2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
4 develop this 3D driver.
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.
26 **********************************************************************/
29 * Keith Whitwell <keith@tungstengraphics.com>
36 #include "brw_structs.h"
37 #include "brw_defines.h"
38 #include "shader/prog_instruction.h"
40 #define BRW_SWIZZLE4(a,b,c,d) (((a)<<0) | ((b)<<2) | ((c)<<4) | ((d)<<6))
41 #define BRW_GET_SWZ(swz, idx) (((swz) >> ((idx)*2)) & 0x3)
43 #define BRW_SWIZZLE_NOOP BRW_SWIZZLE4(0,1,2,3)
44 #define BRW_SWIZZLE_XYZW BRW_SWIZZLE4(0,1,2,3)
45 #define BRW_SWIZZLE_XXXX BRW_SWIZZLE4(0,0,0,0)
46 #define BRW_SWIZZLE_XYXY BRW_SWIZZLE4(0,1,0,1)
49 #define REG_SIZE (8*4)
52 /* These aren't hardware structs, just something useful for us to pass around:
54 * Align1 operation has a lot of control over input ranges. Used in
55 * WM programs to implement shaders decomposed into "channel serial"
56 * or "structure of array" form:
63 GLuint subnr
:5; /* :1 in align16 */
64 GLuint negate
:1; /* source only */
65 GLuint abs
:1; /* source only */
66 GLuint vstride
:4; /* source only */
67 GLuint width
:3; /* src only, align1 only */
68 GLuint hstride
:2; /* align1 only */
69 GLuint address_mode
:1; /* relative addressing, hopefully! */
74 GLuint swizzle
:8; /* src only, align16 only */
75 GLuint writemask
:4; /* dest only, align16 only */
76 GLint indirect_offset
:10; /* relative addressing offset */
77 GLuint pad1
:10; /* two dwords total */
94 #define BRW_EU_MAX_INSN_STACK 5
95 #define BRW_EU_MAX_INSN 1200
98 struct brw_instruction store
[BRW_EU_MAX_INSN
];
101 /* Allow clients to push/pop instruction state:
103 struct brw_instruction stack
[BRW_EU_MAX_INSN_STACK
];
104 struct brw_instruction
*current
;
107 GLboolean single_program_flow
;
108 struct brw_context
*brw
;
113 static INLINE
int type_sz( GLuint type
)
116 case BRW_REGISTER_TYPE_UD
:
117 case BRW_REGISTER_TYPE_D
:
118 case BRW_REGISTER_TYPE_F
:
120 case BRW_REGISTER_TYPE_HF
:
121 case BRW_REGISTER_TYPE_UW
:
122 case BRW_REGISTER_TYPE_W
:
124 case BRW_REGISTER_TYPE_UB
:
125 case BRW_REGISTER_TYPE_B
:
133 * Construct a brw_reg.
134 * \param file one of the BRW_x_REGISTER_FILE values
135 * \param nr register number/index
136 * \param subnr register sub number
137 * \param type one of BRW_REGISTER_TYPE_x
138 * \param vstride one of BRW_VERTICAL_STRIDE_x
139 * \param width one of BRW_WIDTH_x
140 * \param hstride one of BRW_HORIZONTAL_STRIDE_x
141 * \param swizzle one of BRW_SWIZZLE_x
142 * \param writemask WRITEMASK_X/Y/Z/W bitfield
144 static INLINE
struct brw_reg
brw_reg( GLuint file
,
158 reg
.subnr
= subnr
* type_sz(type
);
161 reg
.vstride
= vstride
;
163 reg
.hstride
= hstride
;
164 reg
.address_mode
= BRW_ADDRESS_DIRECT
;
167 /* Could do better: If the reg is r5.3<0;1,0>, we probably want to
168 * set swizzle and writemask to W, as the lower bits of subnr will
169 * be lost when converted to align16. This is probably too much to
170 * keep track of as you'd want it adjusted by suboffset(), etc.
171 * Perhaps fix up when converting to align16?
