2 * Copyright (C) 2008-2009 Advanced Micro Devices, Inc.
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice shall be included
12 * in all copies or substantial portions of the Software.
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
15 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
18 * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
19 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
24 * Richard Li <RichardZ.Li@amd.com>, <richardradeon@gmail.com>
34 #include "main/imports.h"
35 #include "main/mtypes.h"
37 #include "tnl/t_context.h"
38 #include "shader/program.h"
39 #include "shader/prog_parameter.h"
40 #include "shader/prog_statevars.h"
42 #include "radeon_debug.h"
43 #include "r600_context.h"
44 #include "r600_cmdbuf.h"
45 #include "shader/programopt.c"
47 #include "r700_debug.h"
48 #include "r700_vertprog.h"
50 unsigned int Map_Vertex_Output(r700_AssemblerBase
*pAsm
,
51 struct gl_vertex_program
*mesa_vp
,
56 unsigned int unTotal
= unStart
;
58 //!!!!!!! THE ORDER MATCH FS INPUT
60 unBit
= 1 << VERT_RESULT_HPOS
;
61 if(mesa_vp
->Base
.OutputsWritten
& unBit
)
63 pAsm
->ucVP_OutputMap
[VERT_RESULT_HPOS
] = unTotal
++;
66 unBit
= 1 << VERT_RESULT_COL0
;
67 if(mesa_vp
->Base
.OutputsWritten
& unBit
)
69 pAsm
->ucVP_OutputMap
[VERT_RESULT_COL0
] = unTotal
++;
72 unBit
= 1 << VERT_RESULT_COL1
;
73 if(mesa_vp
->Base
.OutputsWritten
& unBit
)
75 pAsm
->ucVP_OutputMap
[VERT_RESULT_COL1
] = unTotal
++;
78 //TODO : dealing back face.
79 unBit
= 1 << VERT_RESULT_BFC0
;
80 if(mesa_vp
->Base
.OutputsWritten
& unBit
)
82 pAsm
->ucVP_OutputMap
[VERT_RESULT_BFC0
] = unTotal
++;
85 unBit
= 1 << VERT_RESULT_BFC1
;
86 if(mesa_vp
->Base
.OutputsWritten
& unBit
)
88 pAsm
->ucVP_OutputMap
[VERT_RESULT_BFC1
] = unTotal
++;
92 unBit
= 1 << VERT_RESULT_FOGC
;
93 if(mesa_vp
->Base
.OutputsWritten
& unBit
)
95 pAsm
->ucVP_OutputMap
[VERT_RESULT_FOGC
] = unTotal
++;
98 //TODO : dealing point size.
99 unBit
= 1 << VERT_RESULT_PSIZ
;
100 if(mesa_vp
->Base
.OutputsWritten
& unBit
)
102 pAsm
->ucVP_OutputMap
[VERT_RESULT_PSIZ
] = unTotal
++;
107 unBit
= 1 << (VERT_RESULT_TEX0
+ i
);
108 if(mesa_vp
->Base
.OutputsWritten
& unBit
)
110 pAsm
->ucVP_OutputMap
[VERT_RESULT_TEX0
+ i
] = unTotal
++;
114 for(i
=VERT_RESULT_VAR0
; i
<VERT_RESULT_MAX
; i
++)
117 if(mesa_vp
->Base
.OutputsWritten
& unBit
)
119 pAsm
->ucVP_OutputMap
[i
] = unTotal
++;
123 return (unTotal
- unStart
);
126 unsigned int Map_Vertex_Input(r700_AssemblerBase
*pAsm
,
127 struct gl_vertex_program
*mesa_vp
,
128 unsigned int unStart
)
132 unsigned int unTotal
= unStart
;
133 for(i
=0; i
<VERT_ATTRIB_MAX
; i
++)
136 if(mesa_vp
->Base
.InputsRead
& unBit
)
138 pAsm
->ucVP_AttributeMap
[i
] = unTotal
++;
141 return (unTotal
