2 * Copyright (C) 2008 Nicolai Haehnle.
6 * Permission is hereby granted, free of charge, to any person obtaining
7 * a copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sublicense, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial
16 * portions of the Software.
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21 * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22 * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23 * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24 * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
28 #ifndef __RADEON_PROGRAM_H_
29 #define __RADEON_PROGRAM_H_
34 #include "radeon_opcodes.h"
35 #include "radeon_code.h"
36 #include "radeon_program_constants.h"
37 #include "radeon_program_pair.h"
39 struct radeon_compiler
;
41 struct rc_src_register
{
44 /** Negative values may be used for relative addressing. */
45 signed int Index
:(RC_REGISTER_INDEX_BITS
+1);
46 unsigned int RelAddr
:1;
48 unsigned int Swizzle
:12;
50 /** Take the component-wise absolute value */
53 /** Post-Abs negation. */
54 unsigned int Negate
:4;
57 struct rc_dst_register
{
60 /** Negative values may be used for relative addressing. */
61 signed int Index
:(RC_REGISTER_INDEX_BITS
+1);
62 unsigned int RelAddr
:1;
64 unsigned int WriteMask
:4;
68 * Instructions are maintained by the compiler in a doubly linked list
69 * of these structures.
71 * This instruction format is intended to be expanded for hardware-specific
72 * trickery. At different stages of compilation, a different set of
73 * instruction types may be valid.
75 struct rc_sub_instruction
{
76 struct rc_src_register SrcReg
[3];
77 struct rc_dst_register DstReg
;
80 * Opcode of this instruction, according to \ref rc_opcode enums.
82 unsigned int Opcode
:8;
85 * Saturate each value of the result to the range [0,1] or [-1,1],
86 * according to \ref rc_saturate_mode enums.
88 unsigned int SaturateMode
:2;
91 * Writing to the special register RC_SPECIAL_ALU_RESULT
94 unsigned int WriteALUResult
:2;
95 unsigned int ALUResultCompare
:3;
99 * \name Extra fields for TEX, TXB, TXD, TXL, TXP instructions.
102 /** Source texture unit. */
103 unsigned int TexSrcUnit
:5;
105 /** Source texture target, one of the \ref rc_texture_target enums */
106 unsigned int TexSrcTarget
:3;
108 /** True if tex instruction should do shadow comparison */
109 unsigned int TexShadow
:1;
114 RC_INSTRUCTION_NORMAL
= 0,
116 } rc_instruction_type
;
118 struct rc_instruction
{
119 struct rc_instruction
* Prev
;
120 struct rc_instruction
* Next
;
122 rc_instruction_type Type
;
124 struct rc_sub_instruction I
;
125 struct rc_pair_instruction P
;
129 * Warning: IPs are not stable. If you want to use them,
130 * you need to recompute them at the beginning of each pass
131 * using \ref rc_recompute_ips
138 * Instructions.Next points to the first instruction,
139 * Instructions.Prev points to the last instruction.
141 struct rc_instruction Instructions
;
143 /* Long term, we should probably remove InputsRead & OutputsWritten,
144 * since updating dependent state can be fragile, and they aren't
145 * actually used very often. */
147 uint32_t OutputsWritten
;
148 uint32_t ShadowSamplers
; /**< Texture units used for shadow sampling. */
150 struct rc_constant_list Constants
;
154 OPCODE_REPL_ALPHA
= MAX_RC_OPCODE
/**< used in paired instructions */
158 static inline rc_swizzle
get_swz(unsigned int swz
, rc_swizzle idx
)
162 return GET_SWZ(swz
, idx
);
165 static inline unsigned int combine_swizzles4(unsigned int src
,
166 rc_swizzle swz_x
, rc_swizzle swz_y
, rc_swizzle swz_z
, rc_swizzle swz_w
)
168 unsigned int ret
= 0;
170 ret
|= get_swz(src
, swz_x
);
171 ret
|= get_swz(src
, swz_y
) << 3;
172 ret
|= get_swz(src
, swz_z
) << 6;
173 ret
|= get_swz(src
, swz_w
) << 9;
178 static inline unsigned int combine_swizzles(unsigned int src
, unsigned int swz
)
180 unsigned int ret
= 0;
182 ret
|= get_swz(src
, GET_SWZ(swz
, RC_SWIZZLE_X
));
183 ret
|= get_swz(src
, GET_SWZ(swz
, RC_SWIZZLE_Y
)) << 3;
184 ret
|= get_swz(src
, GET_SWZ(swz
, RC_SWIZZLE_Z
)) << 6;
185 ret
|= get_swz(src
, GET_SWZ(swz
, RC_SWIZZLE_W
)) << 9;
190 struct rc_src_register
lmul_swizzle(unsigned int swizzle
, struct rc_src_register srcreg
);
192 static inline void reset_srcreg(struct rc_src_register
* reg
)
194 memset(reg
, 0, sizeof(struct rc_src_register
));
195 reg
->Swizzle
= RC_SWIZZLE_XYZW
;
200 * A transformation that can be passed to \ref radeonLocalTransform.
202 * The function will be called once for each instruction.
203 * It has to either emit the appropriate transformed code for the instruction
204 * and return true, or return false if it doesn't understand the
207 * The function gets passed the userData as last parameter.
209 struct radeon_program_transformation
{
211 struct radeon_compiler
*,
212 struct rc_instruction
*,
217 void radeonLocalTransform(
218 struct radeon_compiler
*c
,
219 int num_transformations
,
220 struct radeon_program_transformation
* transformations
);
222 unsigned int rc_find_free_temporary(struct radeon_compiler
* c
);
224 struct rc_instruction
*rc_alloc_instruction(struct radeon_compiler
* c
);
225 struct rc_instruction
*rc_insert_new_instruction(struct radeon_compiler
* c
, struct rc_instruction
* after
);
226 void rc_insert_instruction(struct rc_instruction
* after
, struct rc_instruction
* inst
);
227 void rc_remove_instruction(struct rc_instruction
* inst
);
229 unsigned int rc_recompute_ips(struct radeon_compiler
* c
);
231 void rc_print_program(const struct rc_program
*prog
);