1/*
2 * RadeonHD R6xx, R7xx DRI driver
3 *
4 * Copyright (C) 2008-2009  Alexander Deucher
5 * Copyright (C) 2008-2009  Matthias Hopf
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the "Software"),
9 * to deal in the Software without restriction, including without limitation
10 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
11 * and/or sell copies of the Software, and to permit persons to whom the
12 * Software is furnished to do so, subject to the following conditions:
13 *
14 * The above copyright notice and this permission notice shall be included
15 * in all copies or substantial portions of the Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
18 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
20 * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
21 * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
22 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
23 */
24
25/*
26 * Shader macros
27 */
28
29#ifndef __SHADER_H__
30#define __SHADER_H__
31
32#include "radeon.h"
33
34/* Restrictions of ALU instructions
35 * order of scalar ops is always x,y,z,w,t(rans), last to be indicated by last==1.
36 * max of 3 different src GPRs per instr.
37 * max of 4 different cfile constant components per instr.
38 * max of 2 (different) constants (any type) for t.
39 * bank swizzle (see below).
40 * GPR write stalls read of same register. Auto-replaced by PV/PS, NOP needed if registers are relative to
41 * different indices (gpr,loop,nothing).
42 * may use constant registers or constant cache, but not both.
43 */
44
45/* Bank_swizzle: (pp. 297ff)
46 * Only one of each x,y,z,w GPR component can be loaded per cycle (3 cycles per instr, called 0-2).
47 * per scalar instruction bank_swizzle can select which cycle each operand comes from. e.g.:
48 *   SRC0 SRC1 SRC2  SWIZZLE  cycle0 cycle1 cycle2
49 *   1.x  2.x          012     1.x    2.x     -
50 *   3.x  1.y          201     1.y     -     3.x
51 *   2.x  1.y          102    (1.y)  (2.x)    -
52 * If data is read in a cycle, multiple scalar instructions can reference it.
53 * Special case: square() - i.e. same component in src0+src1 doesn't need read port -> ignores swizzle for src1.
54 * No restrictions for constants or PV/PS.
55 * t can load multiple components in a single cycle slot, but has to share cycles with xyzw.
56 * t with single constant may not load GPRs or PV/PS in cycle 0 (carefull with ALU_TRANS_210).
57 * t with two constants may only load GPRs or PV/PS in cycle 2.
58 */
59
60
61/* Oder of instructions: All CF, All ALU, All Tex/Vtx fetches */
62
63
64// CF insts
65// addr
66#define ADDR(x)  (x)
67// pc
68#define POP_COUNT(x)      (x)
69// const
70#define CF_CONST(x)       (x)
71// cond
72#define COND(x)        (x)		// SQ_COND_*
73// count
74#define I_COUNT(x)        ((x) ? ((x) - 1) : 0)
75//r7xx
76#define COUNT_3(x)        (x)
77// call count
78#define CALL_COUNT(x)     (x)
79// eop
80#define END_OF_PROGRAM(x)   (x)
81// vpm
82#define VALID_PIXEL_MODE(x) (x)
83// cf inst
84#define CF_INST(x)        (x)		// SQ_CF_INST_*
85
86// wqm
87#define WHOLE_QUAD_MODE(x)  (x)
88// barrier
89#define BARRIER(x)          (x)
90//kb0
91#define KCACHE_BANK0(x)          (x)
92//kb1
93#define KCACHE_BANK1(x)          (x)
94// km0/1
95#define KCACHE_MODE0(x)          (x)
96#define KCACHE_MODE1(x)          (x)	// SQ_CF_KCACHE_*
97//
