From: Ben Skeggs Date: Sat, 25 Nov 2006 09:58:35 +0000 (+0000) Subject: Incomplete shader stuff, should mostly work for NV40. Other cards, not so X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=9c9e6abbf82fbf591575a9c352f86721bc72aa90;p=mesa.git Incomplete shader stuff, should mostly work for NV40. Other cards, not so much.. --- diff --git a/src/mesa/drivers/dri/nouveau/Makefile b/src/mesa/drivers/dri/nouveau/Makefile index 4d1e3e6c70c..384713eeeb7 100644 --- a/src/mesa/drivers/dri/nouveau/Makefile +++ b/src/mesa/drivers/dri/nouveau/Makefile @@ -17,13 +17,22 @@ DRIVER_SOURCES = \ nouveau_screen.c \ nouveau_span.c \ nouveau_state.c \ + nouveau_shader.c \ + nouveau_shader_0_arb.c \ + nouveau_shader_1.c \ + nouveau_shader_2.c \ nouveau_tex.c \ nouveau_swtcl.c \ nv10_swtcl.c \ nv10_state.c \ nv20_state.c \ nv30_state.c \ - nouveau_state_cache.c + nouveau_state_cache.c \ + nv20_vertprog.c \ + nv30_fragprog.c \ + nv30_vertprog.c \ + nv40_fragprog.c \ + nv40_vertprog.c C_SOURCES = \ $(COMMON_SOURCES) \ diff --git a/src/mesa/drivers/dri/nouveau/nouveau_context.c b/src/mesa/drivers/dri/nouveau/nouveau_context.c index d3fbdab9f39..4ae0c68fa9a 100644 --- a/src/mesa/drivers/dri/nouveau/nouveau_context.c +++ b/src/mesa/drivers/dri/nouveau/nouveau_context.c @@ -35,6 +35,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. #include "tnl/tnl.h" #include "tnl/t_pipeline.h" +#include "tnl/t_vp_build.h" #include "drivers/common/driverfuncs.h" @@ -130,6 +131,13 @@ GLboolean nouveauCreateContext( const __GLcontextModes *glVisual, nmesa->current_primitive = -1; + nouveauShaderInitFuncs(ctx); + /* Install Mesa's fixed-function shader support */ + if (nmesa->screen->card->type >= NV_40) { + ctx->_MaintainTnlProgram = GL_TRUE; + ctx->_MaintainTexEnvProgram = GL_TRUE; + } + /* Initialize the swrast */ _swrast_CreateContext( ctx ); _ac_CreateContext( ctx ); diff --git a/src/mesa/drivers/dri/nouveau/nouveau_context.h b/src/mesa/drivers/dri/nouveau/nouveau_context.h index 8ae7be015dc..e488f9d42d0 100644 --- a/src/mesa/drivers/dri/nouveau/nouveau_context.h +++ b/src/mesa/drivers/dri/nouveau/nouveau_context.h @@ -38,6 +38,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. #include "nouveau_screen.h" #include "nouveau_state_cache.h" +#include "nouveau_shader.h" #include "xmlconfig.h" @@ -119,6 +120,12 @@ typedef struct nouveau_context { GLenum current_primitive; /* the current primitive enum */ DECLARE_RENDERINPUTS(render_inputs_bitset); /* the current render inputs */ + /* Shader state */ + nvsFunc VPfunc; + nvsFunc FPfunc; + nouveauShader *current_fragprog; + nouveauShader *current_vertprog; + nouveauScreenRec *screen; drm_nouveau_sarea_t *sarea; diff --git a/src/mesa/drivers/dri/nouveau/nouveau_fifo.h b/src/mesa/drivers/dri/nouveau/nouveau_fifo.h index ce465cdca50..44b9f356d15 100644 --- a/src/mesa/drivers/dri/nouveau/nouveau_fifo.h +++ b/src/mesa/drivers/dri/nouveau/nouveau_fifo.h @@ -33,6 +33,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. #include "nouveau_ctrlreg.h" //#define NOUVEAU_RING_DEBUG +//#define NOUVEAU_STATE_CACHE_DISABLE #define NV_READ(reg) *(volatile u_int32_t *)(nmesa->mmio + (reg)) @@ -63,11 +64,11 @@ int i; printf("OUT_RINGp: (size 0x%x dwords)\n",sz); for(i=0;ififo.buffer[nmesa->fifo.current++]=(n); \ #endif +#define BEGIN_RING_SIZE(subchannel,tag,size) do { \ + nouveau_state_cache_flush(nmesa); \ + if (nmesa->fifo.free <= (size)) \ + WAIT_RING(nmesa,(size)); \ + OUT_RING( ((size)<<18) | ((subchannel) << 13) | (tag)); \ + nmesa->fifo.free -= ((size) + 1); \ +}while(0) + extern void WAIT_RING(nouveauContextPtr nmesa,u_int32_t size); extern void nouveau_state_cache_flush(nouveauContextPtr nmesa); extern void nouveau_state_cache_init(nouveauContextPtr nmesa); +#ifdef NOUVEAU_STATE_CACHE_DISABLE +#define BEGIN_RING_CACHE(subc,tag,size) BEGIN_RING_SIZE((subc), (tag), (size)) +#define OUT_RING_CACHE(n) OUT_RING((n)) +#define OUT_RING_CACHEf(n) OUT_RINGf((n)) +#define OUT_RING_CACHEp(ptr, sz) OUT_RINGp((ptr), (sz)) +#else #define BEGIN_RING_CACHE(subchannel,tag,size) do { \ nmesa->state_cache.dirty=1; \ nmesa->state_cache.current_pos=((tag)/4); \ @@ -116,14 +131,7 @@ extern void nouveau_state_cache_init(nouveauContextPtr nmesa); uint32_t* p=(uint32_t*)(ptr); \ int i; for(i=0;ififo.free <= (size)) \ - WAIT_RING(nmesa,(size)); \ - OUT_RING( ((size)<<18) | ((subchannel) << 13) | (tag)); \ - nmesa->fifo.free -= ((size) + 1); \ -}while(0) +#endif #define RING_AVAILABLE() (nmesa->fifo.free-1) diff --git a/src/mesa/drivers/dri/nouveau/nouveau_object.c b/src/mesa/drivers/dri/nouveau/nouveau_object.c index fe3b44df695..cd46feff7cf 100644 --- a/src/mesa/drivers/dri/nouveau/nouveau_object.c +++ b/src/mesa/drivers/dri/nouveau/nouveau_object.c @@ -54,7 +54,7 @@ void nouveauObjectInit(nouveauContextPtr nmesa) nouveauCreateContextObject(nmesa, Nv3D, nmesa->screen->card->class_3d, 0, 0, 0, 0); nouveauObjectOnSubchannel(nmesa, NvSub3D, Nv3D); /* We need to know vram size.. */ -#if 0 +#if 0 nouveauCreateDmaObject( nmesa, NvDmaFB, 0, (256*1024*1024), 0 /*NV_DMA_TARGET_FB*/, 0 /*NV_DMA_ACCESS_RW*/); diff --git a/src/mesa/drivers/dri/nouveau/nouveau_shader.c b/src/mesa/drivers/dri/nouveau/nouveau_shader.c new file mode 100644 index 00000000000..97ea1ee547a --- /dev/null +++ b/src/mesa/drivers/dri/nouveau/nouveau_shader.c @@ -0,0 +1,734 @@ +/* + * Copyright (C) 2006 Ben Skeggs. + * + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial + * portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + */ + +/* + * Authors: + * Ben Skeggs + */ + +#include "glheader.h" +#include "macros.h" +#include "enums.h" +#include "extensions.h" + +#include "program.h" +#include "tnl/tnl.h" + +#include "nouveau_context.h" +#include "nouveau_shader.h" + +/***************************************************************************** + * Mesa entry points + */ +static void +nouveauBindProgram(GLcontext *ctx, GLenum target, struct gl_program *prog) +{ +} + +static struct gl_program * +nouveauNewProgram(GLcontext *ctx, GLenum target, GLuint id) +{ + nouveauShader *nvs; + + nvs = CALLOC_STRUCT(_nouveauShader); + switch (target) { + case GL_VERTEX_PROGRAM_ARB: + return _mesa_init_vertex_program(ctx, &nvs->mesa.vp, target, id); + case GL_FRAGMENT_PROGRAM_ARB: + return _mesa_init_fragment_program(ctx, &nvs->mesa.fp, target, id); + default: + _mesa_problem(ctx, "Unsupported shader target"); + break; + } + + FREE(nvs); + return NULL; +} + +static void +nouveauDeleteProgram(GLcontext *ctx, struct gl_program *prog) +{ + nouveauShader *nvs = (nouveauShader *)prog; + + if (nvs->translated) + FREE(nvs->program); + _mesa_delete_program(ctx, prog); +} + +static void +nouveauProgramStringNotify(GLcontext *ctx, GLenum target, + struct gl_program *prog) +{ + nouveauShader *nvs = (nouveauShader *)prog; + + if (nvs->translated) + FREE(nvs->program); + nvs->translated = 0; + + _tnl_program_string(ctx, target, prog); +} + +static GLboolean +nouveauIsProgramNative(GLcontext * ctx, GLenum target, struct gl_program *prog) +{ + nouveauShader *nvs = (nouveauShader *)prog; + + return nvs->translated; +} + +GLboolean +nvsUpdateShader(GLcontext *ctx, nouveauShader *nvs) +{ + nouveauContextPtr nmesa = NOUVEAU_CONTEXT(ctx); + struct gl_program_parameter_list *plist; + int i; + + /* Translate to HW format now if necessary */ + if (!nvs->translated) { + /* Mesa ASM shader -> nouveauShader */ + if (!nouveau_shader_pass0_arb(ctx, nvs)) + return GL_FALSE; + /* Basic dead code elimination + register usage info */ + if (!nouveau_shader_pass1(nvs)) + return GL_FALSE; + /* nouveauShader -> HW bytecode, HW register alloc */ + if (!nouveau_shader_pass2(nvs)) + return GL_FALSE; + assert(nvs->translated); + assert(nvs->program); + } + + /* Update state parameters */ + plist = nvs->mesa.vp.Base.Parameters; + _mesa_load_state_parameters(ctx, plist); + for (i=0; iNumParameters; i++) { + if (!nvs->on_hardware) { + /* if we've been kicked off the hardware there's no guarantee our + * consts are still there.. reupload them all + */ + nvs->func->UpdateConst(ctx, nvs, i); + } else if (plist->Parameters[i].Type == PROGRAM_STATE_VAR) { + /* update any changed state parameters */ + if (!TEST_EQ_4V(nvs->params[i].val, nvs->params[i].source_val)) + nvs->func->UpdateConst(ctx, nvs, i); + } + } + + /* Upload program to hardware, this must come after state param update + * as >=NV30 fragprogs inline consts into the bytecode. + */ + if (!nvs->on_hardware) { + nouveauShader **current; + + if (nvs->mesa.vp.Base.Target == GL_VERTEX_PROGRAM_ARB) + current = &nmesa->current_vertprog; + else + current = &nmesa->current_fragprog; + if (*current) (*current)->on_hardware = 0; + + nvs->func->UploadToHW(ctx, nvs); + nvs->on_hardware = 1; + + *current = nvs; + } + + return GL_TRUE; +} + +void +nouveauShaderInitFuncs(GLcontext * ctx) +{ + nouveauContextPtr nmesa = NOUVEAU_CONTEXT(ctx); + + switch (nmesa->screen->card->type) { + case NV_20: + NV20VPInitShaderFuncs(&nmesa->VPfunc); + break; + case NV_30: + NV30VPInitShaderFuncs(&nmesa->VPfunc); + NV30FPInitShaderFuncs(&nmesa->FPfunc); + break; + case NV_40: + case G_70: + NV40VPInitShaderFuncs(&nmesa->VPfunc); + NV40FPInitShaderFuncs(&nmesa->FPfunc); + break; + default: + return; + } + + _mesa_enable_extension(ctx, "GL_ARB_vertex_program"); + ctx->Const.VertexProgram.MaxNativeInstructions = nmesa->VPfunc.MaxInst; + ctx->Const.VertexProgram.MaxNativeAluInstructions = nmesa->VPfunc.MaxInst; + ctx->Const.VertexProgram.MaxNativeTexInstructions = nmesa->VPfunc.MaxInst; + ctx->Const.VertexProgram.MaxNativeTexIndirections = + ctx->Const.VertexProgram.MaxNativeTexInstructions; + ctx->Const.VertexProgram.MaxNativeAttribs = nmesa->VPfunc.MaxAttrib; + ctx->Const.VertexProgram.MaxNativeTemps = nmesa->VPfunc.MaxTemp; + ctx->Const.VertexProgram.MaxNativeAddressRegs = nmesa->VPfunc.MaxAddress; + ctx->Const.VertexProgram.MaxNativeParameters = nmesa->VPfunc.MaxConst; + + if (nmesa->screen->card->type >= NV_30) { + _mesa_enable_extension(ctx, "GL_ARB_fragment_program"); + + ctx->Const.FragmentProgram.MaxNativeInstructions = nmesa->FPfunc.MaxInst; + ctx->Const.FragmentProgram.MaxNativeAluInstructions = nmesa->FPfunc.MaxInst; + ctx->Const.FragmentProgram.MaxNativeTexInstructions = nmesa->FPfunc.MaxInst; + ctx->Const.FragmentProgram.MaxNativeTexIndirections = + ctx->Const.FragmentProgram.MaxNativeTexInstructions; + ctx->Const.FragmentProgram.MaxNativeAttribs = nmesa->FPfunc.MaxAttrib; + ctx->Const.FragmentProgram.MaxNativeTemps = nmesa->FPfunc.MaxTemp; + ctx->Const.FragmentProgram.MaxNativeAddressRegs = nmesa->FPfunc.MaxAddress; + ctx->Const.FragmentProgram.MaxNativeParameters = nmesa->FPfunc.MaxConst; + } + + ctx->Driver.NewProgram = nouveauNewProgram; + ctx->Driver.BindProgram = nouveauBindProgram; + ctx->Driver.DeleteProgram = nouveauDeleteProgram; + ctx->Driver.ProgramStringNotify = nouveauProgramStringNotify; + ctx->Driver.IsProgramNative = nouveauIsProgramNative; +} + + +/***************************************************************************** + * Disassembly support structs + */ +#define CHECK_RANGE(idx, arr) ((idx)= (sizeof(ops) / sizeof(struct _opcode_info))) + return NULL; + if (ops[op].name == NULL) + return NULL; + return &ops[op]; +} + +static const char *_SFR_STRING[] = { + [NVS_FR_POSITION] = "position", + [NVS_FR_WEIGHT] = "weight", + [NVS_FR_NORMAL] = "normal", + [NVS_FR_COL0] = "color", + [NVS_FR_COL1] = "color.secondary", + [NVS_FR_BFC0] = "bfc", + [NVS_FR_BFC1] = "bfc.secondary", + [NVS_FR_FOGCOORD] = "fogcoord", + [NVS_FR_POINTSZ] = "pointsize", + [NVS_FR_TEXCOORD0] = "texcoord[0]", + [NVS_FR_TEXCOORD1] = "texcoord[1]", + [NVS_FR_TEXCOORD2] = "texcoord[2]", + [NVS_FR_TEXCOORD3] = "texcoord[3]", + [NVS_FR_TEXCOORD4] = "texcoord[4]", + [NVS_FR_TEXCOORD5] = "texcoord[5]", + [NVS_FR_TEXCOORD6] = "texcoord[6]", + [NVS_FR_TEXCOORD7] = "texcoord[7]", + [NVS_FR_FRAGDATA0] = "data[0]", + [NVS_FR_FRAGDATA1] = "data[1]", + [NVS_FR_FRAGDATA2] = "data[2]", + [NVS_FR_FRAGDATA3] = "data[3]", + [NVS_FR_CLIP0] = "clip_plane[0]", + [NVS_FR_CLIP1] = "clip_plane[1]", + [NVS_FR_CLIP2] = "clip_plane[2]", + [NVS_FR_CLIP3] = "clip_plane[3]", + [NVS_FR_CLIP4] = "clip_plane[4]", + [NVS_FR_CLIP5] = "clip_plane[5]", + [NVS_FR_CLIP6] = "clip_plane[6]", + [NVS_FR_FACING] = "facing", +}; + +#define SFR_STRING(idx) CHECK_RANGE((idx), SFR_STRING) + +static const char *_SWZ_STRING[] = { + [NVS_SWZ_X] = "x", + [NVS_SWZ_Y] = "y", + [NVS_SWZ_Z] = "z", + [NVS_SWZ_W] = "w" +}; + +#define SWZ_STRING(idx) CHECK_RANGE((idx), SWZ_STRING) + +static const char *_NVS_PREC_STRING[] = { + [NVS_PREC_FLOAT32] = "R", + [NVS_PREC_FLOAT16] = "H", + [NVS_PREC_FIXED12] = "X", + [NVS_PREC_UNKNOWN] = "?" +}; + +#define NVS_PREC_STRING(idx) CHECK_RANGE((idx), NVS_PREC_STRING) + +static const char *_NVS_COND_STRING[] = { + [NVS_COND_FL] = "FL", + [NVS_COND_LT] = "LT", + [NVS_COND_EQ] = "EQ", + [NVS_COND_LE] = "LE", + [NVS_COND_GT] = "GT", + [NVS_COND_NE] = "NE", + [NVS_COND_GE] = "GE", + [NVS_COND_TR] = "TR", + [NVS_COND_UNKNOWN] = "??" +}; + +#define NVS_COND_STRING(idx) CHECK_RANGE((idx), NVS_COND_STRING) + +/***************************************************************************** + * ShaderFragment dumping + */ +static void +nvsDumpIndent(int lvl) +{ + while (lvl--) + printf(" "); +} + +static void +nvsDumpSwizzle(nvsSwzComp *swz) +{ + printf(".%s%s%s%s", + SWZ_STRING(swz[0]), + SWZ_STRING(swz[1]), SWZ_STRING(swz[2]), SWZ_STRING(swz[3]) + ); +} + +static void +nvsDumpReg(nvsInstruction * inst, nvsRegister * reg) +{ + if (reg->negate) + printf("-"); + if (reg->abs) + printf("abs("); + + switch (reg->file) { + case NVS_FILE_TEMP: + printf("R%d", reg->index); + nvsDumpSwizzle(reg->swizzle); + break; + case NVS_FILE_ATTRIB: + printf("attrib.%s", SFR_STRING(reg->index)); + nvsDumpSwizzle(reg->swizzle); + break; + case NVS_FILE_ADDRESS: + printf("A%d", reg->index); + break; + case NVS_FILE_CONST: + if (reg->indexed) + printf("const[A%d.%s + %d]", + reg->addr_reg, SWZ_STRING(reg->addr_comp), reg->index); + else + printf("const[%d]", reg->index); + nvsDumpSwizzle(reg->swizzle); + break; + default: + printf("UNKNOWN_FILE"); + break; + } + + if (reg->abs) + printf(")"); +} + +void +nvsDumpInstruction(nvsInstruction * inst, int slot, int lvl) +{ + struct _opcode_info *opr = &ops[inst->op]; + int i; + + nvsDumpIndent(lvl); + printf("%s ", opr->name); + + if (!opr->flags & NODS) { + switch (inst->dest.file) { + case NVS_FILE_RESULT: + printf("result.%s", SFR_STRING(inst->dest.index)); + break; + case NVS_FILE_TEMP: + printf("R%d", inst->dest.index); + break; + case NVS_FILE_ADDRESS: + printf("A%d", inst->dest.index); + break; + default: + printf("UNKNOWN_DST_FILE"); + break; + } + + if (inst->mask != SMASK_ALL) { + printf("."); + if (inst->mask & SMASK_X) + printf("x"); + if (inst->mask & SMASK_Y) + printf("y"); + if (inst->mask & SMASK_Z) + printf("z"); + if (inst->mask & SMASK_W) + printf("w"); + } + + if (opr->numsrc) + printf(", "); + } + + for (i = 0; i < opr->numsrc; i++) { + nvsDumpReg(inst, &inst->src[i]); + if (i != opr->numsrc - 1) + printf(", "); + } + if (opr->flags & TI_UNIT) + printf(", texture[%d]", inst->tex_unit); + + printf("\n"); +} + +void +nvsDumpFragmentList(nvsFragmentList *f, int lvl) +{ + while (f) { + switch (f->fragment->type) { + case NVS_INSTRUCTION: + nvsDumpInstruction((nvsInstruction*)f->fragment, 0, lvl); + break; + default: + fprintf(stderr, "%s: Only NVS_INSTRUCTION fragments can be in" + "nvsFragmentList!\n", __func__); + return; + } + f = f->next; + } +} + +/***************************************************************************** + * HW shader disassembly + */ +static void +nvsDisasmHWShaderOp(nvsFunc * shader, int merged) +{ + struct _opcode_info *opi; + nvsOpcode op; + nvsRegFile file; + nvsSwzComp swz[4]; + int i; + + op = shader->GetOpcode(shader, merged); + opi = _get_op_info(op); + if (!opi) { + printf("NO OPINFO!"); + return; + } + + printf("%s", opi->name); + if (shader->GetPrecision && + (!(opi->flags & BRANCH_ALL)) && (!(opi->flags * NODS)) && + (op != NVS_OP_NOP)) + printf("%s", NVS_PREC_STRING(shader->GetPrecision(shader))); + if (shader->SupportsConditional && shader->SupportsConditional(shader)) { + if (shader->GetConditionUpdate(shader)) { + printf("C%d", shader->GetCondRegID(shader)); + } + } + if (shader->GetSaturate && shader->GetSaturate(shader)) + printf("_SAT"); + + if (!(opi->flags & NODS)) { + int mask = shader->GetDestMask(shader, merged); + + switch (shader->GetDestFile(shader, merged)) { + case NVS_FILE_ADDRESS: + printf(" A%d", shader->GetDestID(shader, merged)); + break; + case NVS_FILE_TEMP: + printf(" R%d", shader->GetDestID(shader, merged)); + break; + case NVS_FILE_RESULT: + printf(" result.%s", (SFR_STRING(shader->GetDestID(shader, merged)))); + break; + default: + printf(" BAD_RESULT_FILE"); + break; + } + + if (mask != SMASK_ALL) { + printf("."); + if (mask & SMASK_X) printf("x"); + if (mask & SMASK_Y) printf("y"); + if (mask & SMASK_Z) printf("z"); + if (mask & SMASK_W) printf("w"); + } + } + + if (shader->SupportsConditional && shader->SupportsConditional(shader) && + shader->GetConditionTest(shader)) { + shader->GetCondRegSwizzle(shader, swz); + + printf(" (%s%d.