- unStart
);
144 GLboolean
Process_Vertex_Program_Vfetch_Instructions(
145 struct r700_vertex_program
*vp
,
146 struct gl_vertex_program
*mesa_vp
)
150 VTX_FETCH_METHOD vtxFetchMethod
;
151 vtxFetchMethod
.bEnableMini
= GL_FALSE
;
152 vtxFetchMethod
.mega_fetch_remainder
= 0;
154 for(i
=0; i
<VERT_ATTRIB_MAX
; i
++)
157 if(mesa_vp
->Base
.InputsRead
& unBit
)
159 assemble_vfetch_instruction(&vp
->r700AsmCode
,
161 vp
->r700AsmCode
.ucVP_AttributeMap
[i
],
162 vp
->aos_desc
[i
].size
,
163 vp
->aos_desc
[i
].type
,
171 GLboolean
Process_Vertex_Program_Vfetch_Instructions2(
173 struct r700_vertex_program
*vp
,
174 struct gl_vertex_program
*mesa_vp
)
177 context_t
*context
= R700_CONTEXT(ctx
);
179 VTX_FETCH_METHOD vtxFetchMethod
;
180 vtxFetchMethod
.bEnableMini
= GL_FALSE
;
181 vtxFetchMethod
.mega_fetch_remainder
= 0;
183 for(i
=0; i
<context
->nNumActiveAos
; i
++)
185 assemble_vfetch_instruction2(&vp
->r700AsmCode
,
186 vp
->r700AsmCode
.ucVP_AttributeMap
[context
->stream_desc
[i
].element
],
187 context
->stream_desc
[i
].type
,
188 context
->stream_desc
[i
].size
,
189 context
->stream_desc
[i
].element
,
190 context
->stream_desc
[i
]._signed
,
191 context
->stream_desc
[i
].normalize
,
198 void Map_Vertex_Program(GLcontext
*ctx
,
199 struct r700_vertex_program
*vp
,
200 struct gl_vertex_program
*mesa_vp
)
203 r700_AssemblerBase
*pAsm
= &(vp
->r700AsmCode
);
204 unsigned int num_inputs
;
206 // R0 will always be used for index into vertex buffer
207 pAsm
->number_used_registers
= 1;
208 pAsm
->starting_vfetch_register_number
= pAsm
->number_used_registers
;
210 // Map Inputs: Add 1 to mapping since R0 is used for index
211 num_inputs
= Map_Vertex_Input(pAsm
, mesa_vp
, pAsm
->number_used_registers
);
212 pAsm
->number_used_registers
+= num_inputs
;
214 // Create VFETCH instructions for inputs
215 if (GL_TRUE
!= Process_Vertex_Program_Vfetch_Instructions2(ctx
, vp
, mesa_vp
) )
217 radeon_error("Calling Process_Vertex_Program_Vfetch_Instructions2 return error. \n");
222 pAsm
->number_of_exports
= Map_Vertex_Output(pAsm
, mesa_vp
, pAsm
->number_used_registers
);
224 pAsm
->starting_export_register_number
= pAsm
->number_used_registers
;
226 pAsm
->number_used_registers
+= pAsm
->number_of_exports
;
228 pAsm
->pucOutMask
= (unsigned char*) MALLOC(pAsm
->number_of_exports
);
230 for(ui
=0; ui
<pAsm
->number_of_exports
; ui
++)
232 pAsm
->pucOutMask
[ui
] = 0x0;
235 /* Map temporary registers (GPRs) */
236 pAsm
->starting_temp_register_number
= pAsm
->number_used_registers
;
238 if(mesa_vp
->Base
.NumNativeTemporaries
>= mesa_vp
->Base
.NumTemporaries
)
239 { /* arb uses NumNativeTemporaries */
240 pAsm
->number_used_registers
+= mesa_vp
->Base
.NumNativeTemporaries
;