98#define KCACHE_ADDR0(x)          (x)
99#define KCACHE_ADDR1(x)          (x)
100// uw
101#define USES_WATERFALL(x)        (x)
102
103#define ARRAY_BASE(x)        (x)
104// export pixel
105#define CF_PIXEL_MRT0         0
106#define CF_PIXEL_MRT1         1
107#define CF_PIXEL_MRT2         2
108#define CF_PIXEL_MRT3         3
109#define CF_PIXEL_MRT4         4
110#define CF_PIXEL_MRT5         5
111#define CF_PIXEL_MRT6         6
112#define CF_PIXEL_MRT7         7
113// *_FOG: r6xx only
114#define CF_PIXEL_MRT0_FOG     16
115#define CF_PIXEL_MRT1_FOG     17
116#define CF_PIXEL_MRT2_FOG     18
117#define CF_PIXEL_MRT3_FOG     19
118#define CF_PIXEL_MRT4_FOG     20
119#define CF_PIXEL_MRT5_FOG     21
120#define CF_PIXEL_MRT6_FOG     22
121#define CF_PIXEL_MRT7_FOG     23
122#define CF_PIXEL_Z            61
123// export pos
124#define CF_POS0               60
125#define CF_POS1               61
126#define CF_POS2               62
127#define CF_POS3               63
128// export param
129// 0...31
130#define TYPE(x)              (x)	// SQ_EXPORT_*
131#if 0
132// type export
133#define SQ_EXPORT_PIXEL              0
134#define SQ_EXPORT_POS                1
135#define SQ_EXPORT_PARAM              2
136// reserved 3
137// type mem
138#define SQ_EXPORT_WRITE              0
139#define SQ_EXPORT_WRITE_IND          1
140#define SQ_EXPORT_WRITE_ACK          2
141#define SQ_EXPORT_WRITE_IND_ACK      3
142#endif
143
144#define RW_GPR(x)            (x)
145#define RW_REL(x)            (x)
146#define ABSOLUTE                  0
147#define RELATIVE                  1
148#define INDEX_GPR(x)            (x)
149#define ELEM_SIZE(x)            (x ? (x - 1) : 0)
150#define COMP_MASK(x)            (x)
151#define R6xx_ELEM_LOOP(x)            (x)
152#define BURST_COUNT(x)          (x ? (x - 1) : 0)
153
154// swiz
155#define SRC_SEL_X(x)    (x)		// SQ_SEL_* each
156#define SRC_SEL_Y(x)    (x)
157#define SRC_SEL_Z(x)    (x)
158#define SRC_SEL_W(x)    (x)
159
160#define CF_DWORD0(addr) cpu_to_le32((addr))
161// R7xx has another entry (COUNT3), but that is only used for adding a bit to count.
162// We allow one more bit for count in the argument of the macro on R7xx instead.
163// R6xx: [0,7]  R7xx: [1,16]
164#define CF_DWORD1(pc, cf_const, cond, count, call_count, eop, vpm, cf_inst, wqm, b) \
165    cpu_to_le32((((pc) << 0) | ((cf_const) << 3) | ((cond) << 8) | (((count) & 7) << 10) | (((count) >> 3) << 19) | \
166		 ((call_count) << 13) | ((eop) << 21) | ((vpm) << 22) | ((cf_inst) << 23) | ((wqm) << 30) | ((b) << 31)))
167
168#define CF_ALU_DWORD0(addr, kb0, kb1, km0) cpu_to_le32((((addr) << 0) | ((kb0) << 22) | ((kb1) << 26) | ((km0) << 30)))
169#define CF_ALU_DWORD1(km1, kcache_addr0, kcache_addr1, count, uw, cf_inst, wqm, b) \
170    cpu_to_le32((((km1) << 0) | ((kcache_addr0) << 2) | ((kcache_addr1) << 10) | \
171		 ((count) << 18) | ((uw) << 25) | ((cf_inst) << 26) | ((wqm) << 30) | ((b) << 31)))
172
173#define CF_ALLOC_IMP_EXP_DWORD0(array_base, type, rw_gpr, rr, index_gpr, es) \
174    cpu_to_le32((((array_base) << 0) | ((type) << 13) | ((rw_gpr) << 15) | ((rr) << 22) | ((index_gpr) << 23) | \
175		 ((es) << 30)))
176// R7xx apparently doesn't have the ELEM_LOOP entry any more
177// We still expose it, but ELEM_LOOP is explicitely R6xx now.
178// TODO: is this just forgotten in the docs, or really not available any more?