%s%s%s%s)", + NVS_COND_STRING(shader->GetCondition(shader)), + shader->GetCondRegID(shader), + SWZ_STRING(swz[NVS_SWZ_X]), + SWZ_STRING(swz[NVS_SWZ_Y]), + SWZ_STRING(swz[NVS_SWZ_Z]), + SWZ_STRING(swz[NVS_SWZ_W]) + ); + } + + /* looping */ + if (opi->flags & COUNT_ALL) { + printf(" { "); + if (opi->flags & COUNT_NUM) { + printf("%d", shader->GetLoopCount(shader)); + } + if (opi->flags & COUNT_IND) { + printf(", %d", shader->GetLoopInitial(shader)); + } + if (opi->flags & COUNT_INC) { + printf(", %d", shader->GetLoopIncrement(shader)); + } + printf(" }"); + } + + /* branching */ + if (opi->flags & BRANCH_TR) + printf(" %d", shader->GetBranch(shader)); + if (opi->flags & BRANCH_EL) + printf(" ELSE %d", shader->GetBranchElse(shader)); + if (opi->flags & BRANCH_EN) + printf(" END %d", shader->GetBranchEnd(shader)); + + if (!(opi->flags & NODS) && opi->numsrc) + printf(","); + printf(" "); + + for (i = 0; i < opi->numsrc; i++) { + if (shader->GetSourceAbs(shader, merged, i)) + printf("abs("); + if (shader->GetSourceNegate(shader, merged, i)) + printf("-"); + + file = shader->GetSourceFile(shader, merged, i); + switch (file) { + case NVS_FILE_TEMP: + printf("R%d", shader->GetSourceID(shader, merged, i)); + break; + case NVS_FILE_CONST: + if (shader->GetSourceIndexed(shader, merged, i)) { + printf("c[A%d.%s + 0x%x]", + shader->GetRelAddressRegID(shader), + SWZ_STRING(shader->GetRelAddressSwizzle(shader)), + shader->GetSourceID(shader, merged, i) + ); + } else { + float val[4]; + + if (shader->GetSourceConstVal) { + shader->GetSourceConstVal(shader, merged, i, val); + printf("{ %.02f, %.02f, %.02f, %.02f }", + val[0], val[1], val[2], val[3]); + } else { + printf("c[0x%x]", shader->GetSourceID(shader, merged, i)); + } + } + break; + case NVS_FILE_ATTRIB: + if (shader->GetSourceIndexed(shader, merged, i)) { + printf("attrib[A%d.%s + %d]", + shader->GetRelAddressRegID(shader), + SWZ_STRING(shader->GetRelAddressSwizzle(shader)), + shader->GetSourceID(shader, merged, i) + ); + } + else { + printf("attrib.%s", + SFR_STRING(shader->GetSourceID(shader, merged, i)) + ); + } + break; + case NVS_FILE_ADDRESS: + printf("A%d", shader->GetRelAddressRegID(shader)); + break; + default: + printf("UNKNOWN_SRC_FILE"); + break; + } + + shader->GetSourceSwizzle(shader, merged, i, swz); + if (file != NVS_FILE_ADDRESS && + (swz[NVS_SWZ_X] != NVS_SWZ_X || swz[NVS_SWZ_Y] != NVS_SWZ_Y || + swz[NVS_SWZ_Z] != NVS_SWZ_Z || swz[NVS_SWZ_W] != NVS_SWZ_W)) { + printf(".%s%s%s%s", SWZ_STRING(swz[NVS_SWZ_X]), + SWZ_STRING(swz[NVS_SWZ_Y]), + SWZ_STRING(swz[NVS_SWZ_Z]), + SWZ_STRING(swz[NVS_SWZ_W])); + } + + if (shader->GetSourceAbs(shader, merged, i)) + printf(")"); + if (shader->GetSourceScale) { + int scale = shader->GetSourceScale(shader, merged, i); + if (scale > 1) + printf("{scaled %dx}", scale); + } + if (i < (opi->numsrc - 1)) + printf(", "); + } + + if (shader->IsLastInst(shader)) + printf(" + END"); +} + +void +nvsDisasmHWShader(nvsPtr nvs) +{ + nvsFunc *shader = nvs->func; + unsigned int iaddr = 0; + + if (!nvs->program) { + fprintf(stderr, "No HW program present"); + return; + } + + shader->inst = nvs->program; + while (1) { + if (shader->inst >= (nvs->program + nvs->program_size)) { + fprintf(stderr, "Reached end of program, but HW inst has no END"); + break; + } + + printf("\t0x%08x:\n", shader->inst[0]); + printf("\t0x%08x:\n", shader->inst[1]); + printf("\t0x%08x:\n", shader->inst[2]); + printf("\t0x%08x:", shader->inst[3]); + + printf("\n\t\tINST %d.0: ", iaddr); + nvsDisasmHWShaderOp(shader, 0); + if (shader->HasMergedInst(shader)) { + printf("\n\t\tINST %d.1: ", iaddr); + nvsDisasmHWShaderOp(shader, 1); + } + printf("\n"); + + if (shader->IsLastInst(shader)) + break; + + shader->inst += shader->GetOffsetNext(shader); + iaddr++; + } + + printf("\n"); +} diff --git a/src/mesa/drivers/dri/nouveau/nouveau_shader.h b/src/mesa/drivers/dri/nouveau/nouveau_shader.h new file mode 100644 index 00000000000..baf59d02596 --- /dev/null +++ b/src/mesa/drivers/dri/nouveau/nouveau_shader.h @@ -0,0 +1,362 @@ +#ifndef __SHADER_COMMON_H__ +#define __SHADER_COMMON_H__ + +#include "mtypes.h" + +typedef struct _nvsFunc nvsFunc; + +#define NVS_MAX_TEMPS 32 +#define NVS_MAX_ATTRIBS 16 +#define NVS_MAX_CONSTS 256 +#define NVS_MAX_ADDRESS 2 +#define NVS_MAX_INSNS 4096 + +typedef struct { + enum { + NVS_INSTRUCTION, + } type; + int position; +} nvsFragmentHeader; + +typedef struct _nvs_fragment_list { + struct _nvs_fragment_list *prev; + struct _nvs_fragment_list *next; + nvsFragmentHeader *fragment; +} nvsFragmentList; + +typedef struct _nouveauShader { + union { + struct gl_vertex_program vp; + struct gl_fragment_program fp; + } mesa; + GLcontext *ctx; + nvsFunc *func; + + /* State of the final program */ + GLboolean translated; + GLboolean on_hardware; + unsigned int *program; + unsigned int program_size; + unsigned int program_alloc_size; + unsigned int program_start_id; + unsigned int program_current; + unsigned int inputs_read; + unsigned int outputs_written; + int inst_count; + + struct { + GLfloat *source_val; /* NULL if invariant */ + float val[4]; + int hw_index; /* hw-specific value */ + } params[NVS_MAX_CONSTS]; + + struct { + int last_use; + } temps[NVS_MAX_TEMPS]; + + /* Pass-private data */ + void *pass_rec; + + nvsFragmentList *list_head; + nvsFragmentList *list_tail; +} nouveauShader, *nvsPtr; + +typedef enum { + NVS_FILE_NONE, + NVS_FILE_TEMP, + NVS_FILE_ATTRIB, + NVS_FILE_CONST, + NVS_FILE_RESULT, + NVS_FILE_ADDRESS, + NVS_FILE_UNKNOWN +} nvsRegFile; + +typedef enum { + NVS_OP_UNKNOWN = 0, + NVS_OP_NOP, + NVS_OP_ABS, NVS_OP_ADD, NVS_OP_ARA, NVS_OP_ARL, NVS_OP_ARR, + NVS_OP_BRA, NVS_OP_BRK, + NVS_OP_CAL, NVS_OP_CMP, NVS_OP_COS, + NVS_OP_DDX, NVS_OP_DDY, NVS_OP_DIV, NVS_OP_DP2, NVS_OP_DP2A, NVS_OP_DP3, + NVS_OP_DP4, NVS_OP_DPH, NVS_OP_DST, + NVS_OP_EX2, NVS_OP_EXP, + NVS_OP_FLR, NVS_OP_FRC, + NVS_OP_IF, + NVS_OP_KIL, + NVS_OP_LG2, NVS_OP_LIT, NVS_OP_LOG, NVS_OP_LOOP, NVS_OP_LRP, + NVS_OP_MAD, NVS_OP_MAX, NVS_OP_MIN, NVS_OP_MOV, NVS_OP_MUL, + NVS_OP_NRM, + NVS_OP_PK2H, NVS_OP_PK2US, NVS_OP_PK4B, NVS_OP_PK4UB, NVS_OP_POW, + NVS_OP_POPA, NVS_OP_PUSHA, + NVS_OP_RCC, NVS_OP_RCP, NVS_OP_REP, NVS_OP_RET, NVS_OP_RFL, NVS_OP_RSQ, + NVS_OP_SCS, NVS_OP_SEQ, NVS_OP_SFL, NVS_OP_SGE, NVS_OP_SGT, NVS_OP_SIN, + NVS_OP_SLE, NVS_OP_SLT, NVS_OP_SNE, NVS_OP_SSG, NVS_OP_STR, NVS_OP_SUB, + NVS_OP_SWZ, + NVS_OP_TEX, NVS_OP_TXB, NVS_OP_TXD, NVS_OP_TXL, NVS_OP_TXP, + NVS_OP_UP2H, NVS_OP_UP2US, NVS_OP_UP4B, NVS_OP_UP4UB, + NVS_OP_X2D, NVS_OP_XPD, + NVS_OP_EMUL +} nvsOpcode; + +typedef enum { + NVS_PREC_FLOAT32, + NVS_PREC_FLOAT16, + NVS_PREC_FIXED12, + NVS_PREC_UNKNOWN +} nvsPrecision; + +typedef enum { + NVS_SWZ_X = 0, + NVS_SWZ_Y = 1, + NVS_SWZ_Z = 2, + NVS_SWZ_W = 3 +} nvsSwzComp; + +typedef enum { + NVS_FR_POSITION, + NVS_FR_WEIGHT, + NVS_FR_NORMAL, + NVS_FR_COL0, + NVS_FR_COL1, + NVS_FR_BFC0, + NVS_FR_BFC1, + NVS_FR_FOGCOORD, + NVS_FR_POINTSZ, + NVS_FR_TEXCOORD0, + NVS_FR_TEXCOORD1, + NVS_FR_TEXCOORD2, + NVS_FR_TEXCOORD3, + NVS_FR_TEXCOORD4, + NVS_FR_TEXCOORD5, + NVS_FR_TEXCOORD6, + NVS_FR_TEXCOORD7, + NVS_FR_FRAGDATA0, + NVS_FR_FRAGDATA1, + NVS_FR_FRAGDATA2, + NVS_FR_FRAGDATA3, + NVS_FR_CLIP0, + NVS_FR_CLIP1, + NVS_FR_CLIP2, + NVS_FR_CLIP3, + NVS_FR_CLIP4, + NVS_FR_CLIP5, + NVS_FR_CLIP6, + NVS_FR_FACING, + NVS_FR_UNKNOWN +} nvsFixedReg; + +typedef enum { + NVS_COND_FL, NVS_COND_LT, NVS_COND_EQ, NVS_COND_LE, NVS_COND_GT, + NVS_COND_NE, NVS_COND_GE, NVS_COND_TR, NVS_COND_UN, + NVS_COND_UNKNOWN +} nvsCond; + +typedef struct { + nvsRegFile file; + unsigned int index; + + unsigned int indexed; + unsigned int addr_reg; + nvsSwzComp addr_comp; + + nvsSwzComp swizzle[4]; + int negate; + int abs; +} nvsRegister; + +static const nvsRegister nvr_unused = { + .file = NVS_FILE_ATTRIB, + .index = 0, + .indexed = 0, + .addr_reg = 0, + .addr_comp = NVS_SWZ_X, + .swizzle = {NVS_SWZ_X, NVS_SWZ_Y, NVS_SWZ_Z, NVS_SWZ_W}, + .negate = 0, + .abs = 0, +}; + +typedef enum { + NVS_TEX_TARGET_1D, + NVS_TEX_TARGET_2D, + NVS_TEX_TARGET_3D, + NVS_TEX_TARGET_CUBE, + NVS_TEX_TARGET_RECT, + NVS_TEX_TARGET_UNKNOWN = 0 +} nvsTexTarget; + +typedef struct { + nvsFragmentHeader header; + + nvsOpcode op; + unsigned int saturate; + + nvsRegister dest; + unsigned int mask; + + nvsRegister