243 { /* fix func t_vp uses NumTemporaries */
244 pAsm
->number_used_registers
+= mesa_vp
->Base
.NumTemporaries
;
247 pAsm
->flag_reg_index
= pAsm
->number_used_registers
++;
249 pAsm
->uFirstHelpReg
= pAsm
->number_used_registers
;
252 GLboolean
Find_Instruction_Dependencies_vp(struct r700_vertex_program
*vp
,
253 struct gl_vertex_program
*mesa_vp
)
256 GLint
* puiTEMPwrites
;
257 struct prog_instruction
*pILInst
;
260 puiTEMPwrites
= (GLint
*) MALLOC(sizeof(GLuint
)*mesa_vp
->Base
.NumTemporaries
);
261 for(i
=0; i
<mesa_vp
->Base
.NumTemporaries
; i
++)
263 puiTEMPwrites
[i
] = -1;
266 pInstDeps
= (InstDeps
*)MALLOC(sizeof(InstDeps
)*mesa_vp
->Base
.NumInstructions
);
268 for(i
=0; i
<mesa_vp
->Base
.NumInstructions
; i
++)
270 pInstDeps
[i
].nDstDep
= -1;
271 pILInst
= &(mesa_vp
->Base
.Instructions
[i
]);
274 if(pILInst
->DstReg
.File
== PROGRAM_TEMPORARY
)
276 //Set lastwrite for the temp
277 puiTEMPwrites
[pILInst
->DstReg
.Index
] = i
;
283 if(pILInst
->SrcReg
[j
].File
== PROGRAM_TEMPORARY
)
286 pInstDeps
[i
].nSrcDeps
[j
] = puiTEMPwrites
[pILInst
->SrcReg
[j
].Index
];
290 pInstDeps
[i
].nSrcDeps
[j
] = -1;
295 vp
->r700AsmCode
.pInstDeps
= pInstDeps
;
302 struct r700_vertex_program
* r700TranslateVertexShader(GLcontext
*ctx
,
303 struct gl_vertex_program
*mesa_vp
)
305 context_t
*context
= R700_CONTEXT(ctx
);
306 struct r700_vertex_program
*vp
;
309 vp
= _mesa_calloc(sizeof(*vp
));
310 vp
->mesa_program
= (struct gl_vertex_program
*)_mesa_clone_program(ctx
, &mesa_vp
->Base
);
312 if (mesa_vp
->IsPositionInvariant
)
314 _mesa_insert_mvp_code(ctx
, vp
->mesa_program
);
317 for(i
=0; i
<context
->nNumActiveAos
; i
++)
319 vp
->aos_desc
[i
].size
= context
->stream_desc
[i
].size
;
320 vp
->aos_desc
[i
].stride
= context
->stream_desc
[i
].stride
;
321 vp
->aos_desc
[i
].type
= context
->stream_desc
[i
].type
;
324 if (context
->radeon
.radeonScreen
->chip_family
< CHIP_FAMILY_RV770
)
326 vp
->r700AsmCode
.bR6xx
= 1;
330 Init_r700_AssemblerBase(SPT_VP
, &(vp
->r700AsmCode
), &(vp
->r700Shader
) );
331 Map_Vertex_Program(ctx
, vp
, vp
->mesa_program
);
333 if(GL_FALSE
== Find_Instruction_Dependencies_vp(vp
, vp
->mesa_program
))
338 InitShaderProgram(&(vp
->r700AsmCode
));
340 for(i
=0; i
< MAX_SAMPLERS
; i
++)
342 vp
->r700AsmCode
.SamplerUnits
[i
] = vp
->mesa_program
->Base
.SamplerUnits
[i
];
345 vp
->r700AsmCode
.unCurNumILInsts
= vp
->mesa_program
->Base
.NumInstructions
;
347 if(GL_FALSE
== AssembleInstr(0,
349 vp
->mesa_program
->Base
.NumInstructions
,
350 &(vp
->mesa_program
->Base
.Instructions
[0]),
351 &(vp
->r700AsmCode
)) )
356 if(GL_FALSE
== Process_Vertex_Exports(&(vp
->r700AsmCode
), vp
->mesa_program
->Base
.OutputsWritten
) )
361 if( GL_FALSE
== RelocProgram(&(vp