173 reg
.dw1
.bits
.swizzle
= swizzle
;
174 reg
.dw1
.bits
.writemask
= writemask
;
175 reg
.dw1
.bits
.indirect_offset
= 0;
176 reg
.dw1
.bits
.pad1
= 0;
180 /** Construct float[16] register */
181 static INLINE
struct brw_reg
brw_vec16_reg( GLuint file
,
189 BRW_VERTICAL_STRIDE_16
,
191 BRW_HORIZONTAL_STRIDE_1
,
196 /** Construct float[8] register */
197 static INLINE
struct brw_reg
brw_vec8_reg( GLuint file
,
205 BRW_VERTICAL_STRIDE_8
,
207 BRW_HORIZONTAL_STRIDE_1
,
212 /** Construct float[4] register */
213 static INLINE
struct brw_reg
brw_vec4_reg( GLuint file
,
221 BRW_VERTICAL_STRIDE_4
,
223 BRW_HORIZONTAL_STRIDE_1
,
228 /** Construct float[2] register */
229 static INLINE
struct brw_reg
brw_vec2_reg( GLuint file
,
237 BRW_VERTICAL_STRIDE_2
,
239 BRW_HORIZONTAL_STRIDE_1
,
244 /** Construct float[1] register */
245 static INLINE
struct brw_reg
brw_vec1_reg( GLuint file
,
253 BRW_VERTICAL_STRIDE_0
,
255 BRW_HORIZONTAL_STRIDE_0
,
261 static INLINE
struct brw_reg
retype( struct brw_reg reg
,
268 static INLINE
struct brw_reg
suboffset( struct brw_reg reg
,
271 reg
.subnr
+= delta
* type_sz(reg
.type
);
276 static INLINE
struct brw_reg
offset( struct brw_reg reg
,
284 static INLINE
struct brw_reg
byte_offset( struct brw_reg reg
,
287 GLuint newoffset
= reg
.nr
* REG_SIZE
+ reg
.subnr
+ bytes
;
288 reg
.nr
= newoffset
/ REG_SIZE
;
289 reg
.subnr
= newoffset
% REG_SIZE
;
294 /** Construct unsigned word[16] register */
295 static INLINE
struct brw_reg
brw_uw16_reg( GLuint file
,
299 return suboffset(retype(brw_vec16_reg(file
, nr
, 0), BRW_REGISTER_TYPE_UW
), subnr
);
302 /** Construct unsigned word[8] register */
303 static INLINE
struct brw_reg
brw_uw8_reg( GLuint file
,
307 return suboffset(retype(brw_vec8_reg(file
, nr
, 0), BRW_REGISTER_TYPE_UW
), subnr
);
310 /** Construct unsigned word[1] register */
311 static INLINE
struct brw_reg
brw_uw1_reg( GLuint file
,
315 return suboffset(retype(brw_vec1_reg(file
, nr
, 0), BRW_REGISTER_TYPE_UW
), subnr
);
318 static INLINE
struct brw_reg
brw_imm_reg( GLuint type
)
320 return brw_reg( BRW_IMMEDIATE_VALUE
,
324 BRW_VERTICAL_STRIDE_0
,
326 BRW_HORIZONTAL_STRIDE_0
,
331 /** Construct float immediate register */
332 static INLINE
struct brw_reg
brw_imm_f( GLfloat f
)
334 struct brw_reg imm
= brw_imm_reg(BRW_REGISTER_TYPE_F
);
339 /** Construct integer immediate register */
340 static INLINE
struct brw_reg
brw_imm_d( GLint d
)
342 struct brw_reg imm
= brw_imm_reg(BRW_REGISTER_TYPE_D
);
347 /** Construct uint immediate register */
348 static INLINE
struct brw_reg
brw_imm_ud( GLuint ud
)
350 struct brw_reg imm
= brw_imm_reg(BRW_REGISTER_TYPE_UD
);
355 /** Construct ushort immediate register */
356 static INLINE
struct brw_reg
brw_imm_uw( GLushort uw
)
358 struct brw_reg imm
= brw_imm_reg(BRW_REGISTER_TYPE_UW
);
359 imm
.dw1
.ud
= uw
| (uw
<< 16);
363 /** Construct short immediate register */
364 static INLINE
struct brw_reg
brw_imm_w( GLshort w
)
366 struct brw_reg imm
= brw_imm_reg(BRW_REGISTER_TYPE_W
);
367 imm
.dw1
.d
= w
| (w
<< 16);
371 /* brw_imm_b and brw_imm_ub aren't supported by hardware - the type
372 * numbers alias with _V and _VF below:
375 /** Construct vector of eight signed half-byte values */
376 static INLINE
struct brw_reg
brw_imm_v( GLuint v
)
378 struct brw_reg imm
= brw_imm_reg(BRW_REGISTER_TYPE_V
);
379 imm
.vstride