179#define CF_ALLOC_IMP_EXP_DWORD1_BUF(array_size, comp_mask, el, bc, eop, vpm, cf_inst, wqm, b) \
180    cpu_to_le32((((array_size) << 0) | ((comp_mask) << 12) | ((el) << 16) | ((bc) << 17) | \
181		 ((eop) << 21) | ((vpm) << 22) | ((cf_inst) << 23) | ((wqm) << 30) | ((b) << 31)))
182#define CF_ALLOC_IMP_EXP_DWORD1_SWIZ(sel_x, sel_y, sel_z, sel_w, el, bc, eop, vpm, cf_inst, wqm, b) \
183    cpu_to_le32((((sel_x) << 0) | ((sel_y) << 3) | ((sel_z) << 6) | ((sel_w) << 9) | ((el) << 16) | \
184		 ((bc) << 17) | ((eop) << 21) | ((vpm) << 22) | ((cf_inst) << 23) | \
185		 ((wqm) << 30) | ((b) << 31)))
186
187// ALU clause insts
188#define SRC0_SEL(x)        (x)
189#define SRC1_SEL(x)        (x)
190#define SRC2_SEL(x)        (x)
191// src[0-2]_sel
192//   0-127 GPR
193// 128-159 kcache constants bank 0
194// 160-191 kcache constants bank 1
195// 248-255 special SQ_ALU_SRC_* (0, 1, etc.)
196#define ALU_SRC_GPR_BASE        0
197#define ALU_SRC_KCACHE0_BASE  128
198#define ALU_SRC_KCACHE1_BASE  160
199#define ALU_SRC_CFILE_BASE    256
200
201#define SRC0_REL(x)        (x)
202#define SRC1_REL(x)        (x)
203#define SRC2_REL(x)        (x)
204// elem
205#define SRC0_ELEM(x)        (x)
206#define SRC1_ELEM(x)        (x)
207#define SRC2_ELEM(x)        (x)
208#define ELEM_X        0
209#define ELEM_Y        1
210#define ELEM_Z        2
211#define ELEM_W        3
212// neg
213#define SRC0_NEG(x)        (x)
214#define SRC1_NEG(x)        (x)
215#define SRC2_NEG(x)        (x)
216// im
217#define INDEX_MODE(x)    (x)		// SQ_INDEX_*
218// ps
219#define PRED_SEL(x)      (x)		// SQ_PRED_SEL_*
220// last
221#define LAST(x)          (x)
222// abs
223#define SRC0_ABS(x)       (x)
224#define SRC1_ABS(x)       (x)
225// uem
226#define UPDATE_EXECUTE_MASK(x) (x)
227// up
228#define UPDATE_PRED(x)      (x)
229// wm
230#define WRITE_MASK(x)   (x)
231// fm
232#define FOG_MERGE(x)    (x)
233// omod
234#define OMOD(x)        (x)		// SQ_ALU_OMOD_*
235// alu inst
236#define ALU_INST(x)        (x)		// SQ_ALU_INST_*
237//bs
238#define BANK_SWIZZLE(x)        (x)	// SQ_ALU_VEC_*
239#define DST_GPR(x)        (x)
240#define DST_REL(x)        (x)
241#define DST_ELEM(x)       (x)
242#define CLAMP(x)          (x)
243
244#define ALU_DWORD0(src0_sel, s0r, s0e, s0n, src1_sel, s1r, s1e, s1n, im, ps, last) \
245    cpu_to_le32((((src0_sel) << 0) | ((s0r) << 9) | ((s0e) << 10) | ((s0n) << 12) | \
246		 ((src1_sel) << 13) | ((s1r) << 22) | ((s1e) << 23) | ((s1n) << 25) | \
247		 ((im) << 26) | ((ps) << 29) | ((last) << 31)))
248// R7xx has alu_inst at a different slot, and no fog merge any more (no fix function fog any more)
249#define R6xx_ALU_DWORD1_OP2(s0a, s1a, uem, up, wm, fm, omod, alu_inst, bs, dst_gpr, dr, de, clamp) \
250    cpu_to_le32((((s0a) << 0) | ((s1a) << 1) | ((uem) << 2) | ((up) << 3) | ((wm) << 4) | \
251		 ((fm) << 5) | ((omod) << 6) | ((alu_inst) << 8) | ((bs) << 18) | ((dst_gpr) << 21) | \
252		 ((dr) << 28) | ((de) << 29) | ((clamp) << 31)))
253#define R7xx_ALU_DWORD1_OP2(s0a, s1a, uem, up, wm, omod, alu_inst, bs, dst_gpr, dr, de, clamp) \
254    cpu_to_le32((((s0a) << 0) | ((s1a) << 1) | ((uem) << 2) | ((up) << 3) | ((wm) << 4) | \
255		 ((omod) << 5) | ((alu_inst) << 7) | ((bs) << 18) | ((dst_gpr) << 21) | \
256		 ((dr) << 28) | ((de) << 29) | ((clamp) << 31)))
257// This is a general chipset macro, but due to selection by chipid typically not usable in static arrays
258// Fog is NOT USED on R7xx, even if specified.