->r700AsmCode
), &(vp
->mesa_program
->Base
)) )
366 vp
->r700Shader
.nRegs
= (vp
->r700AsmCode
.number_used_registers
== 0) ? 0
367 : (vp
->r700AsmCode
.number_used_registers
- 1);
369 vp
->r700Shader
.nParamExports
= vp
->r700AsmCode
.number_of_exports
;
371 vp
->translated
= GL_TRUE
;
376 void r700SelectVertexShader(GLcontext
*ctx
)
378 context_t
*context
= R700_CONTEXT(ctx
);
379 struct r700_vertex_program_cont
*vpc
;
380 struct r700_vertex_program
*vp
;
383 GLbitfield InputsRead
;
385 vpc
= (struct r700_vertex_program_cont
*)ctx
->VertexProgram
._Current
;
387 InputsRead
= vpc
->mesa_program
.Base
.InputsRead
;
388 if (vpc
->mesa_program
.IsPositionInvariant
)
390 InputsRead
|= VERT_BIT_POS
;
393 for (vp
= vpc
->progs
; vp
; vp
= vp
->next
)
396 for(i
=0; i
<context
->nNumActiveAos
; i
++)
398 if (vp
->aos_desc
[i
].size
!= context
->stream_desc
[i
].size
)
406 context
->selected_vp
= vp
;
411 vp
= r700TranslateVertexShader(ctx
, &(vpc
->mesa_program
));
414 radeon_error("Failed to translate vertex shader. \n");
417 vp
->next
= vpc
->progs
;
419 context
->selected_vp
= vp
;
423 int getTypeSize(GLenum type
)
428 return sizeof(GLdouble
);
430 return sizeof(GLfloat
);
432 return sizeof(GLint
);
433 case GL_UNSIGNED_INT
:
434 return sizeof(GLuint
);
436 return sizeof(GLshort
);
437 case GL_UNSIGNED_SHORT
:
438 return sizeof(GLushort
);
440 return sizeof(GLbyte
);
441 case GL_UNSIGNED_BYTE
:
442 return sizeof(GLubyte
);
449 static void r700TranslateAttrib(GLcontext
*ctx
, GLuint unLoc
, int count
, const struct gl_client_array
*input
)
451 context_t
*context
= R700_CONTEXT(ctx
);
453 StreamDesc
* pStreamDesc
= &(context
->stream_desc
[context
->nNumActiveAos
]);
457 stride
= (input
->StrideB
== 0) ? getTypeSize(input
->Type
) * input
->Size
460 if (input
->Type
== GL_DOUBLE
|| input
->Type
== GL_UNSIGNED_INT
|| input
->Type
== GL_INT
||
462 getTypeSize(input
->Type
) != 4 ||
466 pStreamDesc
->type
= GL_FLOAT
;
468 if (input
->StrideB
== 0)
470 pStreamDesc
->stride
= 0;
474 pStreamDesc
->stride
= sizeof(GLfloat
) * input
->Size
;
476 pStreamDesc
->dwords
= input
->Size
;
477 pStreamDesc
->is_named_bo
= GL_FALSE
;
481 pStreamDesc
->type
= input
->Type
;
482 pStreamDesc
->dwords
= (getTypeSize(input
->Type
) * input
->Size
+ 3)/ 4;
483 if (!input
->BufferObj
->Name
)
485 if (input
->StrideB
== 0)
487 pStreamDesc
->stride
= 0;
491 pStreamDesc
->stride
= (getTypeSize(pStreamDesc
->type
) * input
->Size
+ 3) & ~3;
494 pStreamDesc
->is_named_bo
= GL_FALSE
;
498 pStreamDesc
->size
= input
->Size
;
499 pStreamDesc
->dst_loc
= context
->nNumActiveAos
;
500 pStreamDesc
->element
= unLoc
;
502 switch (pStreamDesc
->type
)
505 pStreamDesc
->_signed
= 0;
506 pStreamDesc
->normalize
= GL_FALSE