= BRW_VERTICAL_STRIDE_0
;
380 imm
.width
= BRW_WIDTH_8
;
381 imm
.hstride
= BRW_HORIZONTAL_STRIDE_1
;
386 /** Construct vector of four 8-bit float values */
387 static INLINE
struct brw_reg
brw_imm_vf( GLuint v
)
389 struct brw_reg imm
= brw_imm_reg(BRW_REGISTER_TYPE_VF
);
390 imm
.vstride
= BRW_VERTICAL_STRIDE_0
;
391 imm
.width
= BRW_WIDTH_4
;
392 imm
.hstride
= BRW_HORIZONTAL_STRIDE_1
;
399 #define VF_NEG (1<<7)
401 static INLINE
struct brw_reg
brw_imm_vf4( GLuint v0
,
406 struct brw_reg imm
= brw_imm_reg(BRW_REGISTER_TYPE_VF
);
407 imm
.vstride
= BRW_VERTICAL_STRIDE_0
;
408 imm
.width
= BRW_WIDTH_4
;
409 imm
.hstride
= BRW_HORIZONTAL_STRIDE_1
;
410 imm
.dw1
.ud
= ((v0
<< 0) |
418 static INLINE
struct brw_reg
brw_address( struct brw_reg reg
)
420 return brw_imm_uw(reg
.nr
* REG_SIZE
+ reg
.subnr
);
423 /** Construct float[1] general-purpose register */
424 static INLINE
struct brw_reg
brw_vec1_grf( GLuint nr
, GLuint subnr
)
426 return brw_vec1_reg(BRW_GENERAL_REGISTER_FILE
, nr
, subnr
);
429 /** Construct float[2] general-purpose register */
430 static INLINE
struct brw_reg
brw_vec2_grf( GLuint nr
, GLuint subnr
)
432 return brw_vec2_reg(BRW_GENERAL_REGISTER_FILE
, nr
, subnr
);
435 /** Construct float[4] general-purpose register */
436 static INLINE
struct brw_reg
brw_vec4_grf( GLuint nr
, GLuint subnr
)
438 return brw_vec4_reg(BRW_GENERAL_REGISTER_FILE
, nr
, subnr
);
441 /** Construct float[8] general-purpose register */
442 static INLINE
struct brw_reg
brw_vec8_grf( GLuint nr
, GLuint subnr
)
444 return brw_vec8_reg(BRW_GENERAL_REGISTER_FILE
, nr
, subnr
);
448 static INLINE
struct brw_reg
brw_uw8_grf( GLuint nr
, GLuint subnr
)
450 return brw_uw8_reg(BRW_GENERAL_REGISTER_FILE
, nr
, subnr
);
453 static INLINE
struct brw_reg
brw_uw16_grf( GLuint nr
, GLuint subnr
)
455 return brw_uw16_reg(BRW_GENERAL_REGISTER_FILE
, nr
, subnr
);
459 /** Construct null register (usually used for setting condition codes) */
460 static INLINE
struct brw_reg
brw_null_reg( void )
462 return brw_vec8_reg(BRW_ARCHITECTURE_REGISTER_FILE
,
467 static INLINE
struct brw_reg
brw_address_reg( GLuint subnr
)
469 return brw_uw1_reg(BRW_ARCHITECTURE_REGISTER_FILE
,
474 /* If/else instructions break in align16 mode if writemask & swizzle
475 * aren't xyzw. This goes against the convention for other scalar
478 static INLINE
struct brw_reg
brw_ip_reg( void )
480 return brw_reg(BRW_ARCHITECTURE_REGISTER_FILE
,
483 BRW_REGISTER_TYPE_UD
,
484 BRW_VERTICAL_STRIDE_4
, /* ? */
486 BRW_HORIZONTAL_STRIDE_0
,
487 BRW_SWIZZLE_XYZW
, /* NOTE! */
488 WRITEMASK_XYZW
); /* NOTE! */
491 static INLINE
struct brw_reg
brw_acc_reg( void )
493 return brw_vec8_reg(BRW_ARCHITECTURE_REGISTER_FILE
,
499 static INLINE
struct brw_reg
brw_flag_reg( void )
501 return brw_uw1_reg(BRW_ARCHITECTURE_REGISTER_FILE
,
507 static INLINE
struct brw_reg
brw_mask_reg( GLuint subnr
)
509 return brw_uw1_reg(BRW_ARCHITECTURE_REGISTER_FILE
,
514 static INLINE
struct brw_reg
brw_message_reg( GLuint nr
)
516 return brw_vec8_reg(BRW_MESSAGE_REGISTER_FILE
,
524 /* This is almost always called with a numeric constant argument, so
525 * make things easy to evaluate at compile time:
527 static INLINE GLuint
cvt( GLuint val
)
541 static INLINE
struct brw_reg
stride( struct brw_reg reg
,
546 reg
.vstride
= cvt(vstride
);
547 reg
.width
= cvt(width
) - 1;
548 reg
.hstride