259#define ALU_DWORD1_OP2(chipfamily, s0a, s1a, uem, up, wm, fm, omod, alu_inst, bs, dst_gpr, dr, de, clamp) \
260    ((chipfamily) < CHIP_FAMILY_RV770 ? \
261     R6xx_ALU_DWORD1_OP2(s0a, s1a, uem, up, wm, fm, omod, alu_inst, bs, dst_gpr, dr, de, clamp) : \
262     R7xx_ALU_DWORD1_OP2(s0a, s1a, uem, up, wm, omod, alu_inst, bs, dst_gpr, dr, de, clamp))
263#define ALU_DWORD1_OP3(src2_sel, s2r, s2e, s2n, alu_inst, bs, dst_gpr, dr, de, clamp) \
264    cpu_to_le32((((src2_sel) << 0) | ((s2r) << 9) | ((s2e) << 10) | ((s2n) << 12) | \
265		 ((alu_inst) << 13) | ((bs) << 18) | ((dst_gpr) << 21) | ((dr) << 28) | \
266		 ((de) << 29) | ((clamp) << 31)))
267
268// VTX clause insts
269// vxt insts
270#define VTX_INST(x)        (x)		// SQ_VTX_INST_*
271
272// fetch type
273#define FETCH_TYPE(x)        (x)	// SQ_VTX_FETCH_*
274
275#define FETCH_WHOLE_QUAD(x)        (x)
276#define BUFFER_ID(x)        (x)
277#define SRC_GPR(x)          (x)
278#define SRC_REL(x)          (x)
279#define MEGA_FETCH_COUNT(x)        ((x) ? ((x) - 1) : 0)
280
281#define SEMANTIC_ID(x)        (x)
282#define DST_SEL_X(x)          (x)
283#define DST_SEL_Y(x)          (x)
284#define DST_SEL_Z(x)          (x)
285#define DST_SEL_W(x)          (x)
286#define USE_CONST_FIELDS(x)   (x)
287#define DATA_FORMAT(x)        (x)
288// num format
289#define NUM_FORMAT_ALL(x)     (x)	// SQ_NUM_FORMAT_*
290// format comp
291#define FORMAT_COMP_ALL(x)     (x)	// SQ_FORMAT_COMP_*
292// sma
293#define SRF_MODE_ALL(x)     (x)
294#define SRF_MODE_ZERO_CLAMP_MINUS_ONE      0
295#define SRF_MODE_NO_ZERO                   1
296#define OFFSET(x)     (x)
297// endian swap
298#define ENDIAN_SWAP(x)     (x)		// SQ_ENDIAN_*
299#define CONST_BUF_NO_STRIDE(x)     (x)
300// mf
301#define MEGA_FETCH(x)     (x)
302
303#define VTX_DWORD0(vtx_inst, ft, fwq, buffer_id, src_gpr, sr, ssx, mfc) \
304    cpu_to_le32((((vtx_inst) << 0) | ((ft) << 5) | ((fwq) << 7) | ((buffer_id) << 8) | \
305		 ((src_gpr) << 16) | ((sr) << 23) | ((ssx) << 24) | ((mfc) << 26)))
306#define VTX_DWORD1_SEM(semantic_id, dsx, dsy, dsz, dsw, ucf, data_format, nfa, fca, sma) \
307    cpu_to_le32((((semantic_id) << 0) | ((dsx) << 9) | ((dsy) << 12) | ((dsz) << 15) | ((dsw) << 18) | \
308		 ((ucf) << 21) | ((data_format) << 22) | ((nfa) << 28) | ((fca) << 30) | ((sma) << 31)))
309#define VTX_DWORD1_GPR(dst_gpr, dr, dsx, dsy, dsz, dsw, ucf, data_format, nfa, fca, sma) \
310    cpu_to_le32((((dst_gpr) << 0) | ((dr) << 7) | ((dsx) << 9) | ((dsy) << 12) | ((dsz) << 15) | ((dsw) << 18) | \
