diff options
Diffstat (limited to 'src/mesa/drivers/dri/r300/compiler/radeon_program_tex.c')
-rw-r--r-- | src/mesa/drivers/dri/r300/compiler/radeon_program_tex.c | 339 |
1 files changed, 199 insertions, 140 deletions
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_program_tex.c b/src/mesa/drivers/dri/r300/compiler/radeon_program_tex.c index ddce590ee6..530afa5e08 100644 --- a/src/mesa/drivers/dri/r300/compiler/radeon_program_tex.c +++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_tex.c @@ -1,5 +1,6 @@ /* * Copyright (C) 2010 Corbin Simpson + * Copyright (C) 2010 Marek Olšák <maraeo@gmail.com> * * All Rights Reserved. * @@ -46,31 +47,61 @@ static struct rc_src_register shadow_ambient(struct r300_fragment_program_compil return reg; } -static void lower_texture_rect(struct r300_fragment_program_compiler *compiler, - struct rc_instruction *inst) +static void scale_texcoords(struct r300_fragment_program_compiler *compiler, + struct rc_instruction *inst, + unsigned state_constant) { - struct rc_instruction *inst_rect; + struct rc_instruction *inst_mov; + unsigned temp = rc_find_free_temporary(&compiler->Base); - if (inst->U.I.TexSrcTarget == RC_TEXTURE_RECT || - compiler->state.unit[inst->U.I.TexSrcUnit].non_normalized_coords) { - inst_rect = rc_insert_new_instruction(&compiler->Base, inst->Prev); + inst_mov = rc_insert_new_instruction(&compiler->Base, inst->Prev); - inst_rect->U.I.Opcode = RC_OPCODE_MUL; - inst_rect->U.I.DstReg.File = RC_FILE_TEMPORARY; - inst_rect->U.I.DstReg.Index = temp; - inst_rect->U.I.SrcReg[0] = inst->U.I.SrcReg[0]; - inst_rect->U.I.SrcReg[1].File = RC_FILE_CONSTANT; - inst_rect->U.I.SrcReg[1].Index = - rc_constants_add_state(&compiler->Base.Program.Constants, - RC_STATE_R300_TEXRECT_FACTOR, inst->U.I.TexSrcUnit); + inst_mov->U.I.Opcode = RC_OPCODE_MUL; + inst_mov->U.I.DstReg.File = RC_FILE_TEMPORARY; + inst_mov->U.I.DstReg.Index = temp; + inst_mov->U.I.SrcReg[0] = inst->U.I.SrcReg[0]; + inst_mov->U.I.SrcReg[1].File = RC_FILE_CONSTANT; + inst_mov->U.I.SrcReg[1].Index = + rc_constants_add_state(&compiler->Base.Program.Constants, + state_constant, inst->U.I.TexSrcUnit); - reset_srcreg(&inst->U.I.SrcReg[0]); - inst->U.I.SrcReg[0].File = RC_FILE_TEMPORARY; - inst->U.I.SrcReg[0].Index = temp; + reset_srcreg(&inst->U.I.SrcReg[0]); + inst->U.I.SrcReg[0].File = RC_FILE_TEMPORARY; + inst->U.I.SrcReg[0].Index = temp; +} - inst->U.I.TexSrcTarget = RC_TEXTURE_2D; - } +static void projective_divide(struct r300_fragment_program_compiler *compiler, + struct rc_instruction *inst) +{ + struct rc_instruction *inst_mul, *inst_rcp; + + unsigned temp = rc_find_free_temporary(&compiler->Base); + + inst_rcp = rc_insert_new_instruction(&compiler->Base, inst->Prev); + inst_rcp->U.I.Opcode = RC_OPCODE_RCP; + inst_rcp->U.I.DstReg.File = RC_FILE_TEMPORARY; + inst_rcp->U.I.DstReg.Index = temp; + inst_rcp->U.I.DstReg.WriteMask = RC_MASK_W; + inst_rcp->U.I.SrcReg[0] = inst->U.I.SrcReg[0]; + /* Because the input can be arbitrarily swizzled, + * read the component mapped to W. */ + inst_rcp->U.I.SrcReg[0].Swizzle = + RC_MAKE_SWIZZLE_SMEAR(GET_SWZ(inst->U.I.SrcReg[0].Swizzle, 3)); + + inst_mul = rc_insert_new_instruction(&compiler->Base, inst->Prev); + inst_mul->U.I.Opcode = RC_OPCODE_MUL; + inst_mul->U.I.DstReg.File = RC_FILE_TEMPORARY; + inst_mul->U.I.DstReg.Index = temp; + inst_mul->U.I.SrcReg[0] = inst->U.I.SrcReg[0]; + inst_mul->U.I.SrcReg[1].File = RC_FILE_TEMPORARY; + inst_mul->U.I.SrcReg[1].Index = temp; + inst_mul->U.I.SrcReg[1].Swizzle = RC_SWIZZLE_WWWW; + + reset_srcreg(&inst->U.I.SrcReg[0]); + inst->U.I.Opcode = RC_OPCODE_TEX; + inst->U.I.SrcReg[0].File = RC_FILE_TEMPORARY; + inst->U.I.SrcReg[0].Index = temp; } /** @@ -88,6 +119,9 @@ int radeonTransformTEX( { struct r300_fragment_program_compiler *compiler = (struct r300_fragment_program_compiler*)data; + rc_wrap_mode wrapmode = compiler->state.unit[inst->U.I.TexSrcUnit].wrap_mode; + int is_rect = inst->U.I.TexSrcTarget == RC_TEXTURE_RECT || + compiler->state.unit[inst->U.I.TexSrcUnit].non_normalized_coords; if (inst->U.I.Opcode != RC_OPCODE_TEX && inst->U.I.Opcode != RC_OPCODE_TXB && @@ -141,16 +175,18 @@ int radeonTransformTEX( inst_rcp->U.I.DstReg.Index = tmp_recip_w; inst_rcp->U.I.DstReg.WriteMask = RC_MASK_W; inst_rcp->U.I.SrcReg[0] = inst->U.I.SrcReg[0]; - inst_rcp->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_WWWW; + inst_rcp->U.I.SrcReg[0].Swizzle = + RC_MAKE_SWIZZLE_SMEAR(GET_SWZ(inst->U.I.SrcReg[0].Swizzle, 3)); } - /* Perspective-divide r by W (if it's TXP) and add the texture sample (see below). */ + /* Perspective-divide Z by W (if it's TXP) and add the texture sample (see below). */ tmp_sum = rc_find_free_temporary(c); inst_mad = rc_insert_new_instruction(c, inst_rcp ? inst_rcp : inst); inst_mad->U.I.DstReg.File = RC_FILE_TEMPORARY; inst_mad->U.I.DstReg.Index = tmp_sum; inst_mad->U.I.SrcReg[0] = inst->U.I.SrcReg[0]; - inst_mad->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_ZZZZ; + inst_mad->U.I.SrcReg[0].Swizzle = + RC_MAKE_SWIZZLE_SMEAR(GET_SWZ(inst->U.I.SrcReg[0].Swizzle, 2)); if (inst->U.I.Opcode == RC_OPCODE_TXP) { inst_mad->U.I.Opcode = RC_OPCODE_MAD; inst_mad->U.I.SrcReg[1].File = RC_FILE_TEMPORARY; @@ -206,11 +242,22 @@ int radeonTransformTEX( } } - /* Texture wrap modes don't work on NPOT textures or texrects. - * - * The game plan is simple. We have two flags, fake_npot and - * non_normalized_coords, as well as a tex target. The RECT tex target - * will make the emitted code use non-scaled texcoords. + /* R300 cannot sample from rectangles and the wrap mode fallback needs + * normalized coordinates anyway. */ + if (inst->U.I.Opcode != RC_OPCODE_KIL && + is_rect && (!c->is_r500 || wrapmode != RC_WRAP_NONE)) { + scale_texcoords(compiler, inst, RC_STATE_R300_TEXRECT_FACTOR); + inst->U.I.TexSrcTarget = RC_TEXTURE_2D; + } + + /* Divide by W if needed. */ + if (inst->U.I.Opcode == RC_OPCODE_TXP && + (wrapmode == RC_WRAP_REPEAT || wrapmode == RC_WRAP_MIRRORED_REPEAT || + compiler->state.unit[inst->U.I.TexSrcUnit].clamp_and_scale_before_fetch)) { + projective_divide(compiler, inst); + } + + /* Texture wrap modes don't work on NPOT textures. * * Non-wrapped/clamped texcoords