= cvt(hstride
);
553 static INLINE
struct brw_reg
vec16( struct brw_reg reg
)
555 return stride(reg
, 16,16,1);
558 static INLINE
struct brw_reg
vec8( struct brw_reg reg
)
560 return stride(reg
, 8,8,1);
563 static INLINE
struct brw_reg
vec4( struct brw_reg reg
)
565 return stride(reg
, 4,4,1);
568 static INLINE
struct brw_reg
vec2( struct brw_reg reg
)
570 return stride(reg
, 2,2,1);
573 static INLINE
struct brw_reg
vec1( struct brw_reg reg
)
575 return stride(reg
, 0,1,0);
579 static INLINE
struct brw_reg
get_element( struct brw_reg reg
, GLuint elt
)
581 return vec1(suboffset(reg
, elt
));
584 static INLINE
struct brw_reg
get_element_ud( struct brw_reg reg
, GLuint elt
)
586 return vec1(suboffset(retype(reg
, BRW_REGISTER_TYPE_UD
), elt
));
590 static INLINE
struct brw_reg
brw_swizzle( struct brw_reg reg
,
596 reg
.dw1
.bits
.swizzle
= BRW_SWIZZLE4(BRW_GET_SWZ(reg
.dw1
.bits
.swizzle
, x
),
597 BRW_GET_SWZ(reg
.dw1
.bits
.swizzle
, y
),
598 BRW_GET_SWZ(reg
.dw1
.bits
.swizzle
, z
),
599 BRW_GET_SWZ(reg
.dw1
.bits
.swizzle
, w
));
604 static INLINE
struct brw_reg
brw_swizzle1( struct brw_reg reg
,
607 return brw_swizzle(reg
, x
, x
, x
, x
);
610 static INLINE
struct brw_reg
brw_writemask( struct brw_reg reg
,
613 reg
.dw1
.bits
.writemask
&= mask
;
617 static INLINE
struct brw_reg
brw_set_writemask( struct brw_reg reg
,
620 reg
.dw1
.bits
.writemask
= mask
;
624 static INLINE
struct brw_reg
negate( struct brw_reg reg
)
630 static INLINE
struct brw_reg
brw_abs( struct brw_reg reg
)
636 /***********************************************************************
638 static INLINE
struct brw_reg
brw_vec4_indirect( GLuint subnr
,
641 struct brw_reg reg
= brw_vec4_grf(0, 0);
643 reg
.address_mode
= BRW_ADDRESS_REGISTER_INDIRECT_REGISTER
;
644 reg
.dw1
.bits
.indirect_offset
= offset
;
648 static INLINE
struct brw_reg
brw_vec1_indirect( GLuint subnr
,
651 struct brw_reg reg
= brw_vec1_grf(0, 0);
653 reg
.address_mode
= BRW_ADDRESS_REGISTER_INDIRECT_REGISTER
;
654 reg
.dw1
.bits
.indirect_offset
= offset
;
658 static INLINE
struct brw_reg
deref_4f(struct brw_indirect ptr
, GLint offset
)
660 return brw_vec4_indirect(ptr
.addr_subnr
, ptr
.addr_offset
+ offset
);
663 static INLINE
struct brw_reg
deref_1f(struct brw_indirect ptr
, GLint offset
)
665 return brw_vec1_indirect(ptr
.addr_subnr
, ptr
.addr_offset
+ offset
);
668 static INLINE
struct brw_reg
deref_4b(struct brw_indirect ptr
, GLint offset
)
670 return retype(deref_4f(ptr
, offset
), BRW_REGISTER_TYPE_B
);
673 static INLINE
struct brw_reg
deref_1uw(struct brw_indirect ptr
, GLint offset
)
675 return retype(deref_1f(ptr
, offset
), BRW_REGISTER_TYPE_UW
);
678 static INLINE
struct brw_reg
deref_1d(struct brw_indirect ptr
, GLint offset
)
680 return retype(deref_1f(ptr
, offset
), BRW_REGISTER_TYPE_D
);
683 static INLINE
struct brw_reg
deref_1ud(struct brw_indirect ptr
, GLint offset
)
685 return retype(deref_1f(ptr
, offset
), BRW_REGISTER_TYPE_UD
);
688 static INLINE
struct brw_reg
get_addr_reg(struct brw_indirect ptr
)
690 return brw_address_reg(ptr
.addr_subnr
);
693 static INLINE
struct brw_indirect
brw_indirect_offset( struct brw_indirect ptr
, GLint offset
)
695 ptr
.addr_offset
+= offset
;
699 static INLINE
struct brw_indirect
brw_indirect( GLuint addr_subnr
, GLint offset
)
701 struct brw_indirect ptr
;
702 ptr
.addr_subnr