;
509 pStreamDesc
->_signed
= 1;
510 pStreamDesc
->normalize
= input
->Normalized
;
513 pStreamDesc
->_signed
= 1;
514 pStreamDesc
->normalize
= input
->Normalized
;
516 case GL_UNSIGNED_SHORT
:
517 pStreamDesc
->_signed
= 0;
518 pStreamDesc
->normalize
= input
->Normalized
;
520 case GL_UNSIGNED_BYTE
:
521 pStreamDesc
->_signed
= 0;
522 pStreamDesc
->normalize
= input
->Normalized
;
526 case GL_UNSIGNED_INT
:
531 context
->nNumActiveAos
++;
534 void r700SetVertexFormat(GLcontext
*ctx
, const struct gl_client_array
*arrays
[], int count
)
536 context_t
*context
= R700_CONTEXT(ctx
);
537 struct r700_vertex_program
*vpc
538 = (struct r700_vertex_program
*)ctx
->VertexProgram
._Current
;
540 struct gl_vertex_program
* mesa_vp
= (struct gl_vertex_program
*)&(vpc
->mesa_program
);
541 unsigned int unLoc
= 0;
542 unsigned int unBit
= mesa_vp
->Base
.InputsRead
;
543 context
->nNumActiveAos
= 0;
545 if (mesa_vp
->IsPositionInvariant
)
547 unBit
|= VERT_BIT_POS
;
554 r700TranslateAttrib(ctx
, unLoc
, count
, arrays
[unLoc
]);
560 context
->radeon
.tcl
.aos_count
= context
->nNumActiveAos
;
563 void * r700GetActiveVpShaderBo(GLcontext
* ctx
)
565 context_t
*context
= R700_CONTEXT(ctx
);
566 struct r700_vertex_program
*vp
= context
->selected_vp
;;
574 GLboolean
r700SetupVertexProgram(GLcontext
* ctx
)
576 context_t
*context
= R700_CONTEXT(ctx
);
577 R700_CHIP_CONTEXT
*r700
= (R700_CHIP_CONTEXT
*)(&context
->hw
);
578 struct r700_vertex_program
*vp
= context
->selected_vp
;
580 struct gl_program_parameter_list
*paramList
;
581 unsigned int unNumParamData
;
584 if(GL_FALSE
== vp
->loaded
)
586 if(vp
->r700Shader
.bNeedsAssembly
== GL_TRUE
)
588 Assemble( &(vp
->r700Shader
) );
594 (GLvoid
*)(vp
->r700Shader
.pProgram
),
595 vp
->r700Shader
.uShaderBinaryDWORDSize
,
598 vp
->loaded
= GL_TRUE
;
601 DumpHwBinary(DUMP_VERTEX_SHADER
, (GLvoid
*)(vp
->r700Shader
.pProgram
),
602 vp
->r700Shader
.uShaderBinaryDWORDSize
);
604 /* TODO : enable this after MemUse fixed *=
605 (context->chipobj.MemUse)(context, vp->shadercode.buf->id);
608 R600_STATECHANGE(context
, vs
);
609 R600_STATECHANGE(context
, fs
); /* hack */
611 r700
->vs
.SQ_PGM_RESOURCES_VS
.u32All
= 0;
612 SETbit(r700
->vs
.SQ_PGM_RESOURCES_VS
.u32All
, PGM_RESOURCES__PRIME_CACHE_ON_DRAW_bit
);
614 r700
->vs
.SQ_PGM_START_VS
.u32All
= 0; /* set from buffer object. */
616 SETfield(r700
->vs
.SQ_PGM_RESOURCES_VS
.u32All
, vp
->r700Shader
.nRegs
+ 1,
617 NUM_GPRS_shift
, NUM_GPRS_mask
);
619 if(vp
->r700Shader
.uStackSize
) /* we don't use branch for now, it should be zero. */
621 SETfield(r700
->vs
.SQ_PGM_RESOURCES_VS
.u32All
, vp
->r700Shader
.uStackSize
,
622 STACK_SIZE_shift
, STACK_SIZE_mask
);
625 R600_STATECHANGE(context
, spi
);