src[3]; + + unsigned int tex_unit; + nvsTexTarget tex_target; + + nvsCond cond; + nvsSwzComp cond_swizzle[4]; + int cond_reg; + int cond_test; + int cond_update; +} nvsInstruction; + +#define SMASK_X (1<<0) +#define SMASK_Y (1<<1) +#define SMASK_Z (1<<2) +#define SMASK_W (1<<3) +#define SMASK_ALL (SMASK_X|SMASK_Y|SMASK_Z|SMASK_W) + +#define SPOS_ADDRESS 3 +struct _op_xlat { + unsigned int NV; + nvsOpcode SOP; + int srcpos[3]; +}; +#define MOD_OPCODE(t,hw,sop,s0,s1,s2) do { \ + t[hw].NV = hw; \ + t[hw].SOP = sop; \ + t[hw].srcpos[0] = s0; \ + t[hw].srcpos[1] = s1; \ + t[hw].srcpos[2] = s2; \ +} while(0) + +extern unsigned int NVVP_TX_VOP_COUNT; +extern unsigned int NVVP_TX_NVS_OP_COUNT; +extern struct _op_xlat NVVP_TX_VOP[]; +extern struct _op_xlat NVVP_TX_SOP[]; + +extern unsigned int NVFP_TX_AOP_COUNT; +extern unsigned int NVFP_TX_BOP_COUNT; +extern struct _op_xlat NVFP_TX_AOP[]; +extern struct _op_xlat NVFP_TX_BOP[]; + +#define SCAP_SRC_ABS (1<<0) + +struct _nvsFunc { + unsigned int MaxInst; + unsigned int MaxAttrib; + unsigned int MaxTemp; + unsigned int MaxAddress; + unsigned int MaxConst; + unsigned int caps; + + unsigned int *inst; + void (*UploadToHW) (GLcontext *, nouveauShader *); + void (*UpdateConst) (GLcontext *, nouveauShader *, int); + + struct _op_xlat*(*GetOPTXRec) (nvsFunc *, int merged); + struct _op_xlat*(*GetOPTXFromSOP) (nvsOpcode, int *id); + + int (*SupportsOpcode) (nvsFunc *, nvsOpcode); + void (*SetOpcode) (nvsFunc *, unsigned int opcode, + int slot); + void (*SetCCUpdate) (nvsFunc *); + void (*SetCondition) (nvsFunc *, int on, nvsCond, int reg, + nvsSwzComp *swizzle); + void (*SetResult) (nvsFunc *, nvsRegister *, + unsigned int mask, int slot); + void (*SetSource) (nvsFunc *, nvsRegister *, int pos); + void (*SetUnusedSource) (nvsFunc *, int pos); + void (*SetTexImageUnit) (nvsFunc *, int unit); + void (*SetSaturate) (nvsFunc *); + void (*SetLastInst) (nvsFunc *); + + int (*HasMergedInst) (nvsFunc *); + int (*IsLastInst) (nvsFunc *); + int (*GetOffsetNext) (nvsFunc *); + + int (*GetOpcodeSlot) (nvsFunc *, int merged); + unsigned int (*GetOpcodeHW) (nvsFunc *, int slot); + nvsOpcode (*GetOpcode) (nvsFunc *, int merged); + + nvsPrecision (*GetPrecision) (nvsFunc *); + int (*GetSaturate) (nvsFunc *); + + nvsRegFile (*GetDestFile) (nvsFunc *, int merged); + unsigned int (*GetDestID) (nvsFunc *, int merged); + unsigned int (*GetDestMask) (nvsFunc *, int merged); + + unsigned int (*GetSourceHW) (nvsFunc *, int merged, int pos); + nvsRegFile (*GetSourceFile) (nvsFunc *, int merged, int pos); + int (*GetSourceID) (nvsFunc *, int merged, int pos); + int (*GetTexImageUnit) (nvsFunc *); + int (*GetSourceNegate) (nvsFunc *, int merged, int pos); + int (*GetSourceAbs) (nvsFunc *, int merged, int pos); + void (*GetSourceSwizzle) (nvsFunc *, int merged, int pos, + nvsSwzComp *swz); + int (*GetSourceIndexed) (nvsFunc *, int merged, int pos); + void (*GetSourceConstVal) (nvsFunc *, int merged, int pos, + float *val); + int (*GetSourceScale) (nvsFunc *, int merged, int pos); + + int (*GetRelAddressRegID) (nvsFunc *); + nvsSwzComp (*GetRelAddressSwizzle) (nvsFunc *); + + int (*SupportsConditional) (nvsFunc *); + int (*GetConditionUpdate) (nvsFunc *); + int (*GetConditionTest) (nvsFunc *); + nvsCond (*GetCondition) (nvsFunc *); + void (*GetCondRegSwizzle) (nvsFunc *, nvsSwzComp *swz); + int (*GetCondRegID) (nvsFunc *); + int (*GetBranch) (nvsFunc *); + int (*GetBranchElse) (nvsFunc *); + int (*GetBranchEnd) (nvsFunc *); + + int (*GetLoopCount) (nvsFunc *); + int (*GetLoopInitial) (nvsFunc *); + int (*GetLoopIncrement) (nvsFunc *); +}; + +static inline nvsRegister +nvsNegate(nvsRegister reg) +{ + reg.negate = !reg.negate; + return reg; +} + +static inline nvsRegister +nvsAbs(nvsRegister reg) +{ + reg.abs = 1; + return reg; +} + +static inline nvsRegister +nvsSwizzle(nvsRegister reg, nvsSwzComp x, nvsSwzComp y, + nvsSwzComp z, nvsSwzComp w) +{ + nvsSwzComp sc[4] = { x, y, z, w }; + nvsSwzComp oc[4]; + int i; + + for (i=0;i<4;i++) + oc[i] = reg.swizzle[i]; + for (i=0;i<4;i++) + reg.swizzle[i] = oc[sc[i]]; + return reg; +} + +extern GLboolean nvsUpdateShader(GLcontext *ctx, nouveauShader *nvs); +extern void nvsDisasmHWShader(nvsPtr); + +extern void NV20VPInitShaderFuncs(nvsFunc *); +extern void NV30VPInitShaderFuncs(nvsFunc *); +extern void NV40VPInitShaderFuncs(nvsFunc *); + +extern void NV30FPInitShaderFuncs(nvsFunc *); +extern void NV40FPInitShaderFuncs(nvsFunc *); + +extern void nouveauShaderInitFuncs(GLcontext *ctx); + +extern GLboolean nouveau_shader_pass0_arb(GLcontext *ctx, nouveauShader *nvs); +extern GLboolean nouveau_shader_pass0_slang(GLcontext *ctx, nouveauShader *nvs); +extern GLboolean nouveau_shader_pass1(nvsPtr nvs); +extern GLboolean nouveau_shader_pass2(nvsPtr nvs); + +#endif + diff --git a/src/mesa/drivers/dri/nouveau/nouveau_shader_0_arb.c b/src/mesa/drivers/dri/nouveau/nouveau_shader_0_arb.c new file mode 100644 index 00000000000..8b5222d0693 --- /dev/null +++ b/src/mesa/drivers/dri/nouveau/nouveau_shader_0_arb.c @@ -0,0 +1,694 @@ +/* + * Copyright (C) 2006 Ben Skeggs. + * + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial + * portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + */ + +/* + * Authors: + * Ben Skeggs + */ + +#include "glheader.h" +#include "macros.h" +#include "enums.h" + +#include "program.h" +#include "programopt.h" +#include "program_instruction.h" + +#include "nouveau_context.h" +#include "nouveau_shader.h" + +static nvsFixedReg _tx_mesa_vp_dst_reg[VERT_RESULT_MAX] = { + NVS_FR_POSITION, NVS_FR_COL0, NVS_FR_COL1, NVS_FR_FOGCOORD, + NVS_FR_TEXCOORD0, NVS_FR_TEXCOORD1, NVS_FR_TEXCOORD2, NVS_FR_TEXCOORD3, + NVS_FR_TEXCOORD4, NVS_FR_TEXCOORD5, NVS_FR_TEXCOORD6, NVS_FR_TEXCOORD7, + NVS_FR_POINTSZ, NVS_FR_BFC0, NVS_FR_BFC1, NVS_FR_UNKNOWN /* EDGE */ +}; + +static nvsFixedReg _tx_mesa_fp_dst_reg[FRAG_RESULT_MAX] = { + NVS_FR_FRAGDATA0 /* COLR */, NVS_FR_FRAGDATA0 /* COLH */, + NVS_FR_UNKNOWN /* DEPR */ +}; + +static nvsFixedReg _tx_mesa_vp_src_reg[VERT_ATTRIB_MAX] = { + NVS_FR_POSITION, NVS_FR_WEIGHT, NVS_FR_NORMAL, NVS_FR_COL0, NVS_FR_COL1, + NVS_FR_FOGCOORD, NVS_FR_UNKNOWN /* COLOR_INDEX */, NVS_FR_UNKNOWN, + NVS_FR_TEXCOORD0, NVS_FR_TEXCOORD1, NVS_FR_TEXCOORD2, NVS_FR_TEXCOORD3, + NVS_FR_TEXCOORD4, NVS_FR_TEXCOORD5, NVS_FR_TEXCOORD6, NVS_FR_TEXCOORD7, +/* Generic attribs 0-15, aliased to the above */ + NVS_FR_POSITION, NVS_FR_WEIGHT, NVS_FR_NORMAL, NVS_FR_COL0, NVS_FR_COL1, + NVS_FR_FOGCOORD, NVS_FR_UNKNOWN /* COLOR_INDEX */, NVS_FR_UNKNOWN, + NVS_FR_TEXCOORD0, NVS_FR_TEXCOORD1, NVS_FR_TEXCOORD2, NVS_FR_TEXCOORD3, + NVS_FR_TEXCOORD4, NVS_FR_TEXCOORD5, NVS_FR_TEXCOORD6, NVS_FR_TEXCOORD7 +}; + +static nvsFixedReg _tx_mesa_fp_src_reg[FRAG_ATTRIB_MAX] = { + NVS_FR_POSITION, NVS_FR_COL0, NVS_FR_COL1, NVS_FR_FOGCOORD, + NVS_FR_TEXCOORD0, NVS_FR_TEXCOORD1, NVS_FR_TEXCOORD2, NVS_FR_TEXCOORD3, + NVS_FR_TEXCOORD4, NVS_FR_TEXCOORD5, NVS_FR_TEXCOORD6, NVS_FR_TEXCOORD7 +}; + +static nvsSwzComp _tx_mesa_swizzle[4] = { + NVS_SWZ_X, NVS_SWZ_Y, NVS_SWZ_Z, NVS_SWZ_W +}; + +static nvsOpcode _tx_mesa_opcode[] = { + [OPCODE_ABS] = NVS_OP_ABS, [OPCODE_ADD] = NVS_OP_ADD, + [OPCODE_ARA] = NVS_OP_ARA, [OPCODE_ARL] = NVS_OP_ARL, + [OPCODE_ARL_NV] = NVS_OP_ARL, [OPCODE_ARR] = NVS_OP_ARR, + [OPCODE_CMP] = NVS_OP_CMP, [OPCODE_COS] = NVS_OP_COS, + [OPCODE_DDX] = NVS_OP_DDX, [OPCODE_DDY] = NVS_OP_DDY, + [OPCODE_DP3] = NVS_OP_DP3, [OPCODE_DP4] = NVS_OP_DP4, + [OPCODE_DPH] = NVS_OP_DPH, [OPCODE_DST] = NVS_OP_DST, + [OPCODE_EX2] = NVS_OP_EX2, [OPCODE_EXP] = NVS_OP_EXP, + [OPCODE_FLR] = NVS_OP_FLR, [OPCODE_FRC] = NVS_OP_FRC, + [OPCODE_KIL] = NVS_OP_EMUL, [OPCODE_KIL_NV] = NVS_OP_KIL, + [OPCODE_LG2] = NVS_OP_LG2, [OPCODE_LIT] = NVS_OP_LIT, + [OPCODE_LOG] = NVS_OP_LOG, + [OPCODE_LRP] = NVS_OP_LRP, + [OPCODE_MAD] = NVS_OP_MAD, [OPCODE_MAX] = NVS_OP_MAX, + [OPCODE_MIN] = NVS_OP_MIN, [OPCODE_MOV] = NVS_OP_MOV, + [OPCODE_MUL] = NVS_OP_MUL, + [OPCODE_PK2H] = NVS_OP_PK2H, [OPCODE_PK2US] = NVS_OP_PK2US, + [OPCODE_PK4B] = NVS_OP_PK4B, [OPCODE_PK4UB] = NVS_OP_PK4UB, + [OPCODE_POW] = NVS_OP_POW, [OPCODE_POPA] = NVS_OP_POPA, + [OPCODE_PUSHA] = NVS_OP_PUSHA, + [OPCODE_RCC] = NVS_OP_RCC, [OPCODE_RCP] = NVS_OP_RCP, + [OPCODE_RFL] = NVS_OP_RFL, [OPCODE_RSQ] = NVS_OP_RSQ, + [OPCODE_SCS] = NVS_OP_SCS, [OPCODE_SEQ] = NVS_OP_SEQ, + [OPCODE_SFL] = NVS_OP_SFL, [OPCODE_SGE] = NVS_OP_SGE, + [OPCODE_SGT] = NVS_OP_SGT, [OPCODE_SIN] = NVS_OP_SIN, + [OPCODE_SLE] = NVS_OP_SLE, [OPCODE_SLT] = NVS_OP_SLT, + [OPCODE_SNE] = NVS_OP_SNE, [OPCODE_SSG] = NVS_OP_SSG, + [OPCODE_STR] = NVS_OP_STR, [OPCODE_SUB] = NVS_OP_SUB, + [OPCODE_SWZ] = NVS_OP_MOV, + [OPCODE_TEX] = NVS_OP_TEX, [OPCODE_TXB] = NVS_OP_TXB, + [OPCODE_TXD] = NVS_OP_TXD, + [OPCODE_TXL] = NVS_OP_TXL, [OPCODE_TXP] = NVS_OP_TXP, + [OPCODE_TXP_NV] = NVS_OP_TXP, + [OPCODE_UP2H] = NVS_OP_UP2H, [OPCODE_UP2US] = NVS_OP_UP2US, + [OPCODE_UP4B] = NVS_OP_UP4B, [OPCODE_UP4UB] = NVS_OP_UP4UB, + [OPCODE_X2D] = NVS_OP_X2D, + [OPCODE_XPD] = NVS_OP_XPD +}; + +static nvsCond _tx_mesa_condmask[] = { + NVS_COND_UNKNOWN, NVS_COND_GT, NVS_COND_LT, NVS_COND_UN, NVS_COND_GE, + NVS_COND_LE, NVS_COND_NE, NVS_COND_NE, NVS_COND_TR, NVS_COND_FL +}; + +struct pass0_rec { + int nvs_ipos; + int next_temp; + int swzconst_done; + int swzconst_id; +}; + +#define X NVS_SWZ_X +#define Y NVS_SWZ_Y +#define Z NVS_SWZ_Z +#define W NVS_SWZ_W + +static void +pass0_append_fragment(nouveauShader *nvs, nvsFragmentHeader *fragment) +{ + nvsFragmentList *list = calloc(1, sizeof(nvsFragmentList)); + if (!list) + return; + + list->fragment = fragment; + list->prev = nvs->list_tail; + if ( nvs->list_tail) + nvs->list_tail->next = list; + if (!nvs->list_head) + nvs->list_head = list; + nvs->list_tail = list; + + nvs->inst_count++; +} + +static void +pass0_make_reg(nouveauShader *nvs, nvsRegister *reg, + nvsRegFile file, unsigned int index) +{ + struct pass0_rec *rec = nvs->pass_rec; + + /* defaults */ + *reg = nvr_unused; + /* -1 == quick-and-dirty temp alloc */ + if (file == NVS_FILE_TEMP && index == -1) { + index = rec->next_temp++; + assert(index < NVS_MAX_TEMPS); + } + reg->file = file; + reg->index = index; +} + +static void +pass0_make_swizzle(nvsSwzComp *swz, unsigned int mesa) +{ + int i; + + for (i=0;i<4;i++) + swz[i] = _tx_mesa_swizzle[GET_SWZ(mesa, i)]; +} + +static nvsOpcode +pass0_make_opcode(enum prog_opcode op) +{ + if (op > MAX_OPCODE) + return NVS_OP_UNKNOWN; + return _tx_mesa_opcode[op]; +} + +static nvsCond +pass0_make_condmask(GLuint mesa) +{ + if (mesa > COND_FL) + return NVS_COND_UNKNOWN; + return _tx_mesa_condmask[mesa]; +} + +static unsigned int +pass0_make_mask(GLuint mesa_mask) +{ + unsigned int mask = 0; + + if (mesa_mask & WRITEMASK_X) mask |= SMASK_X; + if (mesa_mask & WRITEMASK_Y) mask |= SMASK_Y; + if (mesa_mask & WRITEMASK_Z) mask |= SMASK_Z; + if (mesa_mask & WRITEMASK_W) mask |= SMASK_W; + + return mask; +} + +static nvsTexTarget +pass0_make_tex_target(GLuint mesa) +{ + switch (mesa) { + case TEXTURE_1D_INDEX: return NVS_TEX_TARGET_1D; + case TEXTURE_2D_INDEX: return NVS_TEX_TARGET_2D; + case TEXTURE_3D_INDEX: return NVS_TEX_TARGET_3D; + case TEXTURE_CUBE_INDEX: return NVS_TEX_TARGET_CUBE; + case TEXTURE_RECT_INDEX: return NVS_TEX_TARGET_RECT; + default: + return NVS_TEX_TARGET_UNKNOWN; + } +} + +static void +pass0_make_dst_reg(nvsPtr nvs, nvsRegister *reg, + struct prog_dst_register *dst) +{ + struct gl_program *mesa = (struct gl_program*)&nvs->mesa.vp; + nvsFixedReg sfr; + + switch (dst->File) { + case PROGRAM_OUTPUT: + if (mesa->Target == GL_VERTEX_PROGRAM_ARB) { + sfr = (dst->Index < VERT_RESULT_MAX) ? + _tx_mesa_vp_dst_reg[dst->Index] : NVS_FR_UNKNOWN; + } else { + sfr = (dst->Index < FRAG_RESULT_MAX) ? + _tx_mesa_fp_dst_reg[dst->Index] : NVS_FR_UNKNOWN; + } + pass0_make_reg(nvs, reg, NVS_FILE_RESULT, sfr); + break; + case PROGRAM_TEMPORARY: + pass0_make_reg(nvs, reg, NVS_FILE_TEMP, dst->Index); + break; + case PROGRAM_ADDRESS: + pass0_make_reg(nvs, reg, NVS_FILE_ADDRESS, dst->Index); + break; + default: + fprintf(stderr, "Unknown dest file %d\n", dst->File); + assert(0); + } +} + +static void +pass0_make_src_reg(nvsPtr nvs, nvsRegister *reg, struct prog_src_register *src) +{ + struct gl_program *mesa = (struct gl_program *)&nvs->mesa.vp.Base; + struct gl_program_parameter_list *p = mesa->Parameters; + + *reg = nvr_unused; + + switch (src->File) { + case PROGRAM_INPUT: + reg->file = NVS_FILE_ATTRIB; + if (mesa->Target == GL_VERTEX_PROGRAM_ARB) { + reg->index = (src->Index < VERT_ATTRIB_MAX) ? + _tx_mesa_vp_src_reg[src->Index] : NVS_FR_UNKNOWN; + } else { + reg->index = (src->Index < FRAG_ATTRIB_MAX) ? + _tx_mesa_fp_src_reg[src->Index] : NVS_FR_UNKNOWN; + } + break; + /* All const types seem to get shoved into here, not really sure why */ + case PROGRAM_STATE_VAR: + switch (p->Parameters[src->Index].Type) { + case PROGRAM_NAMED_PARAM: + case PROGRAM_CONSTANT: + nvs->params[src->Index].source_val = NULL; + COPY_4V(nvs->params[src->Index].val, p->ParameterValues[src->Index]); + break; + case PROGRAM_STATE_VAR: + nvs->params[src->Index].source_val = p->ParameterValues[src->Index]; + break; + default: + fprintf(stderr, "Unknown parameter type %d\n", + p->Parameters[src->Index].Type); + assert(0); + break; + } + + if (src->RelAddr) { + reg->indexed = 1; + reg->addr_reg = 0; + reg->addr_comp = NVS_SWZ_X; + } else + reg->indexed = 0; + reg->file = NVS_FILE_CONST; + reg->index = src->Index; + break; + case PROGRAM_TEMPORARY: + reg->file = NVS_FILE_TEMP; + reg->index = src->Index; + break; + default: + fprintf(stderr, "Unknown source type %d\n", src->File); + assert(0); + } + + /* per-component negate handled elsewhere */ + reg->negate = src->NegateBase != 0; + reg->abs = src->Abs; + pass0_make_swizzle(reg->swizzle, src->Swizzle); +} + +static nvsInstruction * +pass0_emit(nouveauShader *nvs, nvsOpcode op, nvsRegister dst, + unsigned int mask, int saturate, + nvsRegister src0, nvsRegister src1, nvsRegister src2) +{ + struct pass0_rec *rec = nvs->pass_rec; + nvsInstruction *sif = NULL; + + /* Seems mesa doesn't explicitly 0 this.. */ + if (nvs->mesa.vp.Base.Target == GL_VERTEX_PROGRAM_ARB) + saturate = 0; + + sif = calloc(1, sizeof(nvsInstruction)); + if (sif) { + sif->header.type = NVS_INSTRUCTION; + sif->header.position = rec->nvs_ipos++; + sif->op = op; + sif->saturate = saturate; + sif->dest = dst; + sif->mask = mask; + sif->src[0] = src0; + sif->src[1] = src1; + sif->src[2] = src2; + sif->cond = COND_TR; + sif->cond_reg = 0; + sif->cond_test = 0; + sif->cond_update = 0; + pass0_make_swizzle(sif->cond_swizzle, SWIZZLE_NOOP); + pass0_append_fragment(nvs, (nvsFragmentHeader *)sif); + } + + return sif; +} + +static void +pass0_fixup_swizzle(nvsPtr nvs, + struct prog_src_register *src, + unsigned int sm1, + unsigned int sm2) +{ + static const float sc[4] = { 1.0, 0.0, -1.0, 0.0 }; + struct pass0_rec *rec = nvs->pass_rec; + int fixup_1, fixup_2; + nvsRegister sr, dr = nvr_unused; + nvsRegister sm1const, sm2const; + + if (!rec->swzconst_done) { + struct gl_program *prog = &nvs->mesa.vp.Base; + rec->swzconst_id = _mesa_add_unnamed_constant(prog->Parameters, sc, 4); + rec->swzconst_done = 1; + COPY_4V(nvs->params[rec->swzconst_id].val, sc); + } + + fixup_1 = (sm1 != MAKE_SWIZZLE4(0,0,0,0) && sm2 != MAKE_SWIZZLE4(2,2,2,2)); + fixup_2 = (sm2 != MAKE_SWIZZLE4(2,2,2,2)); + + if (src->File != PROGRAM_TEMPORARY && src->File != PROGRAM_INPUT) { + /* We can't use more than one const in an instruction, so move the const + * into a temp, and swizzle from there. + *TODO: should just emit the swizzled const, instead of swizzling it + * in the shader.. would need to reswizzle any state params when they + * change however.. + */ + pass0_make_reg(nvs, &dr, NVS_FILE_TEMP, -1); + pass0_make_src_reg(nvs, &sr, src); + pass0_emit(nvs, NVS_OP_MOV, dr, SMASK_ALL, 0, sr, nvr_unused, nvr_unused); + pass0_make_reg(nvs, &sr, NVS_FILE_TEMP, dr.index); + } else { + if (fixup_1) + src->NegateBase = 0; + pass0_make_src_reg(nvs, &sr, src); + pass0_make_reg(nvs, &dr, NVS_FILE_TEMP, -1); + } + + pass0_make_reg(nvs, &sm1const, NVS_FILE_CONST, rec->swzconst_id); + pass0_make_swizzle(sm1const.swizzle, sm1); + if (fixup_1 && fixup_2) { + /* Any combination with SWIZZLE_ONE */ + pass0_make_reg(nvs, &sm2const, NVS_FILE_CONST, rec->swzconst_id); + pass0_make_swizzle(sm2const.swizzle, sm2); + pass0_emit(nvs, NVS_OP_MAD, dr, SMASK_ALL, 0, sr, sm1const, sm2const); + } else { + /* SWIZZLE_ZERO || arbitrary negate */ + pass0_emit(nvs, NVS_OP_MUL, dr, SMASK_ALL, 0, sr, sm1const, nvr_unused); + } + + src->File = PROGRAM_TEMPORARY; + src->Index = dr.index; + src->Swizzle = SWIZZLE_NOOP; +} + +#define SET_SWZ(fs, cp, c) fs = (fs & ~(0x7<<(cp*3))) | (c<<(cp*3)) +static void +pass0_check_sources(nvsPtr nvs, struct prog_instruction *inst) +{ + unsigned int insrc = -1, constsrc = -1; + int i; + + for (i=0;i<_mesa_num_inst_src_regs(inst->Opcode);i++) { + struct prog_src_register *src = &inst->SrcReg[i]; + unsigned int sm_1 = 0, sm_2 = 