with NPOT are free in HW. Repeat and * mirroring are not. If we need to repeat, we do: @@ -235,128 +282,140 @@ int radeonTransformTEX( * ~ C & M. ;) */ if (inst->U.I.Opcode != RC_OPCODE_KIL && - (inst->U.I.TexSrcTarget == RC_TEXTURE_RECT || - compiler->state.unit[inst->U.I.TexSrcUnit].fake_npot || - compiler->state.unit[inst->U.I.TexSrcUnit].non_normalized_coords)) { - rc_wrap_mode wrapmode = compiler->state.unit[inst->U.I.TexSrcUnit].wrap_mode; - - /* R300 cannot sample from rectangles. */ - if (!c->is_r500) { - lower_texture_rect(compiler, inst); - } - - if (compiler->state.unit[inst->U.I.TexSrcUnit].fake_npot && - wrapmode != RC_WRAP_NONE) { + wrapmode != RC_WRAP_NONE) { + struct rc_instruction *inst_mov; + unsigned temp = rc_find_free_temporary(c); + + if (wrapmode == RC_WRAP_REPEAT) { + /* Both instructions will be paired up. */ + struct rc_instruction *inst_frc = rc_insert_new_instruction(c, inst->Prev); + + inst_frc->U.I.Opcode = RC_OPCODE_FRC; + inst_frc->U.I.DstReg.File = RC_FILE_TEMPORARY; + inst_frc->U.I.DstReg.Index = temp; + inst_frc->U.I.DstReg.WriteMask = RC_MASK_XYZ; + inst_frc->U.I.SrcReg[0] = inst->U.I.SrcReg[0]; + } else if (wrapmode == RC_WRAP_MIRRORED_REPEAT) { + /* + * Function: + * f(v) = 1 - abs(frac(v * 0.5) * 2 - 1) + * + * Code: + * MUL temp, src0, 0.5 + * FRC temp, temp + * MAD temp, temp, 2, -1 + * ADD temp, 1, -abs(temp) + */ + + struct rc_instruction *inst_mul, *inst_frc, *inst_mad, *inst_add; + unsigned two, two_swizzle; + + inst_mul = rc_insert_new_instruction(c, inst->Prev); + + inst_mul->U.I.Opcode = RC_OPCODE_MUL; + inst_mul->U.I.DstReg.File = RC_FILE_TEMPORARY; + inst_mul->U.I.DstReg.Index = temp; + inst_mul->U.I.DstReg.WriteMask = RC_MASK_XYZ; + inst_mul->U.I.SrcReg[0] = inst->U.I.SrcReg[0]; + inst_mul->U.I.SrcReg[1].Swizzle = RC_SWIZZLE_HHHH; + + inst_frc = rc_insert_new_instruction(c, inst->Prev); + + inst_frc->U.I.Opcode = RC_OPCODE_FRC; + inst_frc->U.I.DstReg.File = RC_FILE_TEMPORARY; + inst_frc->U.I.DstReg.Index = temp; + inst_frc->U.I.DstReg.WriteMask = RC_MASK_XYZ; + inst_frc->U.I.SrcReg[0].File = RC_FILE_TEMPORARY; + inst_frc->U.I.SrcReg[0].Index = temp; + inst_frc->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_XYZ0; + + two = rc_constants_add_immediate_scalar(&c->Program.Constants, 2, &two_swizzle); + inst_mad = rc_insert_new_instruction(c, inst->Prev); + + inst_mad->U.I.Opcode = RC_OPCODE_MAD; + inst_mad->U.I.DstReg.File = RC_FILE_TEMPORARY; + inst_mad->U.I.DstReg.Index = temp; + inst_mad->U.I.DstReg.WriteMask = RC_MASK_XYZ; + inst_mad->U.I.SrcReg[0].File = RC_FILE_TEMPORARY; + inst_mad->U.I.SrcReg[0].Index = temp; + inst_mad->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_XYZ0; + inst_mad->U.I.SrcReg[1].File = RC_FILE_CONSTANT; + inst_mad->U.I.SrcReg[1].Index = two; + inst_mad->U.I.SrcReg[1].Swizzle = two_swizzle; + inst_mad->U.I.SrcReg[2].Swizzle = RC_SWIZZLE_1111; + inst_mad->U.I.SrcReg[2].Negate = RC_MASK_XYZ; + + inst_add = rc_insert_new_instruction(c, inst->Prev); + + inst_add->U.I.Opcode = RC_OPCODE_ADD; + inst_add->U.I.DstReg.File = RC_FILE_TEMPORARY; + inst_add->U.I.DstReg.Index = temp; + inst_add->U.I.DstReg.WriteMask = RC_MASK_XYZ; + inst_add->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_1111; + inst_add->U.I.SrcReg[1].File = RC_FILE_TEMPORARY; + inst_add->U.I.SrcReg[1].Index = temp; + inst_add->U.I.SrcReg[1].Swizzle = RC_SWIZZLE_XYZ0; + inst_add->U.I.SrcReg[1].Abs = 1; + inst_add->U.I.SrcReg[1].Negate = RC_MASK_XYZ; + } else if (wrapmode == RC_WRAP_MIRRORED_CLAMP) { + /* + * Mirrored clamp modes are bloody simple, we just use abs + * to mirror [0, 1] into [-1, 0]. This works for + * all modes i.e. CLAMP, CLAMP_TO_EDGE, and CLAMP_TO_BORDER. + */ struct rc_instruction *inst_mov; - unsigned temp = rc_find_free_temporary(c); - - /* For NPOT fallback, we need normalized coordinates anyway. */ - if (c->is_r500) { - lower_texture_rect(compiler, inst); - } - - if (wrapmode == RC_WRAP_REPEAT) { - /* Both instructions will be paired up. */ - struct rc_instruction *inst_frc = rc_insert_new_instruction(c, inst->Prev); - - inst_frc->U.I.Opcode = RC_OPCODE_FRC; - inst_frc->U.I.DstReg.File = RC_FILE_TEMPORARY; - inst_frc->U.I.DstReg.Index = temp; - inst_frc->U.I.DstReg.WriteMask = RC_MASK_XYZ; - inst_frc->U.I.SrcReg[0] = inst->U.I.SrcReg[0]; - } else if (wrapmode == RC_WRAP_MIRRORED_REPEAT) { - /* - * Function: - * f(v) = 1 - abs(frac(v * 0.5) * 2 - 1) - * - * Code: - * MUL temp, src0, 0.5 - * FRC temp, temp - * MAD temp, temp, 2, -1 - * ADD temp, 1, -abs(temp) - */ - - struct rc_instruction *inst_mul, *inst_frc, *inst_mad, *inst_add; - unsigned two, two_swizzle; - - inst_mul = rc_insert_new_instruction(c, inst->Prev); - - inst_mul->U.I.Opcode = RC_OPCODE_MUL; - inst_mul->U.I.DstReg.File = RC_FILE_TEMPORARY; - inst_mul->U.I.DstReg.Index = temp; - inst_mul->U.I.DstReg.WriteMask = RC_MASK_XYZ; - inst_mul->U.I.SrcReg[0] = inst->U.I.SrcReg[0]; - inst_mul->U.I.SrcReg[1].Swizzle = RC_SWIZZLE_HHHH; - - inst_frc = rc_insert_new_instruction(c, inst->Prev); - - inst_frc->U.I.Opcode = RC_OPCODE_FRC; - inst_frc->U.I.DstReg.File = RC_FILE_TEMPORARY; - inst_frc->U.I.DstReg.Index = temp; - inst_frc->U.I.DstReg.WriteMask = RC_MASK_XYZ; - inst_frc->U.I.SrcReg[0].File = RC_FILE_TEMPORARY; - inst_frc->U.I.SrcReg[0].Index = temp; - inst_frc->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_XYZ0; - - two = rc_constants_add_immediate_scalar(&c->Program.Constants, 2, &two_swizzle); - inst_mad = rc_insert_new_instruction(c, inst->Prev); - - inst_mad->U.I.Opcode = RC_OPCODE_MAD; - inst_mad->U.I.DstReg.File = RC_FILE_TEMPORARY; - inst_mad->U.I.DstReg.Index = temp; - inst_mad->U.I.DstReg.WriteMask = RC_MASK_XYZ; - inst_mad->U.I.SrcReg[0].File = RC_FILE_TEMPORARY; - inst_mad->U.I.SrcReg[0].Index = temp; - inst_mad->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_XYZ0; - inst_mad->U.I.SrcReg[1].File = RC_FILE_CONSTANT; - inst_mad->U.I.SrcReg[1].Index = two; - inst_mad->U.I.SrcReg[1].Swizzle = two_swizzle; - inst_mad->U.I.SrcReg[2].Swizzle = RC_SWIZZLE_1111; - inst_mad->U.I.SrcReg[2].Negate = RC_MASK_XYZ; - - inst_add = rc_insert_new_instruction(c, inst->Prev); - - inst_add->U.I.Opcode = RC_OPCODE_ADD; - inst_add->U.I.DstReg.File = RC_FILE_TEMPORARY; - inst_add->U.I.DstReg.Index = temp; - inst_add->U.I.DstReg.WriteMask = RC_MASK_XYZ; - inst_add->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_1111; - inst_add->U.I.SrcReg[1].File = RC_FILE_TEMPORARY; - inst_add->U.I.SrcReg[1].Index = temp; - inst_add->U.I.SrcReg[1].Swizzle = RC_SWIZZLE_XYZ0; - inst_add->U.I.SrcReg[1].Abs = 1; - inst_add->U.I.SrcReg[1].Negate = RC_MASK_XYZ; - } else if (wrapmode == RC_WRAP_MIRRORED_CLAMP) { - /* - * Mirrored clamp modes are bloody simple, we just use abs - * to mirror [0, 1] into [-1, 0]. This works for - * all modes i.e. CLAMP, CLAMP_TO_EDGE, and CLAMP_TO_BORDER. - */ - struct rc_instruction *inst_mov; - - inst_mov = rc_insert_new_instruction(c, inst->Prev); - - inst_mov->U.I.Opcode = RC_OPCODE_MOV; - inst_mov->U.I.DstReg.File = RC_FILE_TEMPORARY; - inst_mov->U.I.DstReg.Index = temp; - inst_mov->U.I.DstReg.WriteMask = RC_MASK_XYZ; - inst_mov->U.I.SrcReg[0] = inst->U.I.SrcReg[0]; - inst_mov->U.I.SrcReg[0].Abs = 1; - } - /* Preserve W for TXP/TXB. */ inst_mov = rc_insert_new_instruction(c, inst->Prev); inst_mov->U.I.Opcode = RC_OPCODE_MOV; inst_mov->U.I.DstReg.File = RC_FILE_TEMPORARY; inst_mov->U.I.DstReg.Index = temp; - inst_mov->U.I.DstReg.WriteMask = RC_MASK_W; + inst_mov->U.I.DstReg.WriteMask = RC_MASK_XYZ; inst_mov->U.I.SrcReg[0] = inst->U.I.SrcReg[0]; - - reset_srcreg(&inst->U.I.SrcReg[0]); - inst->U.I.SrcReg[0].File = RC_FILE_TEMPORARY; - inst->U.I.SrcReg[0].Index = temp; + inst_mov->U.I.SrcReg[0].Abs = 1; } + + /* Preserve W for TXP/TXB. */ + inst_mov = rc_insert_new_instruction(c, inst->Prev); + + inst_mov->U.I.Opcode = RC_OPCODE_MOV; + inst_mov->U.I.DstReg.File = RC_FILE_TEMPORARY; + inst_mov->U.I.DstReg.Index = temp; + inst_mov->U.I.DstReg.WriteMask = RC_MASK_W; + inst_mov->U.I.SrcReg[0] = inst->U.I.SrcReg[0]; + + reset_srcreg(&inst->U.I.SrcReg[0]); + inst->U.I.SrcReg[0].File = RC_FILE_TEMPORARY; + inst->U.I.SrcReg[0].Index = temp; + } + + if (inst->U.I.Opcode != RC_OPCODE_KIL && + compiler->state.unit[inst->U.I.TexSrcUnit].clamp_and_scale_before_fetch) { + struct rc_instruction *inst_mov; + unsigned temp = rc_find_free_temporary(c); + + /* Saturate XYZ. */ + inst_mov = rc_insert_new_instruction(c, inst->Prev); + inst_mov->U.I.Opcode = RC_OPCODE_MOV; + inst_mov->U.I.SaturateMode = RC_SATURATE_ZERO_ONE; + inst_mov->U.I.DstReg.File = RC_FILE_TEMPORARY; + inst_mov->U.I.DstReg.Index = temp; + inst_mov->U.I.DstReg.WriteMask = RC_MASK_XYZ; + inst_mov->U.I.SrcReg[0] = inst->U.I.SrcReg[0]; + + /* Copy W. */ + inst_mov = rc_insert_new_instruction(c, inst->Prev); + inst_mov->U.I.Opcode = RC_OPCODE_MOV; + inst_mov->U.I.DstReg.File = RC_FILE_TEMPORARY; + inst_mov->U.I.DstReg.Index = temp; + inst_mov->U.I.DstReg.WriteMask = RC_MASK_W; + inst_mov->U.I.SrcReg[0] = inst->U.I.SrcReg[0]; + + reset_srcreg(&inst->U.I.SrcReg[0]); + inst->U.I.SrcReg[0].File = RC_FILE_TEMPORARY; + inst->U.I.SrcReg[0].Index = temp; + + scale_texcoords(compiler, inst, RC_STATE_R300_TEXSCALE_FACTOR); } /* Cannot write texture to output registers (all chips) or with masks (non-r500) */ |