diff options
Diffstat (limited to 'src/gallium/auxiliary/tgsi')
21 files changed, 765 insertions, 1179 deletions
diff --git a/src/gallium/auxiliary/tgsi/tgsi_build.c b/src/gallium/auxiliary/tgsi/tgsi_build.c index e0cfc54420..4fa10e2f7e 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_build.c +++ b/src/gallium/auxiliary/tgsi/tgsi_build.c @@ -472,9 +472,9 @@ tgsi_default_full_instruction( void ) unsigned i; full_instruction.Instruction = tgsi_default_instruction(); - full_instruction.InstructionExtNv = tgsi_default_instruction_ext_nv(); full_instruction.InstructionExtLabel = tgsi_default_instruction_ext_label(); full_instruction.InstructionExtTexture = tgsi_default_instruction_ext_texture(); + full_instruction.InstructionExtPredicate = tgsi_default_instruction_ext_predicate(); for( i = 0; i < TGSI_FULL_MAX_DST_REGISTERS; i++ ) { full_instruction.FullDstRegisters[i] = tgsi_default_full_dst_register(); } @@ -512,34 +512,6 @@ tgsi_build_full_instruction( header ); prev_token = (struct tgsi_token *) instruction; - if( tgsi_compare_instruction_ext_nv( - full_inst->InstructionExtNv, - tgsi_default_instruction_ext_nv() ) ) { - struct tgsi_instruction_ext_nv *instruction_ext_nv; - - if( maxsize <= size ) - return 0; - instruction_ext_nv = - (struct tgsi_instruction_ext_nv *) &tokens[size]; - size++; - - *instruction_ext_nv = tgsi_build_instruction_ext_nv( - full_inst->InstructionExtNv.Precision, - full_inst->InstructionExtNv.CondDstIndex, - full_inst->InstructionExtNv.CondFlowIndex, - full_inst->InstructionExtNv.CondMask, - full_inst->InstructionExtNv.CondSwizzleX, - full_inst->InstructionExtNv.CondSwizzleY, - full_inst->InstructionExtNv.CondSwizzleZ, - full_inst->InstructionExtNv.CondSwizzleW, - full_inst->InstructionExtNv.CondDstUpdate, - full_inst->InstructionExtNv.CondFlowEnable, - prev_token, - instruction, - header ); - prev_token = (struct tgsi_token *) instruction_ext_nv; - } - if( tgsi_compare_instruction_ext_label( full_inst->InstructionExtLabel, tgsi_default_instruction_ext_label() ) ) { @@ -578,6 +550,29 @@ tgsi_build_full_instruction( prev_token = (struct tgsi_token *) instruction_ext_texture; } + if (tgsi_compare_instruction_ext_predicate(full_inst->InstructionExtPredicate, + tgsi_default_instruction_ext_predicate())) { + struct tgsi_instruction_ext_predicate *instruction_ext_predicate; + + if (maxsize <= size) { + return 0; + } + instruction_ext_predicate = (struct tgsi_instruction_ext_predicate *)&tokens[size]; + size++; + + *instruction_ext_predicate = + tgsi_build_instruction_ext_predicate(full_inst->InstructionExtPredicate.SrcIndex, + full_inst->InstructionExtPredicate.Negate, + full_inst->InstructionExtPredicate.SwizzleX, + full_inst->InstructionExtPredicate.SwizzleY, + full_inst->InstructionExtPredicate.SwizzleZ, + full_inst->InstructionExtPredicate.SwizzleW, + prev_token, + instruction, + header); + prev_token = (struct tgsi_token *)instruction_ext_predicate; + } + for( i = 0; i < full_inst->Instruction.NumDstRegs; i++ ) { const struct tgsi_full_dst_register *reg = &full_inst->FullDstRegisters[i]; struct tgsi_dst_register *dst_register; @@ -597,30 +592,6 @@ tgsi_build_full_instruction( header ); prev_token = (struct tgsi_token *) dst_register; - if( tgsi_compare_dst_register_ext_concode( - reg->DstRegisterExtConcode, - tgsi_default_dst_register_ext_concode() ) ) { - struct tgsi_dst_register_ext_concode *dst_register_ext_concode; - - if( maxsize <= size ) - return 0; - dst_register_ext_concode = - (struct tgsi_dst_register_ext_concode *) &tokens[size]; - size++; - - *dst_register_ext_concode = tgsi_build_dst_register_ext_concode( - reg->DstRegisterExtConcode.CondMask, - reg->DstRegisterExtConcode.CondSwizzleX, - reg->DstRegisterExtConcode.CondSwizzleY, - reg->DstRegisterExtConcode.CondSwizzleZ, - reg->DstRegisterExtConcode.CondSwizzleW, - reg->DstRegisterExtConcode.CondSrcIndex, - prev_token, - instruction, - header ); - prev_token = (struct tgsi_token *) dst_register_ext_concode; - } - if( tgsi_compare_dst_register_ext_modulate( reg->DstRegisterExtModulate, tgsi_default_dst_register_ext_modulate() ) ) { @@ -687,40 +658,6 @@ tgsi_build_full_instruction( header ); prev_token = (struct tgsi_token *) src_register; - if( tgsi_compare_src_register_ext_swz( - reg->SrcRegisterExtSwz, - tgsi_default_src_register_ext_swz() ) ) { - struct tgsi_src_register_ext_swz *src_register_ext_swz; - - /* Use of the extended swizzle requires the simple swizzle to be identity. - */ - assert( reg->SrcRegister.SwizzleX == TGSI_SWIZZLE_X ); - assert( reg->SrcRegister.SwizzleY == TGSI_SWIZZLE_Y ); - assert( reg->SrcRegister.SwizzleZ == TGSI_SWIZZLE_Z ); - assert( reg->SrcRegister.SwizzleW == TGSI_SWIZZLE_W ); - assert( reg->SrcRegister.Negate == FALSE ); - - if( maxsize <= size ) - return 0; - src_register_ext_swz = - (struct tgsi_src_register_ext_swz *) &tokens[size]; - size++; - - *src_register_ext_swz = tgsi_build_src_register_ext_swz( - reg->SrcRegisterExtSwz.ExtSwizzleX, - reg->SrcRegisterExtSwz.ExtSwizzleY, - reg->SrcRegisterExtSwz.ExtSwizzleZ, - reg->SrcRegisterExtSwz.ExtSwizzleW, - reg->SrcRegisterExtSwz.NegateX, - reg->SrcRegisterExtSwz.NegateY, - reg->SrcRegisterExtSwz.NegateZ, - reg->SrcRegisterExtSwz.NegateW, - prev_token, - instruction, - header ); - prev_token = (struct tgsi_token *) src_register_ext_swz; - } - if( tgsi_compare_src_register_ext_mod( reg->SrcRegisterExtMod, tgsi_default_src_register_ext_mod() ) ) { @@ -809,29 +746,6 @@ tgsi_build_full_instruction( return size; } -struct tgsi_instruction_ext_nv -tgsi_default_instruction_ext_nv( void ) -{ - struct tgsi_instruction_ext_nv instruction_ext_nv; - - instruction_ext_nv.Type = TGSI_INSTRUCTION_EXT_TYPE_NV; - instruction_ext_nv.Precision = TGSI_PRECISION_DEFAULT; - instruction_ext_nv.CondDstIndex = 0; - instruction_ext_nv.CondFlowIndex = 0; - instruction_ext_nv.CondMask = TGSI_CC_TR; - instruction_ext_nv.CondSwizzleX = TGSI_SWIZZLE_X; - instruction_ext_nv.CondSwizzleY = TGSI_SWIZZLE_Y; - instruction_ext_nv.CondSwizzleZ = TGSI_SWIZZLE_Z; - instruction_ext_nv.CondSwizzleW = TGSI_SWIZZLE_W; - instruction_ext_nv.CondDstUpdate = 0; - instruction_ext_nv.CondFlowEnable = 0; - instruction_ext_nv.Padding = 0; - instruction_ext_nv.Extended = 0; - - return instruction_ext_nv; -} - - /** test for inequality of 32-bit values pointed to by a and b */ static INLINE boolean compare32(const void *a, const void *b) @@ -839,53 +753,6 @@ compare32(const void *a, const void *b) return *((uint32_t *) a) != *((uint32_t *) b); } - -unsigned -tgsi_compare_instruction_ext_nv( - struct tgsi_instruction_ext_nv a, - struct tgsi_instruction_ext_nv b ) -{ - a.Padding = b.Padding = 0; - a.Extended = b.Extended = 0; - return compare32(&a, &b); -} - -struct tgsi_instruction_ext_nv -tgsi_build_instruction_ext_nv( - unsigned precision, - unsigned cond_dst_index, - unsigned cond_flow_index, - unsigned cond_mask, - unsigned cond_swizzle_x, - unsigned cond_swizzle_y, - unsigned cond_swizzle_z, - unsigned cond_swizzle_w, - unsigned cond_dst_update, - unsigned cond_flow_enable, - struct tgsi_token *prev_token, - struct tgsi_instruction *instruction, - struct tgsi_header *header ) -{ - struct tgsi_instruction_ext_nv instruction_ext_nv; - - instruction_ext_nv = tgsi_default_instruction_ext_nv(); - instruction_ext_nv.Precision = precision; - instruction_ext_nv.CondDstIndex = cond_dst_index; - instruction_ext_nv.CondFlowIndex = cond_flow_index; - instruction_ext_nv.CondMask = cond_mask; - instruction_ext_nv.CondSwizzleX = cond_swizzle_x; - instruction_ext_nv.CondSwizzleY = cond_swizzle_y; - instruction_ext_nv.CondSwizzleZ = cond_swizzle_z; - instruction_ext_nv.CondSwizzleW = cond_swizzle_w; - instruction_ext_nv.CondDstUpdate = cond_dst_update; - instruction_ext_nv.CondFlowEnable = cond_flow_enable; - - prev_token->Extended = 1; - instruction_grow( instruction, header ); - - return instruction_ext_nv; -} - struct tgsi_instruction_ext_label tgsi_default_instruction_ext_label( void ) { @@ -968,6 +835,60 @@ tgsi_build_instruction_ext_texture( return instruction_ext_texture; } +struct tgsi_instruction_ext_predicate +tgsi_default_instruction_ext_predicate(void) +{ + struct tgsi_instruction_ext_predicate instruction_ext_predicate; + + instruction_ext_predicate.Type = TGSI_INSTRUCTION_EXT_TYPE_PREDICATE; + instruction_ext_predicate.SwizzleX = TGSI_SWIZZLE_X; + instruction_ext_predicate.SwizzleY = TGSI_SWIZZLE_Y; + instruction_ext_predicate.SwizzleZ = TGSI_SWIZZLE_Z; + instruction_ext_predicate.SwizzleW = TGSI_SWIZZLE_W; + instruction_ext_predicate.Negate = 0; + instruction_ext_predicate.SrcIndex = 0; + instruction_ext_predicate.Padding = 0; + instruction_ext_predicate.Extended = 0; + + return instruction_ext_predicate; +} + +unsigned +tgsi_compare_instruction_ext_predicate(struct tgsi_instruction_ext_predicate a, + struct tgsi_instruction_ext_predicate b) +{ + a.Padding = b.Padding = 0; + a.Extended = b.Extended = 0; + return compare32(&a, &b); +} + +struct tgsi_instruction_ext_predicate +tgsi_build_instruction_ext_predicate(unsigned index, + unsigned negate, + unsigned swizzleX, + unsigned swizzleY, + unsigned swizzleZ, + unsigned swizzleW, + struct tgsi_token *prev_token, + struct tgsi_instruction *instruction, + struct tgsi_header *header) +{ + struct tgsi_instruction_ext_predicate instruction_ext_predicate; + + instruction_ext_predicate = tgsi_default_instruction_ext_predicate(); + instruction_ext_predicate.SwizzleX = swizzleX; + instruction_ext_predicate.SwizzleY = swizzleY; + instruction_ext_predicate.SwizzleZ = swizzleZ; + instruction_ext_predicate.SwizzleW = swizzleW; + instruction_ext_predicate.Negate = negate; + instruction_ext_predicate.SrcIndex = index; + + prev_token->Extended = 1; + instruction_grow(instruction, header); + + return instruction_ext_predicate; +} + struct tgsi_src_register tgsi_default_src_register( void ) { @@ -1033,7 +954,6 @@ tgsi_default_full_src_register( void ) struct tgsi_full_src_register full_src_register; full_src_register.SrcRegister = tgsi_default_src_register(); - full_src_register.SrcRegisterExtSwz = tgsi_default_src_register_ext_swz(); full_src_register.SrcRegisterExtMod = tgsi_default_src_register_ext_mod(); full_src_register.SrcRegisterInd = tgsi_default_src_register(); full_src_register.SrcRegisterDim = tgsi_default_dimension(); @@ -1042,76 +962,6 @@ tgsi_default_full_src_register( void ) return full_src_register; } -struct tgsi_src_register_ext_swz -tgsi_default_src_register_ext_swz( void ) -{ - struct tgsi_src_register_ext_swz src_register_ext_swz; - - src_register_ext_swz.Type = TGSI_SRC_REGISTER_EXT_TYPE_SWZ; - src_register_ext_swz.ExtSwizzleX = TGSI_EXTSWIZZLE_X; - src_register_ext_swz.ExtSwizzleY = TGSI_EXTSWIZZLE_Y; - src_register_ext_swz.ExtSwizzleZ = TGSI_EXTSWIZZLE_Z; - src_register_ext_swz.ExtSwizzleW = TGSI_EXTSWIZZLE_W; - src_register_ext_swz.NegateX = 0; - src_register_ext_swz.NegateY = 0; - src_register_ext_swz.NegateZ = 0; - src_register_ext_swz.NegateW = 0; - src_register_ext_swz.Padding = 0; - src_register_ext_swz.Extended = 0; - - return src_register_ext_swz; -} - -unsigned -tgsi_compare_src_register_ext_swz( - struct tgsi_src_register_ext_swz a, - struct tgsi_src_register_ext_swz b ) -{ - a.Padding = b.Padding = 0; - a.Extended = b.Extended = 0; - return compare32(&a, &b); -} - -struct tgsi_src_register_ext_swz -tgsi_build_src_register_ext_swz( - unsigned ext_swizzle_x, - unsigned ext_swizzle_y, - unsigned ext_swizzle_z, - unsigned ext_swizzle_w, - unsigned negate_x, - unsigned negate_y, - unsigned negate_z, - unsigned negate_w, - struct tgsi_token *prev_token, - struct tgsi_instruction *instruction, - struct tgsi_header *header ) -{ - struct tgsi_src_register_ext_swz src_register_ext_swz; - - assert( ext_swizzle_x <= TGSI_EXTSWIZZLE_ONE ); - assert( ext_swizzle_y <= TGSI_EXTSWIZZLE_ONE ); - assert( ext_swizzle_z <= TGSI_EXTSWIZZLE_ONE ); - assert( ext_swizzle_w <= TGSI_EXTSWIZZLE_ONE ); - assert( negate_x <= 1 ); - assert( negate_y <= 1 ); - assert( negate_z <= 1 ); - assert( negate_w <= 1 ); - - src_register_ext_swz = tgsi_default_src_register_ext_swz(); - src_register_ext_swz.ExtSwizzleX = ext_swizzle_x; - src_register_ext_swz.ExtSwizzleY = ext_swizzle_y; - src_register_ext_swz.ExtSwizzleZ = ext_swizzle_z; - src_register_ext_swz.ExtSwizzleW = ext_swizzle_w; - src_register_ext_swz.NegateX = negate_x; - src_register_ext_swz.NegateY = negate_y; - src_register_ext_swz.NegateZ = negate_z; - src_register_ext_swz.NegateW = negate_w; - - prev_token->Extended = 1; - instruction_grow( instruction, header ); - - return src_register_ext_swz; -} struct tgsi_src_register_ext_mod tgsi_default_src_register_ext_mod( void ) @@ -1253,77 +1103,12 @@ tgsi_default_full_dst_register( void ) full_dst_register.DstRegister = tgsi_default_dst_register(); full_dst_register.DstRegisterInd = tgsi_default_src_register(); - full_dst_register.DstRegisterExtConcode = - tgsi_default_dst_register_ext_concode(); full_dst_register.DstRegisterExtModulate = tgsi_default_dst_register_ext_modulate(); return full_dst_register; } -struct tgsi_dst_register_ext_concode -tgsi_default_dst_register_ext_concode( void ) -{ - struct tgsi_dst_register_ext_concode dst_register_ext_concode; - - dst_register_ext_concode.Type = TGSI_DST_REGISTER_EXT_TYPE_CONDCODE; - dst_register_ext_concode.CondMask = TGSI_CC_TR; - dst_register_ext_concode.CondSwizzleX = TGSI_SWIZZLE_X; - dst_register_ext_concode.CondSwizzleY = TGSI_SWIZZLE_Y; - dst_register_ext_concode.CondSwizzleZ = TGSI_SWIZZLE_Z; - dst_register_ext_concode.CondSwizzleW = TGSI_SWIZZLE_W; - dst_register_ext_concode.CondSrcIndex = 0; - dst_register_ext_concode.Padding = 0; - dst_register_ext_concode.Extended = 0; - - return dst_register_ext_concode; -} - -unsigned -tgsi_compare_dst_register_ext_concode( - struct tgsi_dst_register_ext_concode a, - struct tgsi_dst_register_ext_concode b ) -{ - a.Padding = b.Padding = 0; - a.Extended = b.Extended = 0; - return compare32(&a, &b); -} - -struct tgsi_dst_register_ext_concode -tgsi_build_dst_register_ext_concode( - unsigned cc, - unsigned swizzle_x, - unsigned swizzle_y, - unsigned swizzle_z, - unsigned swizzle_w, - int index, - struct tgsi_token *prev_token, - struct tgsi_instruction *instruction, - struct tgsi_header *header ) -{ - struct tgsi_dst_register_ext_concode dst_register_ext_concode; - - assert( cc <= TGSI_CC_FL ); - assert( swizzle_x <= TGSI_SWIZZLE_W ); - assert( swizzle_y <= TGSI_SWIZZLE_W ); - assert( swizzle_z <= TGSI_SWIZZLE_W ); - assert( swizzle_w <= TGSI_SWIZZLE_W ); - assert( index >= -32768 && index <= 32767 ); - - dst_register_ext_concode = tgsi_default_dst_register_ext_concode(); - dst_register_ext_concode.CondMask = cc; - dst_register_ext_concode.CondSwizzleX = swizzle_x; - dst_register_ext_concode.CondSwizzleY = swizzle_y; - dst_register_ext_concode.CondSwizzleZ = swizzle_z; - dst_register_ext_concode.CondSwizzleW = swizzle_w; - dst_register_ext_concode.CondSrcIndex = index; - - prev_token->Extended = 1; - instruction_grow( instruction, header ); - - return dst_register_ext_concode; -} - struct tgsi_dst_register_ext_modulate tgsi_default_dst_register_ext_modulate( void ) { diff --git a/src/gallium/auxiliary/tgsi/tgsi_build.h b/src/gallium/auxiliary/tgsi/tgsi_build.h index 17d977b059..669712eb8f 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_build.h +++ b/src/gallium/auxiliary/tgsi/tgsi_build.h @@ -157,30 +157,6 @@ tgsi_build_full_instruction( struct tgsi_header *header, unsigned maxsize ); -struct tgsi_instruction_ext_nv -tgsi_default_instruction_ext_nv( void ); - -unsigned -tgsi_compare_instruction_ext_nv( - struct tgsi_instruction_ext_nv a, - struct tgsi_instruction_ext_nv b ); - -struct tgsi_instruction_ext_nv -tgsi_build_instruction_ext_nv( - unsigned precision, - unsigned cond_dst_index, - unsigned cond_flow_index, - unsigned cond_mask, - unsigned cond_swizzle_x, - unsigned cond_swizzle_y, - unsigned cond_swizzle_z, - unsigned cond_swizzle_w, - unsigned cond_dst_update, - unsigned cond_flow_enable, - struct tgsi_token *prev_token, - struct tgsi_instruction *instruction, - struct tgsi_header *header ); - struct tgsi_instruction_ext_label tgsi_default_instruction_ext_label( void ); @@ -211,6 +187,24 @@ tgsi_build_instruction_ext_texture( struct tgsi_instruction *instruction, struct tgsi_header *header ); +struct tgsi_instruction_ext_predicate +tgsi_default_instruction_ext_predicate(void); + +unsigned +tgsi_compare_instruction_ext_predicate(struct tgsi_instruction_ext_predicate a, + struct tgsi_instruction_ext_predicate b); + +struct tgsi_instruction_ext_predicate +tgsi_build_instruction_ext_predicate(unsigned index, + unsigned negate, + unsigned swizzleX, + unsigned swizzleY, + unsigned swizzleZ, + unsigned swizzleW, + struct tgsi_token *prev_token, + struct tgsi_instruction *instruction, + struct tgsi_header *header); + struct tgsi_src_register tgsi_default_src_register( void ); @@ -231,28 +225,6 @@ tgsi_build_src_register( struct tgsi_full_src_register tgsi_default_full_src_register( void ); -struct tgsi_src_register_ext_swz -tgsi_default_src_register_ext_swz( void ); - -unsigned -tgsi_compare_src_register_ext_swz( - struct tgsi_src_register_ext_swz a, - struct tgsi_src_register_ext_swz b ); - -struct tgsi_src_register_ext_swz -tgsi_build_src_register_ext_swz( - unsigned ext_swizzle_x, - unsigned ext_swizzle_y, - unsigned ext_swizzle_z, - unsigned ext_swizzle_w, - unsigned negate_x, - unsigned negate_y, - unsigned negate_z, - unsigned negate_w, - struct tgsi_token *prev_token, - struct tgsi_instruction *instruction, - struct tgsi_header *header ); - struct tgsi_src_register_ext_mod tgsi_default_src_register_ext_mod( void ); @@ -297,26 +269,6 @@ tgsi_build_dst_register( struct tgsi_full_dst_register tgsi_default_full_dst_register( void ); -struct tgsi_dst_register_ext_concode -tgsi_default_dst_register_ext_concode( void ); - -unsigned -tgsi_compare_dst_register_ext_concode( - struct tgsi_dst_register_ext_concode a, - struct tgsi_dst_register_ext_concode b ); - -struct tgsi_dst_register_ext_concode -tgsi_build_dst_register_ext_concode( - unsigned cc, - unsigned swizzle_x, - unsigned swizzle_y, - unsigned swizzle_z, - unsigned swizzle_w, - int index, - struct tgsi_token *prev_token, - struct tgsi_instruction *instruction, - struct tgsi_header *header ); - struct tgsi_dst_register_ext_modulate tgsi_default_dst_register_ext_modulate( void ); diff --git a/src/gallium/auxiliary/tgsi/tgsi_dump.c b/src/gallium/auxiliary/tgsi/tgsi_dump.c index 111d95b666..d16e64f9c5 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_dump.c +++ b/src/gallium/auxiliary/tgsi/tgsi_dump.c @@ -100,7 +100,8 @@ static const char *file_names[TGSI_FILE_COUNT] = "SAMP", "ADDR", "IMM", - "LOOP" + "LOOP", + "PRED" }; static const char *interpolate_names[] = @@ -148,15 +149,6 @@ static const char *texture_names[] = "SHADOWRECT" }; -static const char *extswizzle_names[] = -{ - "x", - "y", - "z", - "w", - "0", - "1" -}; static const char *modulate_names[TGSI_MODULATE_COUNT] = { @@ -446,24 +438,6 @@ iter_instruction( ENM( src->SrcRegister.SwizzleZ, swizzle_names ); ENM( src->SrcRegister.SwizzleW, swizzle_names ); } - if (src->SrcRegisterExtSwz.ExtSwizzleX != TGSI_EXTSWIZZLE_X || - src->SrcRegisterExtSwz.ExtSwizzleY != TGSI_EXTSWIZZLE_Y || - src->SrcRegisterExtSwz.ExtSwizzleZ != TGSI_EXTSWIZZLE_Z || - src->SrcRegisterExtSwz.ExtSwizzleW != TGSI_EXTSWIZZLE_W) { - CHR( '.' ); - if (src->SrcRegisterExtSwz.NegateX) - TXT("-"); - ENM( src->SrcRegisterExtSwz.ExtSwizzleX, extswizzle_names ); - if (src->SrcRegisterExtSwz.NegateY) - TXT("-"); - ENM( src->SrcRegisterExtSwz.ExtSwizzleY, extswizzle_names ); - if (src->SrcRegisterExtSwz.NegateZ) - TXT("-"); - ENM( src->SrcRegisterExtSwz.ExtSwizzleZ, extswizzle_names ); - if (src->SrcRegisterExtSwz.NegateW) - TXT("-"); - ENM( src->SrcRegisterExtSwz.ExtSwizzleW, extswizzle_names ); - } if (src->SrcRegisterExtMod.Complement) CHR( ')' ); diff --git a/src/gallium/auxiliary/tgsi/tgsi_dump_c.c b/src/gallium/auxiliary/tgsi/tgsi_dump_c.c index 4a9c02b141..4648051e29 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_dump_c.c +++ b/src/gallium/auxiliary/tgsi/tgsi_dump_c.c @@ -79,7 +79,8 @@ static const char *TGSI_FILES[TGSI_FILE_COUNT] = "FILE_SAMPLER", "FILE_ADDRESS", "FILE_IMMEDIATE", - "FILE_LOOP" + "FILE_LOOP", + "FILE_PREDICATE" }; static const char *TGSI_INTERPOLATES[] = @@ -114,32 +115,11 @@ static const char *TGSI_SATS[] = static const char *TGSI_INSTRUCTION_EXTS[] = { - "INSTRUCTION_EXT_TYPE_NV", + "", "INSTRUCTION_EXT_TYPE_LABEL", "INSTRUCTION_EXT_TYPE_TEXTURE" }; -static const char *TGSI_PRECISIONS[] = -{ - "PRECISION_DEFAULT", - "PRECISION_FLOAT32", - "PRECISION_FLOAT16", - "PRECISION_FIXED12" -}; - -static const char *TGSI_CCS[] = -{ - "CC_GT", - "CC_EQ", - "CC_LT", - "CC_UN", - "CC_GE", - "CC_LE", - "CC_NE", - "CC_TR", - "CC_FL" -}; - static const char *TGSI_SWIZZLES[] = { "SWIZZLE_X", @@ -163,20 +143,10 @@ static const char *TGSI_TEXTURES[] = static const char *TGSI_SRC_REGISTER_EXTS[] = { - "SRC_REGISTER_EXT_TYPE_SWZ", + "", "SRC_REGISTER_EXT_TYPE_MOD" }; -static const char *TGSI_EXTSWIZZLES[] = -{ - "EXTSWIZZLE_X", - "EXTSWIZZLE_Y", - "EXTSWIZZLE_Z", - "EXTSWIZZLE_W", - "EXTSWIZZLE_ZERO", - "EXTSWIZZLE_ONE" -}; - static const char *TGSI_WRITEMASKS[] = { "0", @@ -199,7 +169,7 @@ static const char *TGSI_WRITEMASKS[] = static const char *TGSI_DST_REGISTER_EXTS[] = { - "DST_REGISTER_EXT_TYPE_CONDCODE", + "", "DST_REGISTER_EXT_TYPE_MODULATE" }; @@ -327,60 +297,6 @@ dump_instruction_verbose( UIX( inst->Instruction.Padding ); } - if( deflt || tgsi_compare_instruction_ext_nv( inst->InstructionExtNv, fi->InstructionExtNv ) ) { - EOL(); - TXT( "\nType : " ); - ENM( inst->InstructionExtNv.Type, TGSI_INSTRUCTION_EXTS ); - if( deflt || fi->InstructionExtNv.Precision != inst->InstructionExtNv.Precision ) { - TXT( "\nPrecision : " ); - ENM( inst->InstructionExtNv.Precision, TGSI_PRECISIONS ); - } - if( deflt || fi->InstructionExtNv.CondDstIndex != inst->InstructionExtNv.CondDstIndex ) { - TXT( "\nCondDstIndex : " ); - UID( inst->InstructionExtNv.CondDstIndex ); - } - if( deflt || fi->InstructionExtNv.CondFlowIndex != inst->InstructionExtNv.CondFlowIndex ) { - TXT( "\nCondFlowIndex : " ); - UID( inst->InstructionExtNv.CondFlowIndex ); - } - if( deflt || fi->InstructionExtNv.CondMask != inst->InstructionExtNv.CondMask ) { - TXT( "\nCondMask : " ); - ENM( inst->InstructionExtNv.CondMask, TGSI_CCS ); - } - if( deflt || fi->InstructionExtNv.CondSwizzleX != inst->InstructionExtNv.CondSwizzleX ) { - TXT( "\nCondSwizzleX : " ); - ENM( inst->InstructionExtNv.CondSwizzleX, TGSI_SWIZZLES ); - } - if( deflt || fi->InstructionExtNv.CondSwizzleY != inst->InstructionExtNv.CondSwizzleY ) { - TXT( "\nCondSwizzleY : " ); - ENM( inst->InstructionExtNv.CondSwizzleY, TGSI_SWIZZLES ); - } - if( deflt || fi->InstructionExtNv.CondSwizzleZ != inst->InstructionExtNv.CondSwizzleZ ) { - TXT( "\nCondSwizzleZ : " ); - ENM( inst->InstructionExtNv.CondSwizzleZ, TGSI_SWIZZLES ); - } - if( deflt || fi->InstructionExtNv.CondSwizzleW != inst->InstructionExtNv.CondSwizzleW ) { - TXT( "\nCondSwizzleW : " ); - ENM( inst->InstructionExtNv.CondSwizzleW, TGSI_SWIZZLES ); - } - if( deflt || fi->InstructionExtNv.CondDstUpdate != inst->InstructionExtNv.CondDstUpdate ) { - TXT( "\nCondDstUpdate : " ); - UID( inst->InstructionExtNv.CondDstUpdate ); - } - if( deflt || fi->InstructionExtNv.CondFlowEnable != inst->InstructionExtNv.CondFlowEnable ) { - TXT( "\nCondFlowEnable: " ); - UID( inst->InstructionExtNv.CondFlowEnable ); - } - if( ignored ) { - TXT( "\nPadding : " ); - UIX( inst->InstructionExtNv.Padding ); - if( deflt || fi->InstructionExtNv.Extended != inst->InstructionExtNv.Extended ) { - TXT( "\nExtended : " ); - UID( inst->InstructionExtNv.Extended ); - } - } - } - if( deflt || tgsi_compare_instruction_ext_label( inst->InstructionExtLabel, fi->InstructionExtLabel ) ) { EOL(); TXT( "\nType : " ); @@ -451,44 +367,6 @@ dump_instruction_verbose( } } - if( deflt || tgsi_compare_dst_register_ext_concode( dst->DstRegisterExtConcode, fd->DstRegisterExtConcode ) ) { - EOL(); - TXT( "\nType : " ); - ENM( dst->DstRegisterExtConcode.Type, TGSI_DST_REGISTER_EXTS ); - if( deflt || fd->DstRegisterExtConcode.CondMask != dst->DstRegisterExtConcode.CondMask ) { - TXT( "\nCondMask : " ); - ENM( dst->DstRegisterExtConcode.CondMask, TGSI_CCS ); - } - if( deflt || fd->DstRegisterExtConcode.CondSwizzleX != dst->DstRegisterExtConcode.CondSwizzleX ) { - TXT( "\nCondSwizzleX: " ); - ENM( dst->DstRegisterExtConcode.CondSwizzleX, TGSI_SWIZZLES ); - } - if( deflt || fd->DstRegisterExtConcode.CondSwizzleY != dst->DstRegisterExtConcode.CondSwizzleY ) { - TXT( "\nCondSwizzleY: " ); - ENM( dst->DstRegisterExtConcode.CondSwizzleY, TGSI_SWIZZLES ); - } - if( deflt || fd->DstRegisterExtConcode.CondSwizzleZ != dst->DstRegisterExtConcode.CondSwizzleZ ) { - TXT( "\nCondSwizzleZ: " ); - ENM( dst->DstRegisterExtConcode.CondSwizzleZ, TGSI_SWIZZLES ); - } - if( deflt || fd->DstRegisterExtConcode.CondSwizzleW != dst->DstRegisterExtConcode.CondSwizzleW ) { - TXT( "\nCondSwizzleW: " ); - ENM( dst->DstRegisterExtConcode.CondSwizzleW, TGSI_SWIZZLES ); - } - if( deflt || fd->DstRegisterExtConcode.CondSrcIndex != dst->DstRegisterExtConcode.CondSrcIndex ) { - TXT( "\nCondSrcIndex: " ); - UID( dst->DstRegisterExtConcode.CondSrcIndex ); - } - if( ignored ) { - TXT( "\nPadding : " ); - UIX( dst->DstRegisterExtConcode.Padding ); - if( deflt || fd->DstRegisterExtConcode.Extended != dst->DstRegisterExtConcode.Extended ) { - TXT( "\nExtended : " ); - UID( dst->DstRegisterExtConcode.Extended ); - } - } - } - if( deflt || tgsi_compare_dst_register_ext_modulate( dst->DstRegisterExtModulate, fd->DstRegisterExtModulate ) ) { EOL(); TXT( "\nType : " ); @@ -556,52 +434,6 @@ dump_instruction_verbose( } } - if( deflt || tgsi_compare_src_register_ext_swz( src->SrcRegisterExtSwz, fs->SrcRegisterExtSwz ) ) { - EOL(); - TXT( "\nType : " ); - ENM( src->SrcRegisterExtSwz.Type, TGSI_SRC_REGISTER_EXTS ); - if( deflt || fs->SrcRegisterExtSwz.ExtSwizzleX != src->SrcRegisterExtSwz.ExtSwizzleX ) { - TXT( "\nExtSwizzleX: " ); - ENM( src->SrcRegisterExtSwz.ExtSwizzleX, TGSI_EXTSWIZZLES ); - } - if( deflt || fs->SrcRegisterExtSwz.ExtSwizzleY != src->SrcRegisterExtSwz.ExtSwizzleY ) { - TXT( "\nExtSwizzleY: " ); - ENM( src->SrcRegisterExtSwz.ExtSwizzleY, TGSI_EXTSWIZZLES ); - } - if( deflt || fs->SrcRegisterExtSwz.ExtSwizzleZ != src->SrcRegisterExtSwz.ExtSwizzleZ ) { - TXT( "\nExtSwizzleZ: " ); - ENM( src->SrcRegisterExtSwz.ExtSwizzleZ, TGSI_EXTSWIZZLES ); - } - if( deflt || fs->SrcRegisterExtSwz.ExtSwizzleW != src->SrcRegisterExtSwz.ExtSwizzleW ) { - TXT( "\nExtSwizzleW: " ); - ENM( src->SrcRegisterExtSwz.ExtSwizzleW, TGSI_EXTSWIZZLES ); - } - if( deflt || fs->SrcRegisterExtSwz.NegateX != src->SrcRegisterExtSwz.NegateX ) { - TXT( "\nNegateX : " ); - UID( src->SrcRegisterExtSwz.NegateX ); - } - if( deflt || fs->SrcRegisterExtSwz.NegateY != src->SrcRegisterExtSwz.NegateY ) { - TXT( "\nNegateY : " ); - UID( src->SrcRegisterExtSwz.NegateY ); - } - if( deflt || fs->SrcRegisterExtSwz.NegateZ != src->SrcRegisterExtSwz.NegateZ ) { - TXT( "\nNegateZ : " ); - UID( src->SrcRegisterExtSwz.NegateZ ); - } - if( deflt || fs->SrcRegisterExtSwz.NegateW != src->SrcRegisterExtSwz.NegateW ) { - TXT( "\nNegateW : " ); - UID( src->SrcRegisterExtSwz.NegateW ); - } - if( ignored ) { - TXT( "\nPadding : " ); - UIX( src->SrcRegisterExtSwz.Padding ); - if( deflt || fs->SrcRegisterExtSwz.Extended != src->SrcRegisterExtSwz.Extended ) { - TXT( "\nExtended : " ); - UID( src->SrcRegisterExtSwz.Extended ); - } - } - } - if( deflt || tgsi_compare_src_register_ext_mod( src->SrcRegisterExtMod, fs->SrcRegisterExtMod ) ) { EOL(); TXT( "\nType : " ); diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c index c79c56debd..b7569e74d4 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_exec.c +++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c @@ -107,6 +107,7 @@ #define TEMP_HALF_I TGSI_EXEC_TEMP_HALF_I #define TEMP_HALF_C TGSI_EXEC_TEMP_HALF_C #define TEMP_R0 TGSI_EXEC_TEMP_R0 +#define TEMP_P0 TGSI_EXEC_TEMP_P0 #define IS_CHANNEL_ENABLED(INST, CHAN)\ ((INST).FullDstRegisters[0].DstRegister.WriteMask & (1 << (CHAN))) @@ -210,9 +211,8 @@ tgsi_check_soa_dependencies(const struct tgsi_full_instruction *inst) uint channelsWritten = 0x0; FOR_EACH_ENABLED_CHANNEL(*inst, chan) { /* check if we're reading a channel that's been written */ - uint swizzle = tgsi_util_get_full_src_register_extswizzle(&inst->FullSrcRegisters[i], chan); - if (swizzle <= TGSI_SWIZZLE_W && - (channelsWritten & (1 << swizzle))) { + uint swizzle = tgsi_util_get_full_src_register_swizzle(&inst->FullSrcRegisters[i], chan); + if (channelsWritten & (1 << swizzle)) { return TRUE; } @@ -338,7 +338,7 @@ tgsi_exec_machine_bind_shader( /* XXX we only handle SOA dependencies properly for MOV/SWZ * at this time! */ - if (opcode != TGSI_OPCODE_MOV && opcode != TGSI_OPCODE_SWZ) { + if (opcode != TGSI_OPCODE_MOV) { debug_printf("Warning: SOA dependency in instruction" " is not handled:\n"); tgsi_dump_instruction(&parse.FullToken.FullInstruction, @@ -1130,10 +1130,10 @@ fetch_src_file_channel( union tgsi_exec_channel *chan ) { switch( swizzle ) { - case TGSI_EXTSWIZZLE_X: - case TGSI_EXTSWIZZLE_Y: - case TGSI_EXTSWIZZLE_Z: - case TGSI_EXTSWIZZLE_W: + case TGSI_SWIZZLE_X: + case TGSI_SWIZZLE_Y: + case TGSI_SWIZZLE_Z: + case TGSI_SWIZZLE_W: switch( file ) { case TGSI_FILE_CONSTANT: assert(mach->Consts); @@ -1188,6 +1188,17 @@ fetch_src_file_channel( chan->u[3] = mach->Addrs[index->i[3]].xyzw[swizzle].u[3]; break; + case TGSI_FILE_PREDICATE: + assert(index->i[0] < TGSI_EXEC_NUM_PREDS); + assert(index->i[1] < TGSI_EXEC_NUM_PREDS); + assert(index->i[2] < TGSI_EXEC_NUM_PREDS); + assert(index->i[3] < TGSI_EXEC_NUM_PREDS); + chan->u[0] = mach->Addrs[0].xyzw[swizzle].u[0]; + chan->u[1] = mach->Addrs[0].xyzw[swizzle].u[1]; + chan->u[2] = mach->Addrs[0].xyzw[swizzle].u[2]; + chan->u[3] = mach->Addrs[0].xyzw[swizzle].u[3]; + break; + case TGSI_FILE_OUTPUT: /* vertex/fragment output vars can be read too */ chan->u[0] = mach->Outputs[index->i[0]].xyzw[swizzle].u[0]; @@ -1201,14 +1212,6 @@ fetch_src_file_channel( } break; - case TGSI_EXTSWIZZLE_ZERO: - *chan = mach->Temps[TEMP_0_I].xyzw[TEMP_0_C]; - break; - - case TGSI_EXTSWIZZLE_ONE: - *chan = mach->Temps[TEMP_1_I].xyzw[TEMP_1_C]; - break; - default: assert( 0 ); } @@ -1367,7 +1370,7 @@ fetch_source( */ } - swizzle = tgsi_util_get_full_src_register_extswizzle( reg, chan_index ); + swizzle = tgsi_util_get_full_src_register_swizzle( reg, chan_index ); fetch_src_file_channel( mach, reg->SrcRegister.File, @@ -1475,119 +1478,17 @@ store_dest( dst = &mach->Addrs[index].xyzw[chan_index]; break; + case TGSI_FILE_PREDICATE: + index = reg->DstRegister.Index; + assert(index < TGSI_EXEC_NUM_PREDS); + dst = &mach->Addrs[index].xyzw[chan_index]; + break; + default: assert( 0 ); return; } - if (inst->InstructionExtNv.CondFlowEnable) { - union tgsi_exec_channel *cc = &mach->Temps[TEMP_CC_I].xyzw[TEMP_CC_C]; - uint swizzle; - uint shift; - uint mask; - uint test; - - /* Only CC0 supported. - */ - assert( inst->InstructionExtNv.CondFlowIndex < 1 ); - - switch (chan_index) { - case CHAN_X: - swizzle = inst->InstructionExtNv.CondSwizzleX; - break; - case CHAN_Y: - swizzle = inst->InstructionExtNv.CondSwizzleY; - break; - case CHAN_Z: - swizzle = inst->InstructionExtNv.CondSwizzleZ; - break; - case CHAN_W: - swizzle = inst->InstructionExtNv.CondSwizzleW; - break; - default: - assert( 0 ); - return; - } - - switch (swizzle) { - case TGSI_SWIZZLE_X: - shift = TGSI_EXEC_CC_X_SHIFT; - mask = TGSI_EXEC_CC_X_MASK; - break; - case TGSI_SWIZZLE_Y: - shift = TGSI_EXEC_CC_Y_SHIFT; - mask = TGSI_EXEC_CC_Y_MASK; - break; - case TGSI_SWIZZLE_Z: - shift = TGSI_EXEC_CC_Z_SHIFT; - mask = TGSI_EXEC_CC_Z_MASK; - break; - case TGSI_SWIZZLE_W: - shift = TGSI_EXEC_CC_W_SHIFT; - mask = TGSI_EXEC_CC_W_MASK; - break; - default: - assert( 0 ); - return; - } - - switch (inst->InstructionExtNv.CondMask) { - case TGSI_CC_GT: - test = ~(TGSI_EXEC_CC_GT << shift) & mask; - for (i = 0; i < QUAD_SIZE; i++) - if (cc->u[i] & test) - execmask &= ~(1 << i); - break; - - case TGSI_CC_EQ: - test = ~(TGSI_EXEC_CC_EQ << shift) & mask; - for (i = 0; i < QUAD_SIZE; i++) - if (cc->u[i] & test) - execmask &= ~(1 << i); - break; - - case TGSI_CC_LT: - test = ~(TGSI_EXEC_CC_LT << shift) & mask; - for (i = 0; i < QUAD_SIZE; i++) - if (cc->u[i] & test) - execmask &= ~(1 << i); - break; - - case TGSI_CC_GE: - test = ~((TGSI_EXEC_CC_GT | TGSI_EXEC_CC_EQ) << shift) & mask; - for (i = 0; i < QUAD_SIZE; i++) - if (cc->u[i] & test) - execmask &= ~(1 << i); - break; - - case TGSI_CC_LE: - test = ~((TGSI_EXEC_CC_LT | TGSI_EXEC_CC_EQ) << shift) & mask; - for (i = 0; i < QUAD_SIZE; i++) - if (cc->u[i] & test) - execmask &= ~(1 << i); - break; - - case TGSI_CC_NE: - test = ~((TGSI_EXEC_CC_GT | TGSI_EXEC_CC_LT | TGSI_EXEC_CC_UN) << shift) & mask; - for (i = 0; i < QUAD_SIZE; i++) - if (cc->u[i] & test) - execmask &= ~(1 << i); - break; - - case TGSI_CC_TR: - break; - - case TGSI_CC_FL: - for (i = 0; i < QUAD_SIZE; i++) - execmask &= ~(1 << i); - break; - - default: - assert( 0 ); - return; - } - } - switch (inst->Instruction.Saturate) { case TGSI_SAT_NONE: for (i = 0; i < QUAD_SIZE; i++) @@ -1622,51 +1523,6 @@ store_dest( default: assert( 0 ); } - - if (inst->InstructionExtNv.CondDstUpdate) { - union tgsi_exec_channel *cc = &mach->Temps[TEMP_CC_I].xyzw[TEMP_CC_C]; - uint shift; - uint mask; - - /* Only CC0 supported. - */ - assert( inst->InstructionExtNv.CondDstIndex < 1 ); - - switch (chan_index) { - case CHAN_X: - shift = TGSI_EXEC_CC_X_SHIFT; - mask = ~TGSI_EXEC_CC_X_MASK; - break; - case CHAN_Y: - shift = TGSI_EXEC_CC_Y_SHIFT; - mask = ~TGSI_EXEC_CC_Y_MASK; - break; - case CHAN_Z: - shift = TGSI_EXEC_CC_Z_SHIFT; - mask = ~TGSI_EXEC_CC_Z_MASK; - break; - case CHAN_W: - shift = TGSI_EXEC_CC_W_SHIFT; - mask = ~TGSI_EXEC_CC_W_MASK; - break; - default: - assert( 0 ); - return; - } - - for (i = 0; i < QUAD_SIZE; i++) - if (execmask & (1 << i)) { - cc->u[i] &= mask; - if (dst->f[i] < 0.0f) - cc->u[i] |= TGSI_EXEC_CC_LT << shift; - else if (dst->f[i] > 0.0f) - cc->u[i] |= TGSI_EXEC_CC_GT << shift; - else if (dst->f[i] == 0.0f) - cc->u[i] |= TGSI_EXEC_CC_EQ << shift; - else - cc->u[i] |= TGSI_EXEC_CC_UN << shift; - } - } } #define FETCH(VAL,INDEX,CHAN)\ @@ -1689,10 +1545,8 @@ exec_kil(struct tgsi_exec_machine *mach, uint kilmask = 0; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */ union tgsi_exec_channel r[1]; - /* This mask stores component bits that were already tested. Note that - * we test if the value is less than zero, so 1.0 and 0.0 need not to be - * tested. */ - uniquemask = (1 << TGSI_EXTSWIZZLE_ZERO) | (1 << TGSI_EXTSWIZZLE_ONE); + /* This mask stores component bits that were already tested. */ + uniquemask = 0; for (chan_index = 0; chan_index < 4; chan_index++) { @@ -1700,7 +1554,7 @@ exec_kil(struct tgsi_exec_machine *mach, uint i; /* unswizzle channel */ - swizzle = tgsi_util_get_full_src_register_extswizzle ( + swizzle = tgsi_util_get_full_src_register_swizzle ( &inst->FullSrcRegisters[0], chan_index); @@ -1728,32 +1582,8 @@ exec_kilp(struct tgsi_exec_machine *mach, { uint kilmask; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */ - if (inst->InstructionExtNv.CondFlowEnable) { - uint swizzle[4]; - uint chan_index; - - kilmask = 0x0; - - swizzle[0] = inst->InstructionExtNv.CondSwizzleX; - swizzle[1] = inst->InstructionExtNv.CondSwizzleY; - swizzle[2] = inst->InstructionExtNv.CondSwizzleZ; - swizzle[3] = inst->InstructionExtNv.CondSwizzleW; - - for (chan_index = 0; chan_index < 4; chan_index++) - { - uint i; - - for (i = 0; i < 4; i++) { - /* TODO: evaluate the condition code */ - if (0) - kilmask |= 1 << i; - } - } - } - else { - /* "unconditional" kil */ - kilmask = mach->ExecMask; - } + /* "unconditional" kil */ + kilmask = mach->ExecMask; mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] |= kilmask; } @@ -1981,8 +1811,8 @@ exec_declaration( break; default: - eval = NULL; assert( 0 ); + return; } if( mask == TGSI_WRITEMASK_XYZW ) { @@ -2031,7 +1861,6 @@ exec_instruction( break; case TGSI_OPCODE_MOV: - case TGSI_OPCODE_SWZ: if (inst->Flags & SOA_DEPENDENCY_FLAG) { /* Do all fetches into temp regs, then do all stores to avoid * intermediate/accidental clobbering. This could be done all the @@ -3223,22 +3052,6 @@ exec_instruction( /* no-op */ break; - case TGSI_OPCODE_NOISE1: - assert( 0 ); - break; - - case TGSI_OPCODE_NOISE2: - assert( 0 ); - break; - - case TGSI_OPCODE_NOISE3: - assert( 0 ); - break; - - case TGSI_OPCODE_NOISE4: - assert( 0 ); - break; - case TGSI_OPCODE_NOP: break; diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.h b/src/gallium/auxiliary/tgsi/tgsi_exec.h index c72f76809d..471f591dd6 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_exec.h +++ b/src/gallium/auxiliary/tgsi/tgsi_exec.h @@ -168,13 +168,18 @@ struct tgsi_exec_labels #define TGSI_EXEC_TEMP_ADDR (TGSI_EXEC_NUM_TEMPS + 8) #define TGSI_EXEC_NUM_ADDRS 1 -#define TGSI_EXEC_NUM_TEMP_EXTRAS 9 +/* predicate register */ +#define TGSI_EXEC_TEMP_P0 (TGSI_EXEC_NUM_TEMPS + 9) +#define TGSI_EXEC_NUM_PREDS 1 +#define TGSI_EXEC_NUM_TEMP_EXTRAS 10 -#define TGSI_EXEC_MAX_COND_NESTING 20 -#define TGSI_EXEC_MAX_LOOP_NESTING 20 -#define TGSI_EXEC_MAX_CALL_NESTING 20 + + +#define TGSI_EXEC_MAX_COND_NESTING 32 +#define TGSI_EXEC_MAX_LOOP_NESTING 32 +#define TGSI_EXEC_MAX_CALL_NESTING 32 /* The maximum number of input attributes per vertex. For 2D * input register files, this is the stride between two 1D diff --git a/src/gallium/auxiliary/tgsi/tgsi_info.c b/src/gallium/auxiliary/tgsi/tgsi_info.c index 17af4cb7ad..be375cabb8 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_info.c +++ b/src/gallium/auxiliary/tgsi/tgsi_info.c @@ -134,10 +134,10 @@ static const struct tgsi_opcode_info opcode_info[TGSI_OPCODE_LAST] = { 0, 0, 0, 0, 0, 1, "BGNSUB", TGSI_OPCODE_BGNSUB }, { 0, 0, 0, 1, 1, 0, "ENDLOOP", TGSI_OPCODE_ENDLOOP }, { 0, 0, 0, 0, 1, 0, "ENDSUB", TGSI_OPCODE_ENDSUB }, - { 1, 1, 0, 0, 0, 0, "NOISE1", TGSI_OPCODE_NOISE1 }, - { 1, 1, 0, 0, 0, 0, "NOISE2", TGSI_OPCODE_NOISE2 }, - { 1, 1, 0, 0, 0, 0, "NOISE3", TGSI_OPCODE_NOISE3 }, - { 1, 1, 0, 0, 0, 0, "NOISE4", TGSI_OPCODE_NOISE4 }, + { 0, 0, 0, 0, 0, 0, "", 103 }, /* removed */ + { 0, 0, 0, 0, 0, 0, "", 104 }, /* removed */ + { 0, 0, 0, 0, 0, 0, "", 105 }, /* removed */ + { 0, 0, 0, 0, 0, 0, "", 106 }, /* removed */ { 0, 0, 0, 0, 0, 0, "NOP", TGSI_OPCODE_NOP }, { 0, 0, 0, 0, 0, 0, "", 108 }, /* removed */ { 0, 0, 0, 0, 0, 0, "", 109 }, /* removed */ @@ -149,7 +149,7 @@ static const struct tgsi_opcode_info opcode_info[TGSI_OPCODE_LAST] = { 0, 1, 0, 0, 0, 0, "BREAKC", TGSI_OPCODE_BREAKC }, { 0, 1, 0, 0, 0, 0, "KIL", TGSI_OPCODE_KIL }, { 0, 0, 0, 0, 0, 0, "END", TGSI_OPCODE_END }, - { 1, 1, 0, 0, 0, 0, "SWZ", TGSI_OPCODE_SWZ } + { 0, 0, 0, 0, 0, 0, "", 118 } /* removed */ }; const struct tgsi_opcode_info * diff --git a/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h b/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h index e7bcf4bf75..b34263da48 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h +++ b/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h @@ -139,10 +139,6 @@ OP00_LBL(BGNLOOP) OP00(BGNSUB) OP00_LBL(ENDLOOP) OP00(ENDSUB) -OP11(NOISE1) -OP11(NOISE2) -OP11(NOISE3) -OP11(NOISE4) OP00(NOP) OP11(NRM4) OP01(CALLNZ) @@ -150,7 +146,6 @@ OP01(IFC) OP01(BREAKC) OP01(KIL) OP00(END) -OP11(SWZ) #undef OP00 diff --git a/src/gallium/auxiliary/tgsi/tgsi_parse.c b/src/gallium/auxiliary/tgsi/tgsi_parse.c index 4870f82b6b..83f9df1183 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_parse.c +++ b/src/gallium/auxiliary/tgsi/tgsi_parse.c @@ -181,10 +181,6 @@ tgsi_parse_token( next_token( ctx, &token ); switch( token.Type ) { - case TGSI_INSTRUCTION_EXT_TYPE_NV: - copy_token(&inst->InstructionExtNv, &token); - break; - case TGSI_INSTRUCTION_EXT_TYPE_LABEL: copy_token(&inst->InstructionExtLabel, &token); break; @@ -193,6 +189,10 @@ tgsi_parse_token( copy_token(&inst->InstructionExtTexture, &token); break; + case TGSI_INSTRUCTION_EXT_TYPE_PREDICATE: + copy_token(&inst->InstructionExtPredicate, &token); + break; + default: assert( 0 ); } @@ -220,11 +220,6 @@ tgsi_parse_token( next_token( ctx, &token ); switch( token.Type ) { - case TGSI_DST_REGISTER_EXT_TYPE_CONDCODE: - copy_token(&inst->FullDstRegisters[i].DstRegisterExtConcode, - &token); - break; - case TGSI_DST_REGISTER_EXT_TYPE_MODULATE: copy_token(&inst->FullDstRegisters[i].DstRegisterExtModulate, &token); @@ -264,11 +259,6 @@ tgsi_parse_token( next_token( ctx, &token ); switch( token.Type ) { - case TGSI_SRC_REGISTER_EXT_TYPE_SWZ: - copy_token(&inst->FullSrcRegisters[i].SrcRegisterExtSwz, - &token); - break; - case TGSI_SRC_REGISTER_EXT_TYPE_MOD: copy_token(&inst->FullSrcRegisters[i].SrcRegisterExtMod, &token); diff --git a/src/gallium/auxiliary/tgsi/tgsi_parse.h b/src/gallium/auxiliary/tgsi/tgsi_parse.h index a26ee5ba86..76f1676d85 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_parse.h +++ b/src/gallium/auxiliary/tgsi/tgsi_parse.h @@ -49,14 +49,12 @@ struct tgsi_full_dst_register { struct tgsi_dst_register DstRegister; struct tgsi_src_register DstRegisterInd; - struct tgsi_dst_register_ext_concode DstRegisterExtConcode; struct tgsi_dst_register_ext_modulate DstRegisterExtModulate; }; struct tgsi_full_src_register { struct tgsi_src_register SrcRegister; - struct tgsi_src_register_ext_swz SrcRegisterExtSwz; struct tgsi_src_register_ext_mod SrcRegisterExtMod; struct tgsi_src_register SrcRegisterInd; struct tgsi_dimension SrcRegisterDim; @@ -82,9 +80,9 @@ struct tgsi_full_immediate struct tgsi_full_instruction { struct tgsi_instruction Instruction; - struct tgsi_instruction_ext_nv InstructionExtNv; struct tgsi_instruction_ext_label InstructionExtLabel; struct tgsi_instruction_ext_texture InstructionExtTexture; + struct tgsi_instruction_ext_predicate InstructionExtPredicate; struct tgsi_full_dst_register FullDstRegisters[TGSI_FULL_MAX_DST_REGISTERS]; struct tgsi_full_src_register FullSrcRegisters[TGSI_FULL_MAX_SRC_REGISTERS]; uint Flags; /**< user-defined usage */ diff --git a/src/gallium/auxiliary/tgsi/tgsi_ppc.c b/src/gallium/auxiliary/tgsi/tgsi_ppc.c index 4b1c7d4e01..617fd7f6be 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_ppc.c +++ b/src/gallium/auxiliary/tgsi/tgsi_ppc.c @@ -283,14 +283,14 @@ emit_fetch(struct gen_context *gen, const struct tgsi_full_src_register *reg, const unsigned chan_index) { - uint swizzle = tgsi_util_get_full_src_register_extswizzle(reg, chan_index); + uint swizzle = tgsi_util_get_full_src_register_swizzle(reg, chan_index); int dst_vec = -1; switch (swizzle) { - case TGSI_EXTSWIZZLE_X: - case TGSI_EXTSWIZZLE_Y: - case TGSI_EXTSWIZZLE_Z: - case TGSI_EXTSWIZZLE_W: + case TGSI_SWIZZLE_X: + case TGSI_SWIZZLE_Y: + case TGSI_SWIZZLE_Z: + case TGSI_SWIZZLE_W: switch (reg->SrcRegister.File) { case TGSI_FILE_INPUT: { @@ -349,16 +349,6 @@ emit_fetch(struct gen_context *gen, assert( 0 ); } break; - case TGSI_EXTSWIZZLE_ZERO: - ppc_vzero(gen->f, dst_vec); - break; - case TGSI_EXTSWIZZLE_ONE: - { - int one_vec = gen_one_vec(gen); - dst_vec = ppc_allocate_vec_register(gen->f); - ppc_vmove(gen->f, dst_vec, one_vec); - } - break; default: assert( 0 ); } @@ -418,8 +408,8 @@ equal_src_locs(const struct tgsi_full_src_register *a, uint chan_a, return FALSE; if (a->SrcRegister.Index != b->SrcRegister.Index) return FALSE; - swz_a = tgsi_util_get_full_src_register_extswizzle(a, chan_a); - swz_b = tgsi_util_get_full_src_register_extswizzle(b, chan_b); + swz_a = tgsi_util_get_full_src_register_swizzle(a, chan_a); + swz_b = tgsi_util_get_full_src_register_swizzle(b, chan_b); if (swz_a != swz_b) return FALSE; sign_a = tgsi_util_get_full_src_register_sign_mode(a, chan_a); @@ -635,7 +625,6 @@ emit_unaryop(struct gen_context *gen, struct tgsi_full_instruction *inst) ppc_vlogefp(gen->f, v1, v0); /* v1 = log2(v0) */ break; case TGSI_OPCODE_MOV: - case TGSI_OPCODE_SWZ: if (v0 != v1) ppc_vmove(gen->f, v1, v0); break; @@ -1119,7 +1108,6 @@ emit_instruction(struct gen_context *gen, switch (inst->Instruction.Opcode) { case TGSI_OPCODE_MOV: - case TGSI_OPCODE_SWZ: case TGSI_OPCODE_ABS: case TGSI_OPCODE_FLR: case TGSI_OPCODE_FRC: diff --git a/src/gallium/auxiliary/tgsi/tgsi_sanity.c b/src/gallium/auxiliary/tgsi/tgsi_sanity.c index 8a13885da9..36e27ea52f 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_sanity.c +++ b/src/gallium/auxiliary/tgsi/tgsi_sanity.c @@ -141,7 +141,8 @@ static const char *file_names[TGSI_FILE_COUNT] = "SAMP", "ADDR", "IMM", - "LOOP" + "LOOP", + "PRED" }; static boolean @@ -358,7 +359,7 @@ epilog( boolean tgsi_sanity_check( - struct tgsi_token *tokens ) + const struct tgsi_token *tokens ) { struct sanity_check_ctx ctx; diff --git a/src/gallium/auxiliary/tgsi/tgsi_sanity.h b/src/gallium/auxiliary/tgsi/tgsi_sanity.h index ca45e94c7a..52263ff883 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_sanity.h +++ b/src/gallium/auxiliary/tgsi/tgsi_sanity.h @@ -40,7 +40,7 @@ extern "C" { */ boolean tgsi_sanity_check( - struct tgsi_token *tokens ); + const struct tgsi_token *tokens ); #if defined __cplusplus } diff --git a/src/gallium/auxiliary/tgsi/tgsi_scan.c b/src/gallium/auxiliary/tgsi/tgsi_scan.c index c535788819..f9c16f1b6c 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_scan.c +++ b/src/gallium/auxiliary/tgsi/tgsi_scan.c @@ -132,6 +132,7 @@ tgsi_scan_shader(const struct tgsi_token *tokens, if (file == TGSI_FILE_INPUT) { info->input_semantic_name[reg] = (ubyte)fulldecl->Semantic.SemanticName; info->input_semantic_index[reg] = (ubyte)fulldecl->Semantic.SemanticIndex; + info->input_interpolate[reg] = (ubyte)fulldecl->Declaration.Interpolate; info->num_inputs++; } else if (file == TGSI_FILE_OUTPUT) { @@ -227,11 +228,6 @@ tgsi_is_passthrough_shader(const struct tgsi_token *tokens) src->SrcRegister.SwizzleZ != TGSI_SWIZZLE_Z || src->SrcRegister.SwizzleW != TGSI_SWIZZLE_W || - src->SrcRegisterExtSwz.ExtSwizzleX != TGSI_EXTSWIZZLE_X || - src->SrcRegisterExtSwz.ExtSwizzleY != TGSI_EXTSWIZZLE_Y || - src->SrcRegisterExtSwz.ExtSwizzleZ != TGSI_EXTSWIZZLE_Z || - src->SrcRegisterExtSwz.ExtSwizzleW != TGSI_EXTSWIZZLE_W || - dst->DstRegister.WriteMask != TGSI_WRITEMASK_XYZW) { tgsi_parse_free(&parse); diff --git a/src/gallium/auxiliary/tgsi/tgsi_scan.h b/src/gallium/auxiliary/tgsi/tgsi_scan.h index 2c1a75bc81..8a7ee0c7e4 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_scan.h +++ b/src/gallium/auxiliary/tgsi/tgsi_scan.h @@ -45,6 +45,7 @@ struct tgsi_shader_info ubyte num_outputs; ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS]; /**< TGSI_SEMANTIC_x */ ubyte input_semantic_index[PIPE_MAX_SHADER_INPUTS]; + ubyte input_interpolate[PIPE_MAX_SHADER_INPUTS]; ubyte output_semantic_name[PIPE_MAX_SHADER_OUTPUTS]; /**< TGSI_SEMANTIC_x */ ubyte output_semantic_index[PIPE_MAX_SHADER_OUTPUTS]; diff --git a/src/gallium/auxiliary/tgsi/tgsi_sse2.c b/src/gallium/auxiliary/tgsi/tgsi_sse2.c index 1e719940ec..a96fc94c7a 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_sse2.c +++ b/src/gallium/auxiliary/tgsi/tgsi_sse2.c @@ -39,8 +39,9 @@ #include "tgsi/tgsi_info.h" #include "tgsi/tgsi_parse.h" #include "tgsi/tgsi_util.h" -#include "tgsi_exec.h" -#include "tgsi_sse2.h" +#include "tgsi/tgsi_dump.h" +#include "tgsi/tgsi_exec.h" +#include "tgsi/tgsi_sse2.h" #include "rtasm/rtasm_x86sse.h" @@ -1259,13 +1260,13 @@ emit_fetch( const struct tgsi_full_src_register *reg, const unsigned chan_index ) { - unsigned swizzle = tgsi_util_get_full_src_register_extswizzle( reg, chan_index ); + unsigned swizzle = tgsi_util_get_full_src_register_swizzle( reg, chan_index ); switch (swizzle) { - case TGSI_EXTSWIZZLE_X: - case TGSI_EXTSWIZZLE_Y: - case TGSI_EXTSWIZZLE_Z: - case TGSI_EXTSWIZZLE_W: + case TGSI_SWIZZLE_X: + case TGSI_SWIZZLE_Y: + case TGSI_SWIZZLE_Z: + case TGSI_SWIZZLE_W: switch (reg->SrcRegister.File) { case TGSI_FILE_CONSTANT: emit_const( @@ -1307,22 +1308,6 @@ emit_fetch( } break; - case TGSI_EXTSWIZZLE_ZERO: - emit_tempf( - func, - xmm, - TGSI_EXEC_TEMP_00000000_I, - TGSI_EXEC_TEMP_00000000_C ); - break; - - case TGSI_EXTSWIZZLE_ONE: - emit_tempf( - func, - xmm, - TEMP_ONE_I, - TEMP_ONE_C ); - break; - default: assert( 0 ); } @@ -1360,6 +1345,32 @@ emit_store( const struct tgsi_full_instruction *inst, unsigned chan_index ) { + switch( inst->Instruction.Saturate ) { + case TGSI_SAT_NONE: + break; + + case TGSI_SAT_ZERO_ONE: + sse_maxps( + func, + make_xmm( xmm ), + get_temp( + TGSI_EXEC_TEMP_00000000_I, + TGSI_EXEC_TEMP_00000000_C ) ); + + sse_minps( + func, + make_xmm( xmm ), + get_temp( + TGSI_EXEC_TEMP_ONE_I, + TGSI_EXEC_TEMP_ONE_C ) ); + break; + + case TGSI_SAT_MINUS_PLUS_ONE: + assert( 0 ); + break; + } + + switch( reg->DstRegister.File ) { case TGSI_FILE_OUTPUT: emit_output( @@ -1388,19 +1399,6 @@ emit_store( default: assert( 0 ); } - - switch( inst->Instruction.Saturate ) { - case TGSI_SAT_NONE: - break; - - case TGSI_SAT_ZERO_ONE: - /* assert( 0 ); */ - break; - - case TGSI_SAT_MINUS_PLUS_ONE: - assert( 0 ); - break; - } } #define STORE( FUNC, INST, XMM, INDEX, CHAN )\ @@ -1568,13 +1566,13 @@ emit_kil( /* This mask stores component bits that were already tested. Note that * we test if the value is less than zero, so 1.0 and 0.0 need not to be * tested. */ - uniquemask = (1 << TGSI_EXTSWIZZLE_ZERO) | (1 << TGSI_EXTSWIZZLE_ONE); + uniquemask = 0; FOR_EACH_CHANNEL( chan_index ) { unsigned swizzle; /* unswizzle channel */ - swizzle = tgsi_util_get_full_src_register_extswizzle( + swizzle = tgsi_util_get_full_src_register_swizzle( reg, chan_index ); @@ -1747,14 +1745,6 @@ emit_instruction( if (indirect_temp_reference(inst)) return FALSE; - /* we don't handle saturation/clamping yet */ - if (inst->Instruction.Saturate != TGSI_SAT_NONE) - return FALSE; - - /* need to use extra temps to fix SOA dependencies : */ - if (tgsi_check_soa_dependencies(inst)) - return FALSE; - switch (inst->Instruction.Opcode) { case TGSI_OPCODE_ARL: FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) { @@ -1766,10 +1756,11 @@ emit_instruction( break; case TGSI_OPCODE_MOV: - case TGSI_OPCODE_SWZ: FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) { - FETCH( func, *inst, 0, 0, chan_index ); - STORE( func, *inst, 0, 0, chan_index ); + FETCH( func, *inst, 4 + chan_index, 0, chan_index ); + } + FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) { + STORE( func, *inst, 4 + chan_index, 0, chan_index ); } break; @@ -1847,7 +1838,6 @@ emit_instruction( break; case TGSI_OPCODE_RCP: - /* TGSI_OPCODE_RECIP */ FETCH( func, *inst, 0, 0, CHAN_X ); emit_rcp( func, 0, 0 ); FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) { @@ -1856,7 +1846,6 @@ emit_instruction( break; case TGSI_OPCODE_RSQ: - /* TGSI_OPCODE_RECIPSQRT */ FETCH( func, *inst, 0, 0, CHAN_X ); emit_abs( func, 0 ); emit_rsqrt( func, 1, 0 ); @@ -1954,7 +1943,6 @@ emit_instruction( break; case TGSI_OPCODE_DP3: - /* TGSI_OPCODE_DOT3 */ FETCH( func, *inst, 0, 0, CHAN_X ); FETCH( func, *inst, 1, 1, CHAN_X ); emit_mul( func, 0, 1 ); @@ -1972,7 +1960,6 @@ emit_instruction( break; case TGSI_OPCODE_DP4: - /* TGSI_OPCODE_DOT4 */ FETCH( func, *inst, 0, 0, CHAN_X ); FETCH( func, *inst, 1, 1, CHAN_X ); emit_mul( func, 0, 1 ); @@ -2043,17 +2030,14 @@ emit_instruction( break; case TGSI_OPCODE_SLT: - /* TGSI_OPCODE_SETLT */ emit_setcc( func, inst, cc_LessThan ); break; case TGSI_OPCODE_SGE: - /* TGSI_OPCODE_SETGE */ emit_setcc( func, inst, cc_NotLessThan ); break; case TGSI_OPCODE_MAD: - /* TGSI_OPCODE_MADD */ FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( func, *inst, 0, 0, chan_index ); FETCH( func, *inst, 1, 1, chan_index ); @@ -2283,7 +2267,7 @@ emit_instruction( break; case TGSI_OPCODE_SEQ: - return 0; + emit_setcc( func, inst, cc_Equal ); break; case TGSI_OPCODE_SFL: @@ -2291,7 +2275,7 @@ emit_instruction( break; case TGSI_OPCODE_SGT: - return 0; + emit_setcc( func, inst, cc_NotLessThanEqual ); break; case TGSI_OPCODE_SIN: @@ -2303,11 +2287,11 @@ emit_instruction( break; case TGSI_OPCODE_SLE: - return 0; + emit_setcc( func, inst, cc_LessThanEqual ); break; case TGSI_OPCODE_SNE: - return 0; + emit_setcc( func, inst, cc_NotEqual ); break; case TGSI_OPCODE_STR: @@ -2371,7 +2355,6 @@ emit_instruction( break; case TGSI_OPCODE_SSG: - /* TGSI_OPCODE_SGN */ FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) { FETCH( func, *inst, 0, 0, chan_index ); emit_sgn( func, 0, 0 ); @@ -2929,6 +2912,21 @@ tgsi_emit_sse2( parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_VERTEX ? "vertex shader" : "fragment shader"); } + + if (tgsi_check_soa_dependencies(&parse.FullToken.FullInstruction)) { + uint opcode = parse.FullToken.FullInstruction.Instruction.Opcode; + + /* XXX: we only handle src/dst aliasing in a few opcodes + * currently. Need to use an additional temporay to hold + * the result in the cases where the code is too opaque to + * fix. + */ + if (opcode != TGSI_OPCODE_MOV) { + debug_printf("Warning: src/dst aliasing in instruction" + " is not handled:\n"); + tgsi_dump_instruction(&parse.FullToken.FullInstruction, 1); + } + } break; case TGSI_TOKEN_TYPE_IMMEDIATE: diff --git a/src/gallium/auxiliary/tgsi/tgsi_text.c b/src/gallium/auxiliary/tgsi/tgsi_text.c index d438450b1e..d2b03ffb2f 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_text.c +++ b/src/gallium/auxiliary/tgsi/tgsi_text.c @@ -232,7 +232,8 @@ static const char *file_names[TGSI_FILE_COUNT] = "SAMP", "ADDR", "IMM", - "LOOP" + "LOOP", + "PRED" }; static boolean @@ -538,13 +539,11 @@ static boolean parse_optional_swizzle( struct translate_ctx *ctx, uint swizzle[4], - boolean *parsed_swizzle, - boolean *parsed_extswizzle ) + boolean *parsed_swizzle ) { const char *cur = ctx->cur; *parsed_swizzle = FALSE; - *parsed_extswizzle = FALSE; eat_opt_white( &cur ); if (*cur == '.') { @@ -562,15 +561,8 @@ parse_optional_swizzle( else if (uprcase( *cur ) == 'W') swizzle[i] = TGSI_SWIZZLE_W; else { - if (*cur == '0') - swizzle[i] = TGSI_EXTSWIZZLE_ZERO; - else if (*cur == '1') - swizzle[i] = TGSI_EXTSWIZZLE_ONE; - else { - report_error( ctx, "Expected register swizzle component `x', `y', `z', `w', `0' or `1'" ); - return FALSE; - } - *parsed_extswizzle = TRUE; + report_error( ctx, "Expected register swizzle component `x', `y', `z', `w', `0' or `1'" ); + return FALSE; } cur++; } @@ -595,7 +587,6 @@ parse_src_operand( uint swizzle[4]; boolean parsed_ext_negate_paren = FALSE; boolean parsed_swizzle; - boolean parsed_extswizzle; if (*ctx->cur == '-') { cur = ctx->cur; @@ -690,16 +681,8 @@ parse_src_operand( /* Parse optional swizzle. */ - if (parse_optional_swizzle( ctx, swizzle, &parsed_swizzle, &parsed_extswizzle )) { - if (parsed_extswizzle) { - assert( parsed_swizzle ); - - src->SrcRegisterExtSwz.ExtSwizzleX = swizzle[0]; - src->SrcRegisterExtSwz.ExtSwizzleY = swizzle[1]; - src->SrcRegisterExtSwz.ExtSwizzleZ = swizzle[2]; - src->SrcRegisterExtSwz.ExtSwizzleW = swizzle[3]; - } - else if (parsed_swizzle) { + if (parse_optional_swizzle( ctx, swizzle, &parsed_swizzle )) { + if (parsed_swizzle) { src->SrcRegister.SwizzleX = swizzle[0]; src->SrcRegister.SwizzleY = swizzle[1]; src->SrcRegister.SwizzleZ = swizzle[2]; diff --git a/src/gallium/auxiliary/tgsi/tgsi_ureg.c b/src/gallium/auxiliary/tgsi/tgsi_ureg.c index f7096bd8e2..3f752e9352 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_ureg.c +++ b/src/gallium/auxiliary/tgsi/tgsi_ureg.c @@ -29,8 +29,10 @@ #include "pipe/p_context.h" #include "pipe/p_state.h" #include "tgsi/tgsi_ureg.h" +#include "tgsi/tgsi_build.h" #include "tgsi/tgsi_info.h" #include "tgsi/tgsi_dump.h" +#include "tgsi/tgsi_sanity.h" #include "util/u_memory.h" #include "util/u_math.h" @@ -45,18 +47,14 @@ union tgsi_any_token { struct tgsi_immediate imm; union tgsi_immediate_data imm_data; struct tgsi_instruction insn; - struct tgsi_instruction_ext_nv insn_ext_nv; struct tgsi_instruction_ext_label insn_ext_label; struct tgsi_instruction_ext_texture insn_ext_texture; struct tgsi_instruction_ext_predicate insn_ext_predicate; struct tgsi_src_register src; - struct tgsi_src_register_ext_swz src_ext_swz; struct tgsi_src_register_ext_mod src_ext_mod; struct tgsi_dimension dim; struct tgsi_dst_register dst; - struct tgsi_dst_register_ext_concode dst_ext_code; struct tgsi_dst_register_ext_modulate dst_ext_mod; - struct tgsi_dst_register_ext_predicate dst_ext_pred; unsigned value; }; @@ -70,9 +68,11 @@ struct ureg_tokens { #define UREG_MAX_INPUT PIPE_MAX_ATTRIBS #define UREG_MAX_OUTPUT PIPE_MAX_ATTRIBS +#define UREG_MAX_CONSTANT_RANGE 32 #define UREG_MAX_IMMEDIATE 32 #define UREG_MAX_TEMP 256 #define UREG_MAX_ADDR 2 +#define UREG_MAX_PRED 1 #define DOMAIN_DECL 0 #define DOMAIN_INSN 1 @@ -86,8 +86,10 @@ struct ureg_program unsigned semantic_name; unsigned semantic_index; unsigned interp; - } input[UREG_MAX_INPUT]; - unsigned nr_inputs; + } fs_input[UREG_MAX_INPUT]; + unsigned nr_fs_inputs; + + unsigned vs_inputs[UREG_MAX_INPUT/32]; struct { unsigned semantic_name; @@ -107,9 +109,14 @@ struct ureg_program unsigned temps_active[UREG_MAX_TEMP / 32]; unsigned nr_temps; - unsigned nr_addrs; + struct { + unsigned first; + unsigned last; + } constant_range[UREG_MAX_CONSTANT_RANGE]; + unsigned nr_constant_ranges; - unsigned nr_constants; + unsigned nr_addrs; + unsigned nr_preds; unsigned nr_instructions; struct ureg_tokens domain[2]; @@ -119,6 +126,9 @@ static union tgsi_any_token error_tokens[32]; static void tokens_error( struct ureg_tokens *tokens ) { + if (tokens->tokens && tokens->tokens != error_tokens) + FREE(tokens->tokens); + tokens->tokens = error_tokens; tokens->size = Elements(error_tokens); tokens->count = 0; @@ -130,8 +140,9 @@ static void tokens_expand( struct ureg_tokens *tokens, { unsigned old_size = tokens->size * sizeof(unsigned); - if (tokens->tokens == error_tokens) - goto fail; + if (tokens->tokens == error_tokens) { + return; + } while (tokens->count + count > tokens->size) { tokens->size = (1 << ++tokens->order); @@ -140,13 +151,9 @@ static void tokens_expand( struct ureg_tokens *tokens, tokens->tokens = REALLOC(tokens->tokens, old_size, tokens->size * sizeof(unsigned)); - if (tokens->tokens == NULL) - goto fail; - - return; - -fail: - tokens_error(tokens); + if (tokens->tokens == NULL) { + tokens_error(tokens); + } } static void set_bad( struct ureg_program *ureg ) @@ -196,9 +203,13 @@ ureg_dst_register( unsigned file, dst.IndirectIndex = 0; dst.IndirectSwizzle = 0; dst.Saturate = 0; + dst.Predicate = 0; + dst.PredNegate = 0; + dst.PredSwizzleX = TGSI_SWIZZLE_X; + dst.PredSwizzleY = TGSI_SWIZZLE_Y; + dst.PredSwizzleZ = TGSI_SWIZZLE_Z; + dst.PredSwizzleW = TGSI_SWIZZLE_W; dst.Index = index; - dst.Pad1 = 0; - dst.Pad2 = 0; return dst; } @@ -228,25 +239,25 @@ ureg_src_register( unsigned file, -static struct ureg_src -ureg_DECL_input( struct ureg_program *ureg, - unsigned name, - unsigned index, - unsigned interp_mode ) +struct ureg_src +ureg_DECL_fs_input( struct ureg_program *ureg, + unsigned name, + unsigned index, + unsigned interp_mode ) { unsigned i; - for (i = 0; i < ureg->nr_inputs; i++) { - if (ureg->input[i].semantic_name == name && - ureg->input[i].semantic_index == index) + for (i = 0; i < ureg->nr_fs_inputs; i++) { + if (ureg->fs_input[i].semantic_name == name && + ureg->fs_input[i].semantic_index == index) goto out; } - if (ureg->nr_inputs < UREG_MAX_INPUT) { - ureg->input[i].semantic_name = name; - ureg->input[i].semantic_index = index; - ureg->input[i].interp = interp_mode; - ureg->nr_inputs++; + if (ureg->nr_fs_inputs < UREG_MAX_INPUT) { + ureg->fs_input[i].semantic_name = name; + ureg->fs_input[i].semantic_index = index; + ureg->fs_input[i].interp = interp_mode; + ureg->nr_fs_inputs++; } else { set_bad( ureg ); @@ -257,25 +268,14 @@ out: } - -struct ureg_src -ureg_DECL_fs_input( struct ureg_program *ureg, - unsigned name, - unsigned index, - unsigned interp ) -{ - assert(ureg->processor == TGSI_PROCESSOR_FRAGMENT); - return ureg_DECL_input( ureg, name, index, interp ); -} - - struct ureg_src ureg_DECL_vs_input( struct ureg_program *ureg, - unsigned name, unsigned index ) { assert(ureg->processor == TGSI_PROCESSOR_VERTEX); - return ureg_DECL_input( ureg, name, index, TGSI_INTERPOLATE_CONSTANT ); + + ureg->vs_inputs[index/32] |= 1 << (index % 32); + return ureg_src_register( TGSI_FILE_INPUT, index ); } @@ -313,9 +313,57 @@ out: * value or manage any constant_buffer contents -- that's the * resposibility of the calling code. */ -struct ureg_src ureg_DECL_constant(struct ureg_program *ureg ) +struct ureg_src ureg_DECL_constant(struct ureg_program *ureg, + unsigned index ) { - return ureg_src_register( TGSI_FILE_CONSTANT, ureg->nr_constants++ ); + unsigned minconst = index, maxconst = index; + unsigned i; + + /* Inside existing range? + */ + for (i = 0; i < ureg->nr_constant_ranges; i++) { + if (ureg->constant_range[i].first <= index && + ureg->constant_range[i].last >= index) + goto out; + } + + /* Extend existing range? + */ + for (i = 0; i < ureg->nr_constant_ranges; i++) { + if (ureg->constant_range[i].last == index - 1) { + ureg->constant_range[i].last = index; + goto out; + } + + if (ureg->constant_range[i].first == index + 1) { + ureg->constant_range[i].first = index; + goto out; + } + + minconst = MIN2(minconst, ureg->constant_range[i].first); + maxconst = MAX2(maxconst, ureg->constant_range[i].last); + } + + /* Create new range? + */ + if (ureg->nr_constant_ranges < UREG_MAX_CONSTANT_RANGE) { + i = ureg->nr_constant_ranges++; + ureg->constant_range[i].first = index; + ureg->constant_range[i].last = index; + } + + /* Collapse all ranges down to one: + */ + i = 0; + ureg->constant_range[0].first = minconst; + ureg->constant_range[0].last = maxconst; + ureg->nr_constant_ranges = 1; + +out: + assert(i < ureg->nr_constant_ranges); + assert(ureg->constant_range[i].first <= index); + assert(ureg->constant_range[i].last >= index); + return ureg_src_register( TGSI_FILE_CONSTANT, index ); } @@ -369,6 +417,19 @@ struct ureg_dst ureg_DECL_address( struct ureg_program *ureg ) return ureg_dst_register( TGSI_FILE_ADDRESS, 0 ); } +/* Allocate a new predicate register. + */ +struct ureg_dst +ureg_DECL_predicate(struct ureg_program *ureg) +{ + if (ureg->nr_preds < UREG_MAX_PRED) { + return ureg_dst_register(TGSI_FILE_PREDICATE, ureg->nr_preds++); + } + + assert(0); + return ureg_dst_register(TGSI_FILE_PREDICATE, 0); +} + /* Allocate a new sampler. */ struct ureg_src ureg_DECL_sampler( struct ureg_program *ureg, @@ -566,17 +627,40 @@ ureg_emit_dst( struct ureg_program *ureg, } +static void validate( unsigned opcode, + unsigned nr_dst, + unsigned nr_src ) +{ +#ifdef DEBUG + const struct tgsi_opcode_info *info = tgsi_get_opcode_info( opcode ); + assert(info); + if(info) { + assert(nr_dst == info->num_dst); + assert(nr_src == info->num_src); + } +#endif +} -unsigned +struct ureg_emit_insn_result ureg_emit_insn(struct ureg_program *ureg, unsigned opcode, boolean saturate, + boolean predicate, + boolean pred_negate, + unsigned pred_swizzle_x, + unsigned pred_swizzle_y, + unsigned pred_swizzle_z, + unsigned pred_swizzle_w, unsigned num_dst, unsigned num_src ) { union tgsi_any_token *out; + uint count = predicate ? 