311		 ((ucf) << 21) | ((data_format) << 22) | ((nfa) << 28) | ((fca) << 30) | ((sma) << 31)))
312#define VTX_DWORD2(offset, es, cbns, mf) \
313    cpu_to_le32((((offset) << 0) | ((es) << 16) | ((cbns) << 18) | ((mf) << 19)))
314#define VTX_DWORD_PAD cpu_to_le32(0x00000000)
315
316// TEX clause insts
317// tex insts
318#define TEX_INST(x)     (x)		// SQ_TEX_INST_*
319
320#define BC_FRAC_MODE(x)         (x)
321#define FETCH_WHOLE_QUAD(x)     (x)
322#define RESOURCE_ID(x)          (x)
323#define R7xx_ALT_CONST(x)            (x)
324
325#define LOD_BIAS(x)     (x)
326//ct
327#define COORD_TYPE_X(x)     (x)
328#define COORD_TYPE_Y(x)     (x)
329#define COORD_TYPE_Z(x)     (x)
330#define COORD_TYPE_W(x)     (x)
331#define TEX_UNNORMALIZED                0
332#define TEX_NORMALIZED                  1
333#define OFFSET_X(x) (((int)(x) * 2) & 0x1f) /* 4:1-bits 2's-complement fixed-point: [-8.0..7.5] */
334#define OFFSET_Y(x) (((int)(x) * 2) & 0x1f)
335#define OFFSET_Z(x) (((int)(x) * 2) & 0x1f)
336#define SAMPLER_ID(x)     (x)
337
338// R7xx has an additional parameter ALT_CONST. We always expose it, but ALT_CONST is R7xx only
339#define TEX_DWORD0(tex_inst, bfm, fwq, resource_id, src_gpr, sr, ac) \
340    cpu_to_le32((((tex_inst) << 0) | ((bfm) << 5) | ((fwq) << 7) | ((resource_id) << 8) | \
341		 ((src_gpr) << 16) | ((sr) << 23) | ((ac) << 24)))
342#define TEX_DWORD1(dst_gpr, dr, dsx, dsy, dsz, dsw, lod_bias, ctx, cty, ctz, ctw) \
343    cpu_to_le32((((dst_gpr) << 0) | ((dr) << 7) | ((dsx) << 9) | ((dsy) << 12) | ((dsz) << 15) | ((dsw) << 18) | \
344		 ((lod_bias) << 21) | ((ctx) << 28) | ((cty) << 29) | ((ctz) << 30) | ((ctw) << 31)))
345#define TEX_DWORD2(offset_x, offset_y, offset_z, sampler_id, ssx, ssy, ssz, ssw) \
346    cpu_to_le32((((offset_x) << 0) | ((offset_y) << 5) | ((offset_z) << 10) | ((sampler_id) << 15) | \
347		 ((ssx) << 20) | ((ssy) << 23) | ((ssz) << 26) | ((ssw) << 29)))
348#define TEX_DWORD_PAD cpu_to_le32(0x00000000)
349
350extern int R600_solid_vs(RADEONChipFamily ChipSet, uint32_t* vs);
351extern int R600_solid_ps(RADEONChipFamily ChipSet, uint32_t* ps);
352
353extern int R600_copy_vs(RADEONChipFamily ChipSet, uint32_t* vs);
354extern int R600_copy_ps(RADEONChipFamily ChipSet, uint32_t* ps);
355
356extern int R600_xv_vs(RADEONChipFamily ChipSet, uint32_t* shader);
357extern int R600_xv_ps(RADEONChipFamily ChipSet, uint32_t* shader);
358
359extern int R600_comp_vs(RADEONChipFamily ChipSet, uint32_t* vs);
360extern int R600_comp_ps(RADEONChipFamily ChipSet, uint32_t* ps);
361
362#endif
363