= addr_subnr
;
703 ptr
.addr_offset
= offset
;
708 static INLINE
struct brw_instruction
*current_insn( struct brw_compile
*p
)
710 return &p
->store
[p
->nr_insn
];
713 void brw_pop_insn_state( struct brw_compile
*p
);
714 void brw_push_insn_state( struct brw_compile
*p
);
715 void brw_set_mask_control( struct brw_compile
*p
, GLuint value
);
716 void brw_set_saturate( struct brw_compile
*p
, GLuint value
);
717 void brw_set_access_mode( struct brw_compile
*p
, GLuint access_mode
);
718 void brw_set_compression_control( struct brw_compile
*p
, GLboolean control
);
719 void brw_set_predicate_control_flag_value( struct brw_compile
*p
, GLuint value
);
720 void brw_set_predicate_control( struct brw_compile
*p
, GLuint pc
);
721 void brw_set_conditionalmod( struct brw_compile
*p
, GLuint conditional
);
723 void brw_init_compile( struct brw_context
*, struct brw_compile
*p
);
724 const GLuint
*brw_get_program( struct brw_compile
*p
, GLuint
*sz
);
727 /* Helpers for regular instructions:
730 struct brw_instruction *brw_##OP(struct brw_compile *p, \
731 struct brw_reg dest, \
732 struct brw_reg src0);
735 struct brw_instruction *brw_##OP(struct brw_compile *p, \
736 struct brw_reg dest, \
737 struct brw_reg src0, \
738 struct brw_reg src1);
770 /* Helpers for SEND instruction:
772 void brw_urb_WRITE(struct brw_compile
*p
,
779 GLuint response_length
,
781 GLboolean writes_complete
,
785 void brw_fb_WRITE(struct brw_compile
*p
,
789 GLuint binding_table_index
,
791 GLuint response_length
,
794 void brw_SAMPLE(struct brw_compile
*p
,
798 GLuint binding_table_index
,
802 GLuint response_length
,
806 void brw_math_16( struct brw_compile
*p
,
814 void brw_math( struct brw_compile
*p
,
823 void brw_dp_READ_16( struct brw_compile
*p
,
826 GLuint scratch_offset
);
828 void brw_dp_WRITE_16( struct brw_compile
*p
,
831 GLuint scratch_offset
);
833 /* If/else/endif. Works by manipulating the execution flags on each
836 struct brw_instruction
*brw_IF(struct brw_compile
*p
,
837 GLuint execute_size
);
839 struct brw_instruction
*brw_ELSE(struct brw_compile
*p
,
840 struct brw_instruction
*if_insn
);
842 void brw_ENDIF(struct brw_compile
*p
,
843 struct brw_instruction
*if_or_else_insn
);
848 struct brw_instruction
*brw_DO(struct brw_compile
*p
,
849 GLuint execute_size
);
851 struct brw_instruction
*brw_WHILE(struct brw_compile
*p
,
852 struct brw_instruction
*patch_insn
);
854 struct brw_instruction
*brw_BREAK(struct brw_compile
*p
);
855 struct brw_instruction
*brw_CONT(struct brw_compile
*p
);
858 void brw_land_fwd_jump(struct brw_compile
*p
,
859 struct brw_instruction
*jmp_insn
);
863 void brw_NOP(struct brw_compile
*p
);
865 /* Special case: there is never a destination, execution size will be
868 void brw_CMP(struct brw_compile
*p
,
872 struct brw_reg src1
);
874 void brw_print_reg( struct brw_reg reg
);
877 /***********************************************************************
881 void brw_copy_indirect_to_indirect(struct brw_compile
*p
,
882 struct brw_indirect dst_ptr
,
883 struct brw_indirect src_ptr
,
886 void brw_copy_from_indirect(struct brw_compile
*p
,
888 struct brw_indirect ptr
,
891 void brw_copy4(struct brw_compile
*p
,
896 void brw_copy8(struct brw_compile
*p
,
901 void brw_math_invert( struct brw_compile
*p
,
905 void brw_set_src1( struct brw_instruction
*insn
,
906 struct brw_reg reg
);