0; + nvsRegister sr, dr; + int do_mov = 0, c; + + /* Build up swizzle masks as if we were going to use + * "MAD new, src, const1, const2" to support arbitrary negation + * and SWIZZLE_ZERO/SWIZZLE_ONE. + */ + for (c=0;c<4;c++) { + if (GET_SWZ(src->Swizzle, c) == SWIZZLE_ZERO) { + SET_SWZ(sm_1, c, SWIZZLE_Y); /* 0.0 */ + SET_SWZ(sm_2, c, SWIZZLE_Y); + SET_SWZ(src->Swizzle, c, SWIZZLE_X); + } else if (GET_SWZ(src->Swizzle, c) == SWIZZLE_ONE) { + SET_SWZ(sm_1, c, SWIZZLE_Y); + if (src->NegateBase & (1<Swizzle, c, SWIZZLE_X); + } else { + if (src->NegateBase & (1<File) { + case PROGRAM_INPUT: + if (insrc != -1 && insrc != src->Index) + do_mov = 1; + else insrc = src->Index; + break; + case PROGRAM_STATE_VAR: + if (constsrc != -1 && constsrc != src->Index) + do_mov = 1; + else constsrc = src->Index; + break; + default: + break; + } + + /* Emit any extra ATTRIB/CONST to a temp, and modify the Mesa instruction + * to point at the temp. + */ + if (do_mov) { + pass0_make_src_reg(nvs, &sr, src); + pass0_make_reg(nvs, &dr, NVS_FILE_TEMP, -1); + pass0_emit(nvs, NVS_OP_MOV, dr, SMASK_ALL, 0, + sr, nvr_unused, nvr_unused); + + src->File = PROGRAM_TEMPORARY; + src->Index = dr.index; + src->Swizzle= SWIZZLE_NOOP; + } + } +} + +static GLboolean +pass0_emulate_instruction(nouveauShader *nvs, struct prog_instruction *inst) +{ + nvsFunc *shader = nvs->func; + nvsRegister src[3], dest, temp; + nvsInstruction *nvsinst; + unsigned int mask = pass0_make_mask(inst->DstReg.WriteMask); + int i, sat; + + sat = (inst->SaturateMode == SATURATE_ZERO_ONE); + + /* Build all the "real" regs for the instruction */ + for (i=0; i<_mesa_num_inst_src_regs(inst->Opcode); i++) + pass0_make_src_reg(nvs, &src[i], &inst->SrcReg[i]); + if (inst->Opcode != OPCODE_KIL) + pass0_make_dst_reg(nvs, &dest, &inst->DstReg); + + switch (inst->Opcode) { + case OPCODE_ABS: + if (shader->caps & SCAP_SRC_ABS) + pass0_emit(nvs, NVS_OP_MOV, dest, mask, sat, + nvsAbs(src[0]), nvr_unused, nvr_unused); + else + pass0_emit(nvs, NVS_OP_MAX, dest, mask, sat, + src[0], nvsNegate(src[0]), nvr_unused); + break; + case OPCODE_KIL: + /* This is only in ARB shaders, so we don't have to worry + * about clobbering a CC reg as they aren't supported anyway. + */ + /* MOVC0 temp, src */ + pass0_make_reg(nvs, &temp, NVS_FILE_TEMP, -1); + nvsinst = pass0_emit(nvs, NVS_OP_MOV, temp, SMASK_ALL, 0, + src[0], nvr_unused, nvr_unused); + nvsinst->cond_update = 1; + nvsinst->cond_reg = 0; + /* KIL_NV (LT0.xyzw) temp */ + nvsinst = pass0_emit(nvs, NVS_OP_KIL, nvr_unused, 0, 0, + nvr_unused, nvr_unused, nvr_unused); + nvsinst->cond = COND_LT; + nvsinst->cond_reg = 0; + nvsinst->cond_test = 1; + pass0_make_swizzle(nvsinst->cond_swizzle, MAKE_SWIZZLE4(0,1,2,3)); + break; + case OPCODE_LIT: + break; + case OPCODE_LRP: + pass0_make_reg(nvs, &temp, NVS_FILE_TEMP, -1); + pass0_emit(nvs, NVS_OP_MAD, temp, mask, 0, + nvsNegate(src[0]), src[2], src[2]); + pass0_emit(nvs, NVS_OP_MAD, dest, mask, sat, + src[0], src[1], temp); + break; + case OPCODE_POW: + if (shader->SupportsOpcode(shader, NVS_OP_LG2) && + shader->SupportsOpcode(shader, NVS_OP_EX2)) { + pass0_make_reg(nvs, &temp, NVS_FILE_TEMP, -1); + /* LG2 temp.x, src0.c */ + pass0_emit(nvs, NVS_OP_LG2, temp, SMASK_X, 0, + nvsSwizzle(src[0], X, X, X, X), + nvr_unused, + nvr_unused); + /* MUL temp.x, temp.x, src1.c */ + pass0_emit(nvs, NVS_OP_MUL, temp, SMASK_X, 0, + nvsSwizzle(temp, X, X, X, X), + nvsSwizzle(src[1], X, X, X, X), + nvr_unused); + /* EX2 dest, temp.x */ + pass0_emit(nvs, NVS_OP_EX2, dest, mask, sat, + nvsSwizzle(temp, X, X, X, X), + nvr_unused, + nvr_unused); + } else { + /* can we use EXP/LOG instead of EX2/LG2?? */ + fprintf(stderr, "Implement POW for NV20 vtxprog!\n"); + return GL_FALSE; + } + break; + case OPCODE_RSQ: + pass0_make_reg(nvs, &temp, NVS_FILE_TEMP, -1); + pass0_emit(nvs, NVS_OP_LG2, temp, SMASK_X, 0, + nvsAbs(nvsSwizzle(src[0], X, X, X, X)), nvr_unused, nvr_unused); + pass0_emit(nvs, NVS_OP_EX2, dest, mask, sat, + nvsSwizzle(temp, X, X, X, X), nvr_unused, nvr_unused); + break; + case OPCODE_SCS: + if (mask & SMASK_X) + pass0_emit(nvs, NVS_OP_COS, dest, SMASK_X, sat, + nvsSwizzle(src[0], X, X, X, X), + nvr_unused, + nvr_unused); + if (mask & SMASK_Y) + pass0_emit(nvs, NVS_OP_SIN, dest, SMASK_Y, sat, + nvsSwizzle(src[0], X, X, X, X), + nvr_unused, + nvr_unused); + break; + case OPCODE_SUB: + pass0_emit(nvs, NVS_OP_ADD, dest, mask, sat, + src[0], nvsNegate(src[1]), nvr_unused); + break; + case OPCODE_XPD: + pass0_make_reg(nvs, &temp, NVS_FILE_TEMP, -1); + pass0_emit(nvs, NVS_OP_MUL, temp, SMASK_ALL, 0, + nvsSwizzle(src[0], Z, X, Y, Y), + nvsSwizzle(src[1], Y, Z, X, X), + nvr_unused); + pass0_emit(nvs, NVS_OP_MAD, dest, (mask & ~SMASK_W), sat, + nvsSwizzle(src[0], Y, Z, X, X), + nvsSwizzle(src[1], Z, X, Y, Y), + nvsNegate(temp)); + break; + default: + fprintf(stderr, "hw doesn't support opcode \"%s\", and no emulation found\n", + _mesa_opcode_string(inst->Opcode)); + return GL_FALSE; + } + + return GL_TRUE; +} + +static GLboolean +pass0_translate_instructions(nouveauShader *nvs) +{ + struct gl_program *prog = (struct gl_program *)&nvs->mesa.vp; + struct pass0_rec *rec = nvs->pass_rec; + nvsFunc *shader = nvs->func; + int ipos; + + for (ipos=0; iposNumInstructions; ipos++) { + struct prog_instruction *inst = &prog->Instructions[ipos]; + + if (inst->Opcode == OPCODE_END) + break; + + /* Deal with multiple ATTRIB/PARAM in a single instruction */ + pass0_check_sources(nvs, inst); + + /* Now it's safe to do the prog_instruction->nvsInstruction conversion */ + if (shader->SupportsOpcode(shader, pass0_make_opcode(inst->Opcode))) { + nvsInstruction *nvsinst; + nvsRegister src[3], dest; + int i; + + for (i=0; i<_mesa_num_inst_src_regs(inst->Opcode); i++) + pass0_make_src_reg(nvs, &src[i], &inst->SrcReg[i]); + pass0_make_dst_reg(nvs, &dest, &inst->DstReg); + + nvsinst = pass0_emit(nvs, + pass0_make_opcode(inst->Opcode), + dest, + pass0_make_mask(inst->DstReg.WriteMask), + (inst->SaturateMode != SATURATE_OFF), + src[0], src[1], src[2]); + nvsinst->tex_unit = inst->TexSrcUnit; + nvsinst->tex_target = pass0_make_tex_target(inst->TexSrcTarget); + /* TODO when NV_fp/vp is implemented */ + nvsinst->cond = COND_TR; + } else { + if (!pass0_emulate_instruction(nvs, inst)) + return GL_FALSE; + } + } + + return GL_TRUE; +} + +GLboolean +nouveau_shader_pass0_arb(GLcontext *ctx, nouveauShader *nvs) +{ + nouveauContextPtr nmesa = NOUVEAU_CONTEXT(ctx); + struct gl_program *prog = (struct gl_program*)nvs; + struct gl_vertex_program *vp = (struct gl_vertex_program *)prog; + struct gl_fragment_program *fp = (struct gl_fragment_program *)prog; + struct pass0_rec *rec; + int ret; + + switch (prog->Target) { + case GL_VERTEX_PROGRAM_ARB: + nvs->func = &nmesa->VPfunc; + if (vp->IsPositionInvariant) + _mesa_insert_mvp_code(ctx, vp); +#if 0 + if (IS_FIXEDFUNCTION_PROG && CLIP_PLANES_USED) + pass0_insert_ff_clip_planes(); +#endif + break; + case GL_FRAGMENT_PROGRAM_ARB: + nvs->func = &nmesa->FPfunc; + if (fp->FogOption != GL_NONE) + _mesa_append_fog_code(ctx, fp); + break; + default: + fprintf(stderr, "Unknown program type %d", prog->Target); + return GL_FALSE; + } + + rec = calloc(1, sizeof(struct pass0_rec)); + rec->next_temp = prog->NumTemporaries; + nvs->pass_rec = rec; + + ret = pass0_translate_instructions(nvs); + if (!ret) { + /* DESTROY list */ + } + + free(nvs->pass_rec); + return ret; +} + diff --git a/src/mesa/drivers/dri/nouveau/nouveau_shader_1.c b/src/mesa/drivers/dri/nouveau/nouveau_shader_1.c new file mode 100644 index 00000000000..5de9017f580 --- /dev/null +++ b/src/mesa/drivers/dri/nouveau/nouveau_shader_1.c @@ -0,0 +1,318 @@ +/* + * Copyright (C) 2006 Ben Skeggs. + * + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial + * portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + */ + +/* + * Authors: + * Ben Skeggs + */ + +#include "glheader.h" +#include "macros.h" +#include "enums.h" + +#include "nouveau_shader.h" + +#define PASS1_OK 0 +#define PASS1_KILL 1 +#define PASS1_FAIL 2 + +struct pass1_rec { + unsigned int temp[NVS_MAX_TEMPS]; + unsigned int result[NVS_MAX_ATTRIBS]; + unsigned int address[NVS_MAX_ADDRESS]; + unsigned int cc[2]; +}; + +static void +pass1_remove_fragment(nvsPtr nvs, nvsFragmentList *item) +{ + if (item->prev) item->prev->next = item->next; + if (item->next) item->next->prev = item->prev; + if (nvs->list_head == item) nvs->list_head = item->next; + if (nvs->list_tail == item) nvs->list_tail = item->prev; + + nvs->inst_count--; +} + +static int +pass1_result_needed(struct pass1_rec *rec, nvsInstruction *inst) +{ + if (inst->cond_update && rec->cc[inst->cond_reg]) + return 1; + /* Only write components that are read later */ + if (inst->dest.file == NVS_FILE_TEMP) + return (inst->mask & rec->temp[inst->dest.index]); + if (inst->dest.file == NVS_FILE_ADDRESS) + return (inst->mask & rec->address[inst->dest.index]); + /* No point writing result components that are written later */ + if (inst->dest.file == NVS_FILE_RESULT) + return (inst->mask & ~rec->result[inst->dest.index]); + assert(0); +} + +static void +pass1_track_result(struct pass1_rec *rec, nvsInstruction *inst) +{ + if (inst->cond_test) + rec->cc[inst->cond_reg] = 1; + if (inst->dest.file == NVS_FILE_TEMP) { + inst->mask &= rec->temp[inst->dest.index]; + } else if (inst->dest.file == NVS_FILE_RESULT) { + inst->mask &= ~rec->result[inst->dest.index]; + rec->result[inst->dest.index] |= inst->mask; + } else if (inst->dest.file == NVS_FILE_ADDRESS) { + inst->mask &= rec->address[inst->dest.index]; + } +} + +static void +pass1_track_source(nouveauShader *nvs, nvsInstruction *inst, int pos, + unsigned int read) +{ + struct pass1_rec *rec = nvs->pass_rec; + nvsRegister *src = &inst->src[pos]; + unsigned int really_read = 0; + int i,sc; + + /* Account for swizzling */ + for (i=0; i<4; i++) + if (read & (1<swizzle[i]); + + /* Track register reads */ + if (src->file == NVS_FILE_TEMP) { + if (nvs->temps[src->index].last_use == -1) + nvs->temps[src->index].last_use = inst->header.position; + rec->temp [src->index] |= really_read; + } else if (src->indexed) { + rec->address[src->addr_reg] |= (1<addr_comp); + } + + /* Modify swizzle to only access read components */ + /* Find a component that is used.. */ + for (sc=0;sc<4;sc++) + if (really_read & (1<swizzle[i] = sc; +} + +static int +pass1_check_instruction(nouveauShader *nvs, nvsInstruction *inst) +{ + struct pass1_rec *rec = nvs->pass_rec; + unsigned int read0, read1, read2; + + if (inst->op != NVS_OP_KIL) { + if (!pass1_result_needed(rec, inst)) + return PASS1_KILL; + } + pass1_track_result(rec, inst); + + read0 = read1 = read2 = 0; + + switch (inst->op) { + case NVS_OP_FLR: + case NVS_OP_FRC: + case NVS_OP_MOV: + case NVS_OP_SSG: + case NVS_OP_ARL: + read0 = inst->mask; + break; + case NVS_OP_ADD: + case NVS_OP_MAX: + case NVS_OP_MIN: + case NVS_OP_MUL: + case NVS_OP_SEQ: + case NVS_OP_SFL: + case NVS_OP_SGE: + case NVS_OP_SGT: + case NVS_OP_SLE: + case NVS_OP_SLT: + case NVS_OP_SNE: + case NVS_OP_STR: + case NVS_OP_SUB: + read0 = inst->mask; + read1 = inst->mask; + break; + case NVS_OP_CMP: + case NVS_OP_LRP: + case NVS_OP_MAD: + read0 = inst->mask; + read1 = inst->mask; + read2 = inst->mask; + break; + case NVS_OP_XPD: + if (inst->mask & SMASK_X) read0 |= SMASK_Y|SMASK_Z; + if (inst->mask & SMASK_Y) read0 |= SMASK_X|SMASK_Z; + if (inst->mask & SMASK_Z) read0 |= SMASK_X|SMASK_Y; + read1 = read0; + break; + case NVS_OP_COS: + case NVS_OP_EX2: + case NVS_OP_EXP: + case NVS_OP_LG2: + case NVS_OP_LOG: + case NVS_OP_RCC: + case NVS_OP_RCP: + case NVS_OP_RSQ: + case NVS_OP_SCS: + case NVS_OP_SIN: + read0 = SMASK_X; + break; + case NVS_OP_POW: + read0 = SMASK_X; + read1 = SMASK_X; + break; + case NVS_OP_DIV: + read0 = inst->mask; + read1 = SMASK_X; + break; + case NVS_OP_DP2: + read0 = SMASK_X|SMASK_Y; + read1 = SMASK_X|SMASK_Y; + break; + case NVS_OP_DP3: + case NVS_OP_RFL: + read0 = SMASK_X|SMASK_Y|SMASK_Z; + read1 = SMASK_X|SMASK_Y|SMASK_Z; + break; + case NVS_OP_DP4: + read0 = SMASK_ALL; + read1 = SMASK_ALL; + break; + case NVS_OP_DPH: + read0 = SMASK_X|SMASK_Y|SMASK_Z; + read1 = SMASK_ALL; + break; + case NVS_OP_DST: + if (inst->mask & SMASK_Y) read0 = read1 = SMASK_Y; + if (inst->mask & SMASK_Z) read0 |= SMASK_Z; + if (inst->mask & SMASK_W) read1 |= SMASK_W; + break; + case NVS_OP_NRM: + read0 = SMASK_X|SMASK_Y|SMASK_Z; + break; + case NVS_OP_PK2H: + case NVS_OP_PK2US: + read0 = SMASK_X|SMASK_Y; + break; + case NVS_OP_DDX: + case NVS_OP_DDY: + case NVS_OP_UP2H: + case NVS_OP_UP2US: + case NVS_OP_PK4B: + case NVS_OP_PK4UB: + case NVS_OP_UP4B: + case NVS_OP_UP4UB: + read0 = SMASK_ALL; + break; + case NVS_OP_X2D: + read1 = SMASK_X|SMASK_Y; + if (inst->mask & (SMASK_X|SMASK_Z)) { + read0 |= SMASK_X; + read2 |= SMASK_X|SMASK_Y; + } + if (inst->mask & (SMASK_Y|SMASK_W)) { + read0 |= SMASK_Y; + read2 |= SMASK_Z|SMASK_W; + } + break; + case NVS_OP_LIT: + read0 |= SMASK_X|SMASK_Y|SMASK_W; + break; + case NVS_OP_TEX: + case NVS_OP_TXP: + case NVS_OP_TXL: + case NVS_OP_TXB: + read0 = SMASK_ALL; + break; + case NVS_OP_TXD: + read0 = SMASK_ALL; + read1 = SMASK_ALL; + read2 = SMASK_ALL; + break; + case NVS_OP_KIL: + break; + default: + fprintf(stderr, "Unknown sop=%d", inst->op); + return PASS1_FAIL; + } + + /* Any values that are written by this inst can't have been read further up */ + if (inst->dest.file == NVS_FILE_TEMP) + rec->temp[inst->dest.index] &= ~inst->mask; + + if (read0) pass1_track_source(nvs, inst, 0, read0); + if (read1) pass1_track_source(nvs, inst, 1, read1); + if (read2) pass1_track_source(nvs, inst, 2, read2); + + return PASS1_OK; +} + +/* Some basic dead code elimination + * - Remove unused instructions + * - Don't write unused register components + * - Modify swizzles to not reference unneeded components. + */ +GLboolean +nouveau_shader_pass1(nvsPtr nvs) +{ + nvsFragmentList *list = nvs->list_tail; + int i; + + for (i=0; itemps[i].last_use = -1; + + nvs->pass_rec = calloc(1, sizeof(struct pass1_rec)); + + while (list) { + assert(list->fragment->type == NVS_INSTRUCTION); + + switch(pass1_check_instruction(nvs, (nvsInstruction *)list->fragment)) { + case PASS1_OK: + break; + case PASS1_KILL: + pass1_remove_fragment(nvs, list); + break; + case PASS1_FAIL: + default: + free(nvs->pass_rec); + nvs->pass_rec = NULL; + return GL_FALSE; + } + + list = list->prev; + } + + free(nvs->pass_rec); + nvs->pass_rec = NULL; + + return GL_TRUE; +} + + diff --git a/src/mesa/drivers/dri/nouveau/nouveau_shader_2.c b/src/mesa/drivers/dri/nouveau/nouveau_shader_2.c new file mode 100644 index 00000000000..1f09b6d4536 --- /dev/null +++ b/src/mesa/drivers/dri/nouveau/nouveau_shader_2.c @@ -0,0 +1,238 @@ +/* + * Copyright (C) 2006 Ben Skeggs. + * + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial + * portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + */ + +/* + * Authors: + * Ben Skeggs + */ + +#include "glheader.h" +#include "macros.h" +#include "enums.h" + +#include "nouveau_shader.h" + +struct pass2_rec { + /* Map nvsRegister temp ID onto hw temp ID */ + unsigned int temps[NVS_MAX_TEMPS]; + /* Track free hw registers */ + unsigned int hw_temps[NVS_MAX_TEMPS]; +}; + +static int +pass2_alloc_hw_temp(nvsPtr nvs) +{ + struct pass2_rec *rec = nvs->pass_rec; + int i; + + for (i=0; ifunc->MaxTemp; i++) { + /* This is a *horrible* hack.. R0 is both temp0 and result.color + * in NV30/40 fragprogs, we can use R0 as a temp before result is + * written however.. + */ + if (nvs->mesa.vp.Base.Target == GL_FRAGMENT_PROGRAM_ARB && i==0) + continue; + + if (rec->hw_temps[i] == 0) { + rec->hw_temps[i] = 1; + return i; + } + } + return -1; +} + +static void +pass2_free_hw_temp(nvsPtr nvs, int reg) +{ + struct pass2_rec *rec = nvs->pass_rec; + rec->hw_temps[reg] = 0; +} + +static nvsRegister +pass2_mangle_reg(nvsPtr nvs, nvsInstruction *inst, nvsRegister reg) +{ + struct pass2_rec *rec = nvs->pass_rec; + + if (reg.file == NVS_FILE_TEMP) { + int hwidx; + + if (rec->temps[reg.index] == -1) + rec->temps[reg.index] = pass2_alloc_hw_temp(nvs); + hwidx = rec->temps[reg.index]; + + if (nvs->temps[reg.index].last_use <= inst->header.position) + pass2_free_hw_temp(nvs, hwidx); + + reg.index = hwidx; + } + + return reg; +} + +static void +pass2_add_instruction(nvsPtr nvs, nvsInstruction *inst, + struct _op_xlat *op, int slot) +{ + nvsSwzComp default_swz[4] = { NVS_SWZ_X, NVS_SWZ_Y, NVS_SWZ_Z, NVS_SWZ_W }; + nvsFunc *shader = nvs->func; + nvsRegister reg; + int i, srcpos_used = ~7; + + shader->SetOpcode(shader, op->NV, slot); + if (inst->saturate ) shader->SetSaturate(shader); + if (inst->cond_update) shader->SetCCUpdate(shader); + if (inst->cond_test ) shader->SetCondition(shader, 1, inst->cond, + inst->cond_reg, + inst->cond_swizzle); + else shader->SetCondition(shader, 0, NVS_COND_TR, + 0, + default_swz); + switch (inst->op) { + case NVS_OP_TEX: + case NVS_OP_TXB: + case NVS_OP_TXL: + case NVS_OP_TXP: + case NVS_OP_TXD: + shader->SetTexImageUnit(shader, inst->tex_unit); + break; + default: + break; + } + + for (i = 0; i < 3; i++) { + if (op->srcpos[i] != -1) { + reg = pass2_mangle_reg(nvs, inst, inst->src[i]); + if (reg.file == NVS_FILE_ATTRIB) + nvs->inputs_read |= (1 << reg.index); + shader->SetSource(shader, ®, op->srcpos[i]); + srcpos_used |= (1<srcpos[i]); + if (reg.file == NVS_FILE_CONST && shader->GetSourceConstVal) + nvs->params[reg.index].hw_index = nvs->program_current + 4; + } + } + for (i = 0; i < 3; i++) { + if (!