627 SETfield(r700
->SPI_VS_OUT_CONFIG
.u32All
,
628 vp
->r700Shader
.nParamExports
? (vp
->r700Shader
.nParamExports
- 1) : 0,
629 VS_EXPORT_COUNT_shift
, VS_EXPORT_COUNT_mask
);
630 SETfield(r700
->SPI_PS_IN_CONTROL_0
.u32All
, vp
->r700Shader
.nParamExports
,
631 NUM_INTERP_shift
, NUM_INTERP_mask
);
634 SETbit(r700->SPI_PS_IN_CONTROL_0.u32All, PERSP_GRADIENT_ENA_bit);
635 CLEARbit(r700->SPI_PS_IN_CONTROL_0.u32All, LINEAR_GRADIENT_ENA_bit);
638 /* sent out shader constants. */
639 paramList
= vp
->mesa_program
->Base
.Parameters
;
641 if(NULL
!= paramList
) {
642 /* vp->mesa_program was cloned, not updated by glsl shader api. */
643 /* _mesa_reference_program has already checked glsl shProg is ok and set ctx->VertexProgem._Current */
644 /* so, use ctx->VertexProgem._Current */
645 struct gl_program_parameter_list
*paramListOrginal
=
646 paramListOrginal
= ctx
->VertexProgram
._Current
->Base
.Parameters
;
648 _mesa_load_state_parameters(ctx
, paramList
);
650 if (paramList
->NumParameters
> R700_MAX_DX9_CONSTS
)
653 R600_STATECHANGE(context
, vs_consts
);
655 r700
->vs
.num_consts
= paramList
->NumParameters
;
657 unNumParamData
= paramList
->NumParameters
;
659 for(ui
=0; ui
<unNumParamData
; ui
++) {
660 if(paramList
->Parameters
[ui
].Type
== PROGRAM_UNIFORM
)
662 r700
->vs
.consts
[ui
][0].f32All
= paramListOrginal
->ParameterValues
[ui
][0];
663 r700
->vs
.consts
[ui
][1].f32All
= paramListOrginal
->ParameterValues
[ui
][1];
664 r700
->vs
.consts
[ui
][2].f32All
= paramListOrginal
->ParameterValues
[ui
][2];
665 r700
->vs
.consts
[ui
][3].f32All
= paramListOrginal
->ParameterValues
[ui
][3];
669 r700
->vs
.consts
[ui
][0].f32All
= paramList
->ParameterValues
[ui
][0];
670 r700
->vs
.consts
[ui
][1].f32All
= paramList
->ParameterValues
[ui
][1];
671 r700
->vs
.consts
[ui
][2].f32All
= paramList
->ParameterValues
[ui
][2];
672 r700
->vs
.consts
[ui
][3].f32All
= paramList
->ParameterValues
[ui
][3];
676 r700
->vs
.num_consts
= 0;
678 COMPILED_SUB
* pCompiledSub
;
680 GLuint unConstOffset
= r700
->vs
.num_consts
;
681 for(ui
=0; ui
<vp
->r700AsmCode
.unNumPresub
; ui
++)
683 pCompiledSub
= vp
->r700AsmCode
.presubs
[ui
].pCompiledSub
;
685 r700
->vs
.num_consts
+= pCompiledSub
->NumParameters
;
687 for(uj
=0; uj
<pCompiledSub
->NumParameters
; uj
++)
689 r700
->vs
.consts
[uj
+ unConstOffset
][0].f32All
= pCompiledSub
->ParameterValues
[uj
][0];
690 r700
->vs
.consts
[uj
+ unConstOffset
][1].f32All
= pCompiledSub
->ParameterValues
[uj
][1];
691 r700
->vs
.consts
[uj
+ unConstOffset
][2].f32All
= pCompiledSub
->ParameterValues
[uj
][2];
692 r700
->vs
.consts
[uj
+ unConstOffset
][3].f32All
= pCompiledSub
->ParameterValues
[uj
][3];
694 unConstOffset
+= pCompiledSub
->NumParameters
;