2 : 1; + struct ureg_emit_insn_result result; - out = get_tokens( ureg, DOMAIN_INSN, 1 ); + validate( opcode, num_dst, num_src ); + + out = get_tokens( ureg, DOMAIN_INSN, count ); out[0].value = 0; out[0].insn.Type = TGSI_TOKEN_TYPE_INSTRUCTION; out[0].insn.NrTokens = 0; @@ -585,17 +669,34 @@ ureg_emit_insn(struct ureg_program *ureg, out[0].insn.NumDstRegs = num_dst; out[0].insn.NumSrcRegs = num_src; out[0].insn.Padding = 0; - out[0].insn.Extended = 0; - + + result.insn_token = ureg->domain[DOMAIN_INSN].count - count; + + if (predicate) { + out[0].insn.Extended = 1; + out[1].insn_ext_predicate = tgsi_default_instruction_ext_predicate(); + out[1].insn_ext_predicate.Negate = pred_negate; + out[1].insn_ext_predicate.SwizzleX = pred_swizzle_x; + out[1].insn_ext_predicate.SwizzleY = pred_swizzle_y; + out[1].insn_ext_predicate.SwizzleZ = pred_swizzle_z; + out[1].insn_ext_predicate.SwizzleW = pred_swizzle_w; + + result.extended_token = result.insn_token + 1; + } else { + out[0].insn.Extended = 0; + + result.extended_token = result.insn_token; + } + ureg->nr_instructions++; - return ureg->domain[DOMAIN_INSN].count - 1; + return result; } void ureg_emit_label(struct ureg_program *ureg, - unsigned insn_token, + unsigned extended_token, unsigned *label_token ) { union tgsi_any_token *out, *insn; @@ -604,9 +705,9 @@ ureg_emit_label(struct ureg_program *ureg, return; out = get_tokens( ureg, DOMAIN_INSN, 1 ); - insn = retrieve_token( ureg, DOMAIN_INSN, insn_token ); + insn = retrieve_token( ureg, DOMAIN_INSN, extended_token ); - insn->insn.Extended = 1; + insn->token.Extended = 1; out[0].value = 0; out[0].insn_ext_label.Type = TGSI_INSTRUCTION_EXT_TYPE_LABEL; @@ -640,15 +741,15 @@ ureg_fixup_label(struct ureg_program *ureg, void ureg_emit_texture(struct ureg_program *ureg, - unsigned insn_token, + unsigned extended_token, unsigned target ) { union tgsi_any_token *out, *insn; out = get_tokens( ureg, DOMAIN_INSN, 1 ); - insn = retrieve_token( ureg, DOMAIN_INSN, insn_token ); + insn = retrieve_token( ureg, DOMAIN_INSN, extended_token ); - insn->insn.Extended = 1; + insn->token.Extended = 1; out[0].value = 0; out[0].insn_ext_texture.Type = TGSI_INSTRUCTION_EXT_TYPE_TEXTURE; @@ -675,23 +776,83 @@ ureg_insn(struct ureg_program *ureg, const struct ureg_src *src, unsigned nr_src ) { - unsigned insn, i; + struct ureg_emit_insn_result insn; + unsigned i; boolean saturate; + boolean predicate; + boolean negate; + unsigned swizzle[4]; -#ifdef DEBUG - { - const struct tgsi_opcode_info *info = tgsi_get_opcode_info( opcode ); - assert(info); - if(info) { - assert(nr_dst == info->num_dst); - assert(nr_src == info->num_src); - } + saturate = nr_dst ? dst[0].Saturate : FALSE; + predicate = nr_dst ? dst[0].Predicate : FALSE; + if (predicate) { + negate = dst[0].PredNegate; + swizzle[0] = dst[0].PredSwizzleX; + swizzle[1] = dst[0].PredSwizzleY; + swizzle[2] = dst[0].PredSwizzleZ; + swizzle[3] = dst[0].PredSwizzleW; } -#endif - + + insn = ureg_emit_insn(ureg, + opcode, + saturate, + predicate, + negate, + swizzle[0], + swizzle[1], + swizzle[2], + swizzle[3], + nr_dst, + nr_src); + + for (i = 0; i < nr_dst; i++) + ureg_emit_dst( ureg, dst[i] ); + + for (i = 0; i < nr_src; i++) + ureg_emit_src( ureg, src[i] ); + + ureg_fixup_insn_size( ureg, insn.insn_token ); +} + +void +ureg_tex_insn(struct ureg_program *ureg, + unsigned opcode, + const struct ureg_dst *dst, + unsigned nr_dst, + unsigned target, + const struct ureg_src *src, + unsigned nr_src ) +{ + struct ureg_emit_insn_result insn; + unsigned i; + boolean saturate; + boolean predicate; + boolean negate; + unsigned swizzle[4]; + saturate = nr_dst ? dst[0].Saturate : FALSE; + predicate = nr_dst ? dst[0].Predicate : FALSE; + if (predicate) { + negate = dst[0].PredNegate; + swizzle[0] = dst[0].PredSwizzleX; + swizzle[1] = dst[0].PredSwizzleY; + swizzle[2] = dst[0].PredSwizzleZ; + swizzle[3] = dst[0].PredSwizzleW; + } + + insn = ureg_emit_insn(ureg, + opcode, + saturate, + predicate, + negate, + swizzle[0], + swizzle[1], + swizzle[2], + swizzle[3], + nr_dst, + nr_src); - insn = ureg_emit_insn( ureg, opcode, saturate, nr_dst, nr_src ); + ureg_emit_texture( ureg, insn.extended_token, target ); for (i = 0; i < nr_dst; i++) ureg_emit_dst( ureg, dst[i] ); @@ -699,7 +860,38 @@ ureg_insn(struct ureg_program *ureg, for (i = 0; i < nr_src; i++) ureg_emit_src( ureg, src[i] ); - ureg_fixup_insn_size( ureg, insn ); + ureg_fixup_insn_size( ureg, insn.insn_token ); +} + + +void +ureg_label_insn(struct ureg_program *ureg, + unsigned opcode, + const struct ureg_src *src, + unsigned nr_src, + unsigned *label_token ) +{ + struct ureg_emit_insn_result insn; + unsigned i; + + insn = ureg_emit_insn(ureg, + opcode, + FALSE, + FALSE, + FALSE, + TGSI_SWIZZLE_X, + TGSI_SWIZZLE_Y, + TGSI_SWIZZLE_Z, + TGSI_SWIZZLE_W, + 0, + nr_src); + + ureg_emit_label( ureg, insn.extended_token, label_token ); + + for (i = 0; i < nr_src; i++) + ureg_emit_src( ureg, src[i] ); + + ureg_fixup_insn_size( ureg, insn.insn_token ); } @@ -777,13 +969,22 @@ static void emit_decls( struct ureg_program *ureg ) { unsigned i; - for (i = 0; i < ureg->nr_inputs; i++) { - emit_decl( ureg, - TGSI_FILE_INPUT, - i, - ureg->input[i].semantic_name, - ureg->input[i].semantic_index, - ureg->input[i].interp ); + if (ureg->processor == TGSI_PROCESSOR_VERTEX) { + for (i = 0; i < UREG_MAX_INPUT; i++) { + if (ureg->vs_inputs[i/32] & (1 << (i%32))) { + emit_decl_range( ureg, TGSI_FILE_INPUT, i, 1 ); + } + } + } + else { + for (i = 0; i < ureg->nr_fs_inputs; i++) { + emit_decl( ureg, + TGSI_FILE_INPUT, + i, + ureg->fs_input[i].semantic_name, + ureg->fs_input[i].semantic_index, + ureg->fs_input[i].interp ); + } } for (i = 0; i < ureg->nr_outputs; i++) { @@ -801,10 +1002,13 @@ static void emit_decls( struct ureg_program *ureg ) ureg->sampler[i].Index, 1 ); } - if (ureg->nr_constants) { - emit_decl_range( ureg, - TGSI_FILE_CONSTANT, - 0, ureg->nr_constants ); + if (ureg->nr_constant_ranges) { + for (i = 0; i < ureg->nr_constant_ranges; i++) + emit_decl_range( ureg, + TGSI_FILE_CONSTANT, + ureg->constant_range[i].first, + (ureg->constant_range[i].last + 1 - + ureg->constant_range[i].first) ); } if (ureg->nr_temps) { @@ -819,6 +1023,13 @@ static void emit_decls( struct ureg_program *ureg ) 0, ureg->nr_addrs ); } + if (ureg->nr_preds) { + emit_decl_range(ureg, + TGSI_FILE_PREDICATE, + 0, + ureg->nr_preds); + } + for (i = 0; i < ureg->nr_immediates; i++) { emit_immediate( ureg, ureg->immediate[i].v ); @@ -890,6 +1101,15 @@ const struct tgsi_token *ureg_finalize( struct ureg_program *ureg ) ureg->domain[DOMAIN_DECL].count); tgsi_dump( tokens, 0 ); } + +#if DEBUG + if (tokens && !tgsi_sanity_check(tokens)) { + debug_printf("tgsi_ureg.c, sanity check failed on generated tokens:\n"); + tgsi_dump(tokens, 0); + assert(0); + } +#endif + return tokens; } @@ -911,6 +1131,25 @@ void *ureg_create_shader( struct ureg_program *ureg, } +const struct tgsi_token *ureg_get_tokens( struct ureg_program *ureg, + unsigned *nr_tokens ) +{ + const struct tgsi_token *tokens; + + ureg_finalize(ureg); + + tokens = &ureg->domain[DOMAIN_DECL].tokens[0].token; + + if (nr_tokens) + *nr_tokens = ureg->domain[DOMAIN_DECL].size; + + ureg->domain[DOMAIN_DECL].tokens = 0; + ureg->domain[DOMAIN_DECL].size = 0; + ureg->domain[DOMAIN_DECL].order = 0; + ureg->domain[DOMAIN_DECL].count = 0; + + return tokens; +} struct ureg_program *ureg_create( unsigned processor ) diff --git a/src/gallium/auxiliary/tgsi/tgsi_ureg.h b/src/gallium/auxiliary/tgsi/tgsi_ureg.h index acbca59040..dae4291194 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_ureg.h +++ b/src/gallium/auxiliary/tgsi/tgsi_ureg.h @@ -67,9 +67,13 @@ struct ureg_dst unsigned WriteMask : 4; /* TGSI_WRITEMASK_ */ unsigned Indirect : 1; /* BOOL */ unsigned Saturate : 1; /* BOOL */ + unsigned Predicate : 1; + unsigned PredNegate : 1; /* BOOL */ + unsigned PredSwizzleX: 2; /* TGSI_SWIZZLE_ */ + unsigned PredSwizzleY: 2; /* TGSI_SWIZZLE_ */ + unsigned PredSwizzleZ: 2; /* TGSI_SWIZZLE_ */ + unsigned PredSwizzleW: 2; /* TGSI_SWIZZLE_ */ int Index : 16; /* SINT */ - unsigned Pad1 : 5; - unsigned Pad2 : 1; /* BOOL */ int IndirectIndex : 16; /* SINT */ int IndirectSwizzle : 2; /* TGSI_SWIZZLE_ */ }; @@ -82,10 +86,21 @@ ureg_create( unsigned processor ); const struct tgsi_token * ureg_finalize( struct ureg_program * ); +/* Create and return a shader: + */ void * ureg_create_shader( struct ureg_program *, struct pipe_context *pipe ); + +/* Alternately, return the built token stream and hand ownership of + * that memory to the caller: + */ +const struct tgsi_token * +ureg_get_tokens( struct ureg_program *ureg, + unsigned *nr_tokens ); + + void ureg_destroy( struct ureg_program * ); @@ -116,8 +131,7 @@ ureg_DECL_fs_input( struct ureg_program *, struct ureg_src ureg_DECL_vs_input( struct ureg_program *, - unsigned semantic_name, - unsigned semantic_index ); + unsigned index ); struct ureg_dst ureg_DECL_output( struct ureg_program *, @@ -130,7 +144,8 @@ ureg_DECL_immediate( struct ureg_program *, unsigned nr ); struct ureg_src -ureg_DECL_constant( struct ureg_program * ); +ureg_DECL_constant( struct ureg_program *, + unsigned index ); struct ureg_dst ureg_DECL_temporary( struct ureg_program * ); @@ -142,6 +157,9 @@ ureg_release_temporary( struct ureg_program *ureg, struct ureg_dst ureg_DECL_address( struct ureg_program * ); +struct ureg_dst +ureg_DECL_predicate(struct ureg_program *); + /* Supply an index to the sampler declaration as this is the hook to * the external pipe_sampler state. Users of this function probably * don't want just any sampler, but a specific one which they've set @@ -233,14 +251,43 @@ ureg_insn(struct ureg_program *ureg, unsigned nr_src ); +void +ureg_tex_insn(struct ureg_program *ureg, + unsigned opcode, + const struct ureg_dst *dst, + unsigned nr_dst, + unsigned target, + const struct ureg_src *src, + unsigned nr_src ); + + +void +ureg_label_insn(struct ureg_program *ureg, + unsigned opcode, + const struct ureg_src *src, + unsigned nr_src, + unsigned *label); + + /*********************************************************************** * Internal instruction helpers, don't call these directly: */ -unsigned +struct ureg_emit_insn_result { + unsigned insn_token; /*< Used to fixup insn size. */ + unsigned extended_token; /*< Used to set the Extended bit, usually the same as insn_token. */ +}; + +struct ureg_emit_insn_result ureg_emit_insn(struct ureg_program *ureg, unsigned opcode, boolean saturate, + boolean predicate, + boolean pred_negate, + unsigned pred_swizzle_x, + unsigned pred_swizzle_y, + unsigned pred_swizzle_z, + unsigned pred_swizzle_w, unsigned num_dst, unsigned num_src ); @@ -271,7 +318,17 @@ ureg_fixup_insn_size(struct ureg_program *ureg, static INLINE void ureg_##op( struct ureg_program *ureg ) \ { \ unsigned opcode = TGSI_OPCODE_##op; \ - unsigned insn = ureg_emit_insn( ureg, opcode, FALSE, 0, 0 ); \ + unsigned insn = ureg_emit_insn(ureg, \ + opcode, \ + FALSE, \ + FALSE, \ + FALSE, \ + TGSI_SWIZZLE_X, \ + TGSI_SWIZZLE_Y, \ + TGSI_SWIZZLE_Z, \ + TGSI_SWIZZLE_W, \ + 0, \ + 0).insn_token; \ ureg_fixup_insn_size( ureg, insn ); \ } @@ -280,7 +337,17 @@ static INLINE void ureg_##op( struct ureg_program *ureg, \ struct ureg_src src ) \ { \ unsigned opcode = TGSI_OPCODE_##op; \ - unsigned insn = ureg_emit_insn( ureg, opcode, FALSE, 0, 1 ); \ + unsigned insn = ureg_emit_insn(ureg, \ + opcode, \ + FALSE, \ + FALSE, \ + FALSE, \ + TGSI_SWIZZLE_X, \ + TGSI_SWIZZLE_Y, \ + TGSI_SWIZZLE_Z, \ + TGSI_SWIZZLE_W, \ + 0, \ + 1).insn_token; \ ureg_emit_src( ureg, src ); \ ureg_fixup_insn_size( ureg, insn ); \ } @@ -290,9 +357,20 @@ static INLINE void ureg_##op( struct ureg_program *ureg, \ unsigned *label_token ) \ { \ unsigned opcode = TGSI_OPCODE_##op; \ - unsigned insn = ureg_emit_insn( ureg, opcode, FALSE, 0, 0 ); \ - ureg_emit_label( ureg, insn, label_token ); \ - ureg_fixup_insn_size( ureg, insn ); \ + struct ureg_emit_insn_result insn; \ + insn = ureg_emit_insn(ureg, \ + opcode, \ + FALSE, \ + FALSE, \ + FALSE, \ + TGSI_SWIZZLE_X, \ + TGSI_SWIZZLE_Y, \ + TGSI_SWIZZLE_Z, \ + TGSI_SWIZZLE_W, \ + 0, \ + 0); \ + ureg_emit_label( ureg, insn.extended_token, label_token ); \ + ureg_fixup_insn_size( ureg, insn.insn_token ); \ } #define OP01_LBL( op ) \ @@ -301,10 +379,21 @@ static INLINE void ureg_##op( struct ureg_program *ureg, \ unsigned *label_token ) \ { \ unsigned opcode = TGSI_OPCODE_##op; \ - unsigned insn = ureg_emit_insn( ureg, opcode, FALSE, 0, 1 ); \ - ureg_emit_label( ureg, insn, label_token ); \ + struct ureg_emit_insn_result insn; \ + insn = ureg_emit_insn(ureg, \ + opcode, \ + FALSE, \ + FALSE, \ + FALSE, \ + TGSI_SWIZZLE_X, \ + TGSI_SWIZZLE_Y, \ + TGSI_SWIZZLE_Z, \ + TGSI_SWIZZLE_W, \ + 0, \ + 1); \ + ureg_emit_label( ureg, insn.extended_token, label_token ); \ ureg_emit_src( ureg, src ); \ - ureg_fixup_insn_size( ureg, insn ); \ + ureg_fixup_insn_size( ureg, insn.insn_token ); \ } #define OP10( op ) \ @@ -312,7 +401,17 @@ static INLINE void ureg_##op( struct ureg_program *ureg, \ struct ureg_dst dst ) \ { \ unsigned opcode = TGSI_OPCODE_##op; \ - unsigned insn = ureg_emit_insn( ureg, opcode, dst.Saturate, 1, 0 ); \ + unsigned insn = ureg_emit_insn(ureg, \ + opcode, \ + dst.Saturate, \ + dst.Predicate, \ + dst.PredNegate, \ + dst.PredSwizzleX, \ + dst.PredSwizzleY, \ + dst.PredSwizzleZ, \ + dst.PredSwizzleW, \ + 1, \ + 0).insn_token; \ ureg_emit_dst( ureg, dst ); \ ureg_fixup_insn_size( ureg, insn ); \ } @@ -324,7 +423,17 @@ static INLINE void ureg_##op( struct ureg_program *ureg, \ struct ureg_src src ) \ { \ unsigned opcode = TGSI_OPCODE_##op; \ - unsigned insn = ureg_emit_insn( ureg, opcode, dst.Saturate, 1, 1 ); \ + unsigned insn = ureg_emit_insn(ureg, \ + opcode, \ + dst.Saturate, \ + dst.Predicate, \ + dst.PredNegate, \ + dst.PredSwizzleX, \ + dst.PredSwizzleY, \ + dst.PredSwizzleZ, \ + dst.PredSwizzleW, \ + 1, \ + 1).insn_token; \ ureg_emit_dst( ureg, dst ); \ ureg_emit_src( ureg, src ); \ ureg_fixup_insn_size( ureg, insn ); \ @@ -337,7 +446,17 @@ static INLINE void ureg_##op( struct ureg_program *ureg, \ struct ureg_src src1 ) \ { \ unsigned opcode = TGSI_OPCODE_##op; \ - unsigned insn = ureg_emit_insn( ureg, opcode, dst.Saturate, 1, 2 ); \ + unsigned insn = ureg_emit_insn(ureg, \ + opcode, \ + dst.Saturate, \ + dst.Predicate, \ + dst.PredNegate, \ + dst.PredSwizzleX, \ + dst.PredSwizzleY, \ + dst.PredSwizzleZ, \ + dst.PredSwizzleW, \ + 1, \ + 2).insn_token; \ ureg_emit_dst( ureg, dst ); \ ureg_emit_src( ureg, src0 ); \ ureg_emit_src( ureg, src1 ); \ @@ -352,12 +471,23 @@ static INLINE void ureg_##op( struct ureg_program *ureg, \ struct ureg_src src1 ) \ { \ unsigned opcode = TGSI_OPCODE_##op; \ - unsigned insn = ureg_emit_insn( ureg, opcode, dst.Saturate, 1, 2 ); \ - ureg_emit_texture( ureg, insn, target ); \ + struct ureg_emit_insn_result insn; \ + insn = ureg_emit_insn(ureg, \ + opcode, \ + dst.Saturate, \ + dst.Predicate, \ + dst.PredNegate, \ + dst.PredSwizzleX, \ + dst.PredSwizzleY, \ + dst.PredSwizzleZ, \ + dst.PredSwizzleW, \ + 1, \ + 2); \ + ureg_emit_texture( ureg, insn.extended_token, target ); \ ureg_emit_dst( ureg, dst ); \ ureg_emit_src( ureg, src0 ); \ ureg_emit_src( ureg, src1 ); \ - ureg_fixup_insn_size( ureg, insn ); \ + ureg_fixup_insn_size( ureg, insn.insn_token ); \ } #define OP13( op ) \ @@ -368,7 +498,17 @@ static INLINE void ureg_##op( struct ureg_program *ureg, \ struct ureg_src src2 ) \ { \ unsigned opcode = TGSI_OPCODE_##op; \ - unsigned insn = ureg_emit_insn( ureg, opcode, dst.Saturate, 1, 3 ); \ + unsigned insn = ureg_emit_insn(ureg, \ + opcode, \ + dst.Saturate, \ + dst.Predicate, \ + dst.PredNegate, \ + dst.PredSwizzleX, \ + dst.PredSwizzleY, \ + dst.PredSwizzleZ, \ + dst.PredSwizzleW, \ + 1, \ + 3).insn_token; \ ureg_emit_dst( ureg, dst ); \ ureg_emit_src( ureg, src0 ); \ ureg_emit_src( ureg, src1 ); \ @@ -386,14 +526,25 @@ static INLINE void ureg_##op( struct ureg_program *ureg, \ struct ureg_src src3 ) \ { \ unsigned opcode = TGSI_OPCODE_##op; \ - unsigned insn = ureg_emit_insn( ureg, opcode, dst.Saturate, 1, 4 ); \ - ureg_emit_texture( ureg, insn, target ); \ + struct ureg_emit_insn_result insn; \ + insn = ureg_emit_insn(ureg, \ + opcode, \ + dst.Saturate, \ + dst.Predicate, \ + dst.PredNegate, \ + dst.PredSwizzleX, \ + dst.PredSwizzleY, \ + dst.PredSwizzleZ, \ + dst.PredSwizzleW, \ + 1, \ + 4); \ + ureg_emit_texture( ureg, insn.extended_token, target ); \ ureg_emit_dst( ureg, dst ); \ ureg_emit_src( ureg, src0 ); \ ureg_emit_src( ureg, src1 ); \ ureg_emit_src( ureg, src2 ); \ ureg_emit_src( ureg, src3 ); \ - ureg_fixup_insn_size( ureg, insn ); \ + ureg_fixup_insn_size( ureg, insn.insn_token ); \ } @@ -468,6 +619,24 @@ ureg_saturate( struct ureg_dst reg ) return reg; } +static INLINE struct ureg_dst +ureg_predicate(struct ureg_dst reg, + boolean negate, + unsigned swizzle_x, + unsigned swizzle_y, + unsigned swizzle_z, + unsigned swizzle_w) +{ + assert(reg.File != TGSI_FILE_NULL); + reg.Predicate = 1; + reg.PredNegate = negate; + reg.PredSwizzleX = swizzle_x; + reg.PredSwizzleY = swizzle_y; + reg.PredSwizzleZ = swizzle_z; + reg.PredSwizzleW = swizzle_w; + return reg; +} + static INLINE struct ureg_dst ureg_dst_indirect( struct ureg_dst reg, struct ureg_src addr ) { @@ -501,9 +670,13 @@ ureg_dst( struct ureg_src src ) dst.IndirectIndex = src.IndirectIndex; dst.IndirectSwizzle = src.IndirectSwizzle; dst.Saturate = 0; + dst.Predicate = 0; + dst.PredNegate = 0; + dst.PredSwizzleX = TGSI_SWIZZLE_X; + dst.PredSwizzleY = TGSI_SWIZZLE_Y; + dst.PredSwizzleZ = TGSI_SWIZZLE_Z; + dst.PredSwizzleW = TGSI_SWIZZLE_W; dst.Index = src.Index; - dst.Pad1 = 0; - dst.Pad2 = 0; return dst; } @@ -542,9 +715,13 @@ ureg_dst_undef( void ) dst.IndirectIndex = 0; dst.IndirectSwizzle = 0; dst.Saturate = 0; + dst.Predicate = 0; + dst.PredNegate = 0; + dst.PredSwizzleX = TGSI_SWIZZLE_X; + dst.PredSwizzleY = TGSI_SWIZZLE_Y; + dst.PredSwizzleZ = TGSI_SWIZZLE_Z; + dst.PredSwizzleW = TGSI_SWIZZLE_W; dst.Index = 0; - dst.Pad1 = 0; - dst.Pad2 = 0; return dst; } diff --git a/src/gallium/auxiliary/tgsi/tgsi_util.c b/src/gallium/auxiliary/tgsi/tgsi_util.c index 71f8a6ca40..4dee1be9e8 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_util.c +++ b/src/gallium/auxiliary/tgsi/tgsi_util.c @@ -69,59 +69,15 @@ tgsi_util_get_src_register_swizzle( return 0; } -unsigned -tgsi_util_get_src_register_extswizzle( - const struct tgsi_src_register_ext_swz *reg, - unsigned component ) -{ - switch( component ) { - case 0: - return reg->ExtSwizzleX; - case 1: - return reg->ExtSwizzleY; - case 2: - return reg->ExtSwizzleZ; - case 3: - return reg->ExtSwizzleW; - default: - assert( 0 ); - } - return 0; -} unsigned -tgsi_util_get_full_src_register_extswizzle( +tgsi_util_get_full_src_register_swizzle( const struct tgsi_full_src_register *reg, unsigned component ) { - unsigned swizzle; - - /* - * First, calculate the extended swizzle for a given channel. This will give - * us either a channel index into the simple swizzle or a constant 1 or 0. - */ - swizzle = tgsi_util_get_src_register_extswizzle( - ®->SrcRegisterExtSwz, + return tgsi_util_get_src_register_swizzle( + ®->SrcRegister, component ); - - assert (TGSI_SWIZZLE_X == TGSI_EXTSWIZZLE_X); - assert (TGSI_SWIZZLE_Y == TGSI_EXTSWIZZLE_Y); - assert (TGSI_SWIZZLE_Z == TGSI_EXTSWIZZLE_Z); - assert (TGSI_SWIZZLE_W == TGSI_EXTSWIZZLE_W); - assert (TGSI_EXTSWIZZLE_ZERO > TGSI_SWIZZLE_W); - assert (TGSI_EXTSWIZZLE_ONE > TGSI_SWIZZLE_W); - - /* - * Second, calculate the simple swizzle for the unswizzled channel index. - * Leave the constants intact, they are not affected by the simple swizzle. - */ - if( swizzle <= TGSI_SWIZZLE_W ) { - swizzle = tgsi_util_get_src_register_swizzle( - ®->SrcRegister, - swizzle ); - } - - return swizzle; } void @@ -148,74 +104,6 @@ tgsi_util_set_src_register_swizzle( } } -void -tgsi_util_set_src_register_extswizzle( - struct tgsi_src_register_ext_swz *reg, - unsigned swizzle, - unsigned component ) -{ - switch( component ) { - case 0: - reg->ExtSwizzleX = swizzle; - break; - case 1: - reg->ExtSwizzleY = swizzle; - break; - case 2: - reg->ExtSwizzleZ = swizzle; - break; - case 3: - reg->ExtSwizzleW = swizzle; - break; - default: - assert( 0 ); - } -} - -unsigned -tgsi_util_get_src_register_extnegate( - const struct tgsi_src_register_ext_swz *reg, - unsigned component ) -{ - switch( component ) { - case 0: - return reg->NegateX; - case 1: - return reg->NegateY; - case 2: - return reg->NegateZ; - case 3: - return reg->NegateW; - default: - assert( 0 ); - } - return 0; -} - -void -tgsi_util_set_src_register_extnegate( - struct tgsi_src_register_ext_swz *reg, - unsigned negate, - unsigned component ) -{ - switch( component ) { - case 0: - reg->NegateX = negate; - break; - case 1: - reg->NegateY = negate; - break; - case 2: - reg->NegateZ = negate; - break; - case 3: - reg->NegateW = negate; - break; - default: - assert( 0 ); - } -} - unsigned tgsi_util_get_full_src_register_sign_mode( const struct tgsi_full_src_register *reg, @@ -239,9 +127,7 @@ tgsi_util_get_full_src_register_sign_mode( unsigned negate; negate = reg->SrcRegister.Negate; - if( tgsi_util_get_src_register_extnegate( ®->SrcRegisterExtSwz, component ) ) { - negate = !negate; - } + if( reg->SrcRegisterExtMod.Negate ) { negate = !negate; } @@ -262,11 +148,6 @@ tgsi_util_set_full_src_register_sign_mode( struct tgsi_full_src_register *reg, unsigned sign_mode ) { - reg->SrcRegisterExtSwz.NegateX = 0; - reg->SrcRegisterExtSwz.NegateY = 0; - reg->SrcRegisterExtSwz.NegateZ = 0; - reg->SrcRegisterExtSwz.NegateW = 0; - switch (sign_mode) { case TGSI_UTIL_SIGN_CLEAR: diff --git a/src/gallium/auxiliary/tgsi/tgsi_util.h b/src/gallium/auxiliary/tgsi/tgsi_util.h index 21eb656327..19ee2e7cf2 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_util.h +++ b/src/gallium/auxiliary/tgsi/tgsi_util.h @@ -33,7 +33,6 @@ extern "C" { #endif struct tgsi_src_register; -struct tgsi_src_register_ext_swz; struct tgsi_full_src_register; void * @@ -45,13 +44,9 @@ tgsi_util_get_src_register_swizzle( const struct tgsi_src_register *reg, unsigned component ); -unsigned -tgsi_util_get_src_register_extswizzle( - const struct tgsi_src_register_ext_swz *reg, - unsigned component); unsigned -tgsi_util_get_full_src_register_extswizzle( +tgsi_util_get_full_src_register_swizzle( const struct tgsi_full_src_register *reg, unsigned component ); @@ -61,23 +56,6 @@ tgsi_util_set_src_register_swizzle( unsigned swizzle, unsigned component ); -void -tgsi_util_set_src_register_extswizzle( - struct tgsi_src_register_ext_swz *reg, - unsigned swizzle, - unsigned component ); - -unsigned -tgsi_util_get_src_register_extnegate( - const struct tgsi_src_register_ext_swz *reg, - unsigned component ); - -void -tgsi_util_set_src_register_extnegate( - struct tgsi_src_register_ext_swz *reg, - unsigned negate, - unsigned component ); - #define TGSI_UTIL_SIGN_CLEAR 0 /* Force positive */ #define TGSI_UTIL_SIGN_SET 1 /* Force negative */ #define TGSI_UTIL_SIGN_TOGGLE 2 /* Negate */ |