_class(src[0]. - File), - src[0]. - NegateBase ? - VSF_FLAG_ALL : - VSF_FLAG_NONE) | - (src[0].RelAddr << 4); - o_inst->src[2] = MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), t_swizzle(GET_SWZ(src[0].Swizzle, 1)), // y - t_swizzle(GET_SWZ(src[0].Swizzle, 0)), // x - VSF_IN_COMPONENT_ZERO, // z - t_swizzle(GET_SWZ(src[0].Swizzle, 3)), // w - t_src_class(src[0]. - File), - src[0]. - NegateBase ? - VSF_FLAG_ALL : - VSF_FLAG_NONE) | - (src[0].RelAddr << 4); - goto next; - - case OPCODE_DPH: //DOT RESULT 1.X Y Z W PARAM 0{} {X Y Z ONE} PARAM 0{} {X Y Z W} - o_inst->op = - MAKE_VSF_OP(R300_VPI_OUT_OP_DOT, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - - o_inst->src[0] = - MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), - t_swizzle(GET_SWZ - (src[0].Swizzle, 0)), - t_swizzle(GET_SWZ - (src[0].Swizzle, 1)), - t_swizzle(GET_SWZ - (src[0].Swizzle, 2)), - VSF_IN_COMPONENT_ONE, - t_src_class(src[0].File), - src[0]. - NegateBase ? VSF_FLAG_XYZ : - VSF_FLAG_NONE) | (src[0]. - RelAddr << 4); - o_inst->src[1] = t_src(vp, &src[1]); - o_inst->src[2] = ZERO_SRC_1; - goto next; - - case OPCODE_XPD: - /* mul r0, r1.yzxw, r2.zxyw - mad r0, -r2.yzxw, r1.zxyw, r0 - NOTE: might need MAD_2 - */ - - o_inst->op = MAKE_VSF_OP(R300_VPI_OUT_OP_MAD, u_temp_i, - t_dst_mask(vpi->DstReg. - WriteMask), - VSF_OUT_CLASS_TMP); - - o_inst->src[0] = MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), t_swizzle(GET_SWZ(src[0].Swizzle, 1)), // y - t_swizzle(GET_SWZ(src[0].Swizzle, 2)), // z - t_swizzle(GET_SWZ(src[0].Swizzle, 0)), // x - t_swizzle(GET_SWZ(src[0].Swizzle, 3)), // w - t_src_class(src[0]. - File), - src[0]. - NegateBase ? - VSF_FLAG_ALL : - VSF_FLAG_NONE) | - (src[0].RelAddr << 4); - - o_inst->src[1] = MAKE_VSF_SOURCE(t_src_index(vp, &src[1]), t_swizzle(GET_SWZ(src[1].Swizzle, 2)), // z - t_swizzle(GET_SWZ(src[1].Swizzle, 0)), // x - t_swizzle(GET_SWZ(src[1].Swizzle, 1)), // y - t_swizzle(GET_SWZ(src[1].Swizzle, 3)), // w - t_src_class(src[1]. - File), - src[1]. - NegateBase ? - VSF_FLAG_ALL : - VSF_FLAG_NONE) | - (src[1].RelAddr << 4); - - o_inst->src[2] = ZERO_SRC_1; - o_inst++; - u_temp_i--; - - o_inst->op = - MAKE_VSF_OP(R300_VPI_OUT_OP_MAD, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - - o_inst->src[0] = MAKE_VSF_SOURCE(t_src_index(vp, &src[1]), t_swizzle(GET_SWZ(src[1].Swizzle, 1)), // y - t_swizzle(GET_SWZ(src[1].Swizzle, 2)), // z - t_swizzle(GET_SWZ(src[1].Swizzle, 0)), // x - t_swizzle(GET_SWZ(src[1].Swizzle, 3)), // w - t_src_class(src[1]. - File), - (!src[1]. - NegateBase) ? - VSF_FLAG_ALL : - VSF_FLAG_NONE) | - (src[1].RelAddr << 4); - - o_inst->src[1] = MAKE_VSF_SOURCE(t_src_index(vp, &src[0]), t_swizzle(GET_SWZ(src[0].Swizzle, 2)), // z - t_swizzle(GET_SWZ(src[0].Swizzle, 0)), // x - t_swizzle(GET_SWZ(src[0].Swizzle, 1)), // y - t_swizzle(GET_SWZ(src[0].Swizzle, 3)), // w - t_src_class(src[0]. - File), - src[0]. - NegateBase ? - VSF_FLAG_ALL : - VSF_FLAG_NONE) | - (src[0].RelAddr << 4); - - o_inst->src[2] = MAKE_VSF_SOURCE(u_temp_i + 1, - VSF_IN_COMPONENT_X, - VSF_IN_COMPONENT_Y, - VSF_IN_COMPONENT_Z, - VSF_IN_COMPONENT_W, - VSF_IN_CLASS_TMP, - VSF_FLAG_NONE); - - goto next; - - case OPCODE_RCC: - fprintf(stderr, "Dont know how to handle op %d yet\n", - vpi->Opcode); - _mesa_exit(-1); + /* FIXME */ + o_inst = t_opcode_flr(vp, vpi, o_inst, src, &u_temp_i); + break; + case OPCODE_FRC: + t_opcode_frc(vp, vpi, o_inst, src); + break; + case OPCODE_LG2: + t_opcode_lg2(vp, vpi, o_inst, src); + break; + case OPCODE_LIT: + t_opcode_lit(vp, vpi, o_inst, src); + break; + case OPCODE_LOG: + t_opcode_log(vp, vpi, o_inst, src); + break; + case OPCODE_MAD: + t_opcode_mad(vp, vpi, o_inst, src); + break; + case OPCODE_MAX: + t_opcode_max(vp, vpi, o_inst, src); + break; + case OPCODE_MIN: + t_opcode_min(vp, vpi, o_inst, src); + break; + case OPCODE_MOV: + t_opcode_mov(vp, vpi, o_inst, src); break; - case OPCODE_END: + case OPCODE_MUL: + t_opcode_mul(vp, vpi, o_inst, src); + break; + case OPCODE_POW: + t_opcode_pow(vp, vpi, o_inst, src); + break; + case OPCODE_RCP: + t_opcode_rcp(vp, vpi, o_inst, src); + break; + case OPCODE_RSQ: + t_opcode_rsq(vp, vpi, o_inst, src); + break; + case OPCODE_SGE: + t_opcode_sge(vp, vpi, o_inst, src); + break; + case OPCODE_SLT: + t_opcode_slt(vp, vpi, o_inst, src); + break; + case OPCODE_SUB: + t_opcode_sub(vp, vpi, o_inst, src); + break; + case OPCODE_SWZ: + t_opcode_swz(vp, vpi, o_inst, src); + break; + case OPCODE_XPD: + /* FIXME */ + o_inst = t_opcode_xpd(vp, vpi, o_inst, src, &u_temp_i); break; default: + assert(0); break; } - - o_inst->op = - MAKE_VSF_OP(t_opcode(vpi->Opcode), - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - - if (are_srcs_scalar) { - switch (operands) { - case 1: - o_inst->src[0] = t_src_scalar(vp, &src[0]); - o_inst->src[1] = ZERO_SRC_0; - o_inst->src[2] = ZERO_SRC_0; - break; - - case 2: - o_inst->src[0] = t_src_scalar(vp, &src[0]); - o_inst->src[1] = t_src_scalar(vp, &src[1]); - o_inst->src[2] = ZERO_SRC_1; - break; - - case 3: - o_inst->src[0] = t_src_scalar(vp, &src[0]); - o_inst->src[1] = t_src_scalar(vp, &src[1]); - o_inst->src[2] = t_src_scalar(vp, &src[2]); - break; - - default: - fprintf(stderr, - "scalars and op RCC not handled yet"); - _mesa_exit(-1); - break; - } - } else { - switch (operands) { - case 1: - o_inst->src[0] = t_src(vp, &src[0]); - o_inst->src[1] = ZERO_SRC_0; - o_inst->src[2] = ZERO_SRC_0; - break; - - case 2: - o_inst->src[0] = t_src(vp, &src[0]); - o_inst->src[1] = t_src(vp, &src[1]); - o_inst->src[2] = ZERO_SRC_1; - break; - - case 3: - o_inst->src[0] = t_src(vp, &src[0]); - o_inst->src[1] = t_src(vp, &src[1]); - o_inst->src[2] = t_src(vp, &src[2]); - break; - - default: - fprintf(stderr, - "scalars and op RCC not handled yet"); - _mesa_exit(-1); - break; - } - } - next:; } - /* Will most likely segfault before we get here... fix later. */ - if (o_inst - vp->program.body.i >= VSF_MAX_FRAGMENT_LENGTH / 4) { + vp->program.length = (o_inst - vp->program.body.i) * 4; + if (vp->program.length >= VSF_MAX_FRAGMENT_LENGTH) { vp->program.length = 0; vp->native = GL_FALSE; - return; } - vp->program.length = (o_inst - vp->program.body.i) * 4; #if 0 fprintf(stderr, "hw program:\n"); for (i = 0; i < vp->program.length; i++) @@ -1065,7 +1291,8 @@ static void position_invariant(struct gl_program *prog) struct gl_program_parameter_list *paramList; int i; - gl_state_index tokens[STATE_LENGTH] = { STATE_MVP_MATRIX, 0, 0, 0, 0 }; + gl_state_index tokens[STATE_LENGTH] = + { STATE_MVP_MATRIX, 0, 0, 0, 0 }; /* tokens[4] = matrix modifier */ #ifdef PREFER_DP4 @@ -1159,8 +1386,8 @@ static void insert_wpos(struct r300_vertex_program *vp, prog->NumInstructions - 1); /* END */ _mesa_copy_instructions(&vpi[prog->NumInstructions + 1], - &prog->Instructions[prog->NumInstructions - 1], - 1); + &prog->Instructions[prog->NumInstructions - + 1], 1); vpi_insert = &vpi[prog->NumInstructions - 1]; vpi_insert[i].Opcode = OPCODE_MOV; @@ -1206,8 +1433,8 @@ static void pos_as_texcoord(struct r300_vertex_program *vp, prog->NumTemporaries++; for (vpi = prog->Instructions; vpi->Opcode != OPCODE_END; vpi++) { - if (vpi->DstReg.File == PROGRAM_OUTPUT && - vpi->DstReg.Index == VERT_RESULT_HPOS) { + if (vpi->DstReg.File == PROGRAM_OUTPUT + && vpi->DstReg.Index == VERT_RESULT_HPOS) { vpi->DstReg.File = PROGRAM_TEMPORARY; vpi->DstReg.Index = tempregi; } @@ -1223,20 +1450,18 @@ static struct r300_vertex_program *build_program(struct r300_vertex_program_key vp = _mesa_calloc(sizeof(*vp)); _mesa_memcpy(&vp->key, wanted_key, sizeof(vp->key)); - vp->wpos_idx = wpos_idx; if (mesa_vp->IsPositionInvariant) { position_invariant(&mesa_vp->Base); } - if (wpos_idx > -1) + if (wpos_idx > -1) { pos_as_texcoord(vp, &mesa_vp->Base); + } assert(mesa_vp->Base.NumInstructions); - vp->num_temporaries = mesa_vp->Base.NumTemporaries; - r300TranslateVertexShader(vp, mesa_vp->Base.Instructions); return vp; @@ -1252,11 +1477,10 @@ void r300SelectVertexShader(r300ContextPtr r300) struct r300_vertex_program *vp; GLint wpos_idx; - vpc = (struct r300_vertex_program_cont *)ctx->VertexProgram._Current; + vpc = + (struct r300_vertex_program_cont *)ctx->VertexProgram._Current; InputsRead = ctx->FragmentProgram._Current->Base.InputsRead; - wanted_key.OutputsWritten |= 1 << VERT_RESULT_HPOS; - wpos_idx = -1; if (InputsRead & FRAG_BIT_WPOS) { for (i = 0; i < ctx->Const.MaxTextureUnits; i++) @@ -1271,28 +1495,35 @@ void r300SelectVertexShader(r300ContextPtr r300) InputsRead |= (FRAG_BIT_TEX0 << i); wpos_idx = i; } + wanted_key.InputsRead = vpc->mesa_program.Base.InputsRead; + wanted_key.OutputsWritten = vpc->mesa_program.Base.OutputsWritten; + + wanted_key.OutputsWritten |= 1 << VERT_RESULT_HPOS; - if (InputsRead & FRAG_BIT_COL0) + if (InputsRead & FRAG_BIT_COL0) { wanted_key.OutputsWritten |= 1 << VERT_RESULT_COL0; + } - if ((InputsRead & FRAG_BIT_COL1) /*|| - (InputsRead & FRAG_BIT_FOGC) */ ) + if ((InputsRead & FRAG_BIT_COL1)) { wanted_key.OutputsWritten |= 1 << VERT_RESULT_COL1; + } - for (i = 0; i < ctx->Const.MaxTextureUnits; i++) - if (InputsRead & (FRAG_BIT_TEX0 << i)) + for (i = 0; i < ctx->Const.MaxTextureUnits; i++) { + if (InputsRead & (FRAG_BIT_TEX0 << i)) { wanted_key.OutputsWritten |= 1 << (VERT_RESULT_TEX0 + i); + } + } - wanted_key.InputsRead = vpc->mesa_program.Base.InputsRead; if (vpc->mesa_program.IsPositionInvariant) { /* we wan't position don't we ? */ wanted_key.InputsRead |= (1 << VERT_ATTRIB_POS); + wanted_key.OutputsWritten |= (1 << VERT_RESULT_HPOS); } for (vp = vpc->progs; vp; vp = vp->next) - if (_mesa_memcmp(&vp->key, &wanted_key, sizeof(wanted_key)) == - 0) { + if (_mesa_memcmp(&vp->key, &wanted_key, sizeof(wanted_key)) + == 0) { r300->selected_vp = vp; return; } diff --git a/src/mesa/drivers/dri/r300/r300_vertprog.h b/src/mesa/drivers/dri/r300/r300_vertprog.h index 252d5a901f..2d399e243a 100644 --- a/src/mesa/drivers/dri/r300/r300_vertprog.h +++ b/src/mesa/drivers/dri/r300/r300_vertprog.h @@ -3,10 +3,10 @@ #include "r300_reg.h" -typedef struct { - GLuint op; +struct r300_vertprog_instruction { + GLuint opcode; GLuint src[3]; -} VERTEX_SHADER_INSTRUCTION; +}; #define VSF_FLAG_X 1 #define VSF_FLAG_Y 2 |