(srcpos_used & (1<SetUnusedSource(shader, i); + } + + reg = pass2_mangle_reg(nvs, inst, inst->dest); + if (reg.file == NVS_FILE_RESULT) + nvs->outputs_written |= (1 << reg.index); + shader->SetResult(shader, ®, inst->mask, slot); +} + +static int +pass2_assemble_instruction(nvsPtr nvs, nvsInstruction *inst, int last) +{ + nvsFunc *shader = nvs->func; + struct _op_xlat *op, *op2; + unsigned int hw_inst[8] = {0,0,0,0,0,0,0,0,0}; + int slot, slot2; + int instsz; + int i; + + shader->inst = hw_inst; + + /* Assemble this instruction */ + if (!(op = shader->GetOPTXFromSOP(inst->op, &slot))) + return 0; + pass2_add_instruction(nvs, inst, op, slot); + if (last) + shader->SetLastInst(shader); + + instsz = shader->GetOffsetNext(nvs->func); + if (nvs->program_size + instsz >= nvs->program_alloc_size) { + nvs->program_alloc_size *= 2; + nvs->program = realloc(nvs->program, + nvs->program_alloc_size * sizeof(uint32_t)); + } + + for (i=0; iprogram[nvs->program_current++] = hw_inst[i]; + nvs->program_size = nvs->program_current; + return 1; +} + +/* Translate program into hardware format */ +GLboolean +nouveau_shader_pass2(nvsPtr nvs) +{ + nvsFragmentList *list = nvs->list_head; + struct pass2_rec *rec; + int i; + + rec = calloc(1, sizeof(struct pass2_rec)); + for (i=0; itemps[i] = -1; + nvs->pass_rec = rec; + + /* Start off with allocating 4 uint32_t's for each inst, will be grown + * if necessary.. + */ + nvs->program_alloc_size = nvs->inst_count * 4; + nvs->program = calloc(nvs->program_alloc_size, sizeof(uint32_t)); + nvs->program_size = 0; + nvs->program_current = 0; + + while (list) { + assert(list->fragment->type == NVS_INSTRUCTION); + + if (!pass2_assemble_instruction(nvs, (nvsInstruction *)list->fragment, list->next ? 0 : 1)) { + free(nvs->program); + nvs->program = NULL; + return GL_FALSE; + } + + list = list->next; + } + + /* Shrink allocated memory to only what we need */ + nvs->program = realloc(nvs->program, nvs->program_size * sizeof(uint32_t)); + nvs->program_alloc_size = nvs->program_size; + + nvs->translated = 1; + nvs->on_hardware = 0; + +#if 1 + fflush(stdout); fflush(stderr); + fprintf(stderr, "----------------MESA PROGRAM\n"); + fflush(stdout); fflush(stderr); + _mesa_print_program(&nvs->mesa.vp.Base); + fflush(stdout); fflush(stderr); + fprintf(stderr, "^^^^^^^^^^^^^^^^MESA PROGRAM\n"); + fflush(stdout); fflush(stderr); + fprintf(stderr, "----------------NV40 PROGRAM\n"); + fflush(stdout); fflush(stderr); + nvsDisasmHWShader(nvs); + fflush(stdout); fflush(stderr); + fprintf(stderr, "^^^^^^^^^^^^^^^^NV40 PROGRAM\n"); + fflush(stdout); fflush(stderr); +#endif + + return GL_TRUE; +} + diff --git a/src/mesa/drivers/dri/nouveau/nouveau_state.h b/src/mesa/drivers/dri/nouveau/nouveau_state.h index f8fd0cea504..37f04f41bd4 100644 --- a/src/mesa/drivers/dri/nouveau/nouveau_state.h +++ b/src/mesa/drivers/dri/nouveau/nouveau_state.h @@ -37,6 +37,7 @@ extern void nv20InitStateFuncs(struct dd_function_table *func); extern void nv30InitStateFuncs(struct dd_function_table *func); extern void nouveauInitState(GLcontext *ctx); + /* extern void nouveauDDUpdateState(GLcontext *ctx); extern void nouveauDDUpdateHWState(GLcontext *ctx); diff --git a/src/mesa/drivers/dri/nouveau/nv20_shader.h b/src/mesa/drivers/dri/nouveau/nv20_shader.h new file mode 100644 index 00000000000..7d2e29db668 --- /dev/null +++ b/src/mesa/drivers/dri/nouveau/nv20_shader.h @@ -0,0 +1,121 @@ +/* NV20_TCL_PRIMITIVE_3D_0x0B00 */ +#define NV20_VP_INST_0B00 0x00000000 /* always 0? */ +#define NV20_VP_INST0_KNOWN 0 + +/* NV20_TCL_PRIMITIVE_3D_0x0B04 */ +#define NV20_VP_INST_SCA_OPCODE_SHIFT 25 +#define NV20_VP_INST_SCA_OPCODE_MASK (0x0F << 25) +#define NV20_VP_INST_OPCODE_RCP 0x2 +#define NV20_VP_INST_OPCODE_RCC 0x3 +#define NV20_VP_INST_OPCODE_RSQ 0x4 +#define NV20_VP_INST_OPCODE_EXP 0x5 +#define NV20_VP_INST_OPCODE_LOG 0x6 +#define NV20_VP_INST_OPCODE_LIT 0x7 +#define NV20_VP_INST_VEC_OPCODE_SHIFT 21 +#define NV20_VP_INST_VEC_OPCODE_MASK (0x0F << 21) +#define NV20_VP_INST_OPCODE_NOP 0x0 /* guess */ +#define NV20_VP_INST_OPCODE_MOV 0x1 +#define NV20_VP_INST_OPCODE_MUL 0x2 +#define NV20_VP_INST_OPCODE_ADD 0x3 +#define NV20_VP_INST_OPCODE_MAD 0x4 +#define NV20_VP_INST_OPCODE_DP3 0x5 +#define NV20_VP_INST_OPCODE_DPH 0x6 +#define NV20_VP_INST_OPCODE_DP4 0x7 +#define NV20_VP_INST_OPCODE_DST 0x8 +#define NV20_VP_INST_OPCODE_MIN 0x9 +#define NV20_VP_INST_OPCODE_MAX 0xA +#define NV20_VP_INST_OPCODE_SLT 0xB +#define NV20_VP_INST_OPCODE_SGE 0xC +#define NV20_VP_INST_OPCODE_ARL 0xD +#define NV20_VP_INST_CONST_SRC_SHIFT 13 +#define NV20_VP_INST_CONST_SRC_MASK (0xFF << 13) +#define NV20_VP_INST_INPUT_SRC_SHIFT 9 +#define NV20_VP_INST_INPUT_SRC_MASK (0xF << 9) /* guess */ +#define NV20_VP_INST_INPUT_SRC_POS 0 +#define NV20_VP_INST_INPUT_SRC_COL0 3 +#define NV20_VP_INST_INPUT_SRC_COL1 4 +#define NV20_VP_INST_INPUT_SRC_TC(n) (9+n) +#define NV20_VP_INST_SRC0H_SHIFT 0 +#define NV20_VP_INST_SRC0H_MASK (0x1FF << 0) +#define NV20_VP_INST1_KNOWN ( \ + NV20_VP_INST_OPCODE_MASK | \ + NV20_VP_INST_CONST_SRC_MASK | \ + NV20_VP_INST_INPUT_SRC_MASK | \ + NV20_VP_INST_SRC0H_MASK \ + ) + +/* NV20_TCL_PRIMITIVE_3D_0x0B08 */ +#define NV20_VP_INST_SRC0L_SHIFT 26 +#define NV20_VP_INST_SRC0L_MASK (0x3F <<26) +#define NV20_VP_INST_SRC1_SHIFT 11 +#define NV20_VP_INST_SRC1_MASK (0x7FFF<<11) +#define NV20_VP_INST_SRC2H_SHIFT 0 +#define NV20_VP_INST_SRC2H_MASK (0x7FF << 0) + +/* NV20_TCL_PRIMITIVE_3D_0x0B0C */ +#define NV20_VP_INST_SRC2L_SHIFT 28 +#define NV20_VP_INST_SRC2L_MASK (0x0F <<28) +#define NV20_VP_INST_VTEMP_WRITEMASK_SHIFT 24 +#define NV20_VP_INST_VTEMP_WRITEMASK_MASK (0x0F <<24) +# define NV20_VP_INST_TEMP_WRITEMASK_X (1<<27) +# define NV20_VP_INST_TEMP_WRITEMASK_Y (1<<26) +# define NV20_VP_INST_TEMP_WRITEMASK_Z (1<<25) +# define NV20_VP_INST_TEMP_WRITEMASK_W (1<<24) +#define NV20_VP_INST_DEST_TEMP_ID_SHIFT 20 +#define NV20_VP_INST_DEST_TEMP_ID_MASK (0x0F <<20) +#define NV20_VP_INST_STEMP_WRITEMASK_SHIFT 16 +#define NV20_VP_INST_STEMP_WRITEMASK_MASK (0x0F <<16) +# define NV20_VP_INST_STEMP_WRITEMASK_X (1<<19) +# define NV20_VP_INST_STEMP_WRITEMASK_Y (1<<18) +# define NV20_VP_INST_STEMP_WRITEMASK_Z (1<<17) +# define NV20_VP_INST_STEMP_WRITEMASK_W (1<<16) +#define NV20_VP_INST_DEST_WRITEMASK_SHIFT 12 +#define NV20_VP_INST_DEST_WRITEMASK_MASK (0x0F <<12) +# define NV20_VP_INST_DEST_WRITEMASK_X (1<<15) +# define NV20_VP_INST_DEST_WRITEMASK_Y (1<<14) +# define NV20_VP_INST_DEST_WRITEMASK_Z (1<<13) +# define NV20_VP_INST_DEST_WRITEMASK_W (1<<12) +#define NV20_VP_INST_DEST_SHIFT 3 +#define NV20_VP_INST_DEST_MASK (0xF << 3) /* guess */ +#define NV20_VP_INST_DEST_POS 0 +#define NV20_VP_INST_DEST_COL0 3 +#define NV20_VP_INST_DEST_COL1 4 +#define NV20_VP_INST_DEST_TC(n) (9+n) +#define NV20_VP_INST_INDEX_CONST (1<<1) +#define NV20_VP_INST3_KNOWN ( \ + NV20_VP_INST_SRC2L_MASK | \ + NV20_VP_INST_TEMP_WRITEMASK_MASK | \ + NV20_VP_INST_DEST_TEMP_ID_MASK | \ + NV20_VP_INST_STEMP_WRITEMASK_MASK | \ + NV20_VP_INST_DEST_WRITEMASK_MASK | \ + NV20_VP_INST_DEST_MASK | \ + NV20_VP_INST_INDEX_CONST \ + ) + +/* Useful to split the source selection regs into their pieces */ +#define NV20_VP_SRC0_HIGH_SHIFT 6 +#define NV20_VP_SRC0_HIGH_MASK 0x00007FC0 +#define NV20_VP_SRC0_LOW_MASK 0x0000003F +#define NV20_VP_SRC2_HIGH_SHIFT 4 +#define NV20_VP_SRC2_HIGH_MASK 0x00007FF0 +#define NV20_VP_SRC2_LOW_MASK 0x0000000F + +#define NV20_VP_SRC_REG_NEGATE (1<<14) +#define NV20_VP_SRC_REG_SWZ_X_SHIFT 12 +#define NV20_VP_SRC_REG_SWZ_X_MASK (0x03 <<12) +#define NV20_VP_SRC_REG_SWZ_Y_SHIFT 10 +#define NV20_VP_SRC_REG_SWZ_Y_MASK (0x03 <<10) +#define NV20_VP_SRC_REG_SWZ_Z_SHIFT 8 +#define NV20_VP_SRC_REG_SWZ_Z_MASK (0x03 << 8) +#define NV20_VP_SRC_REG_SWZ_W_SHIFT 6 +#define NV20_VP_SRC_REG_SWZ_W_MASK (0x03 << 6) +#define NV20_VP_SRC_REG_SWZ_ALL_SHIFT 6 +#define NV20_VP_SRC_REG_SWZ_ALL_MASK (0xFF << 6) +#define NV20_VP_SRC_REG_TEMP_ID_SHIFT 2 +#define NV20_VP_SRC_REG_TEMP_ID_MASK (0x0F << 0) +#define NV20_VP_SRC_REG_TYPE_SHIFT 0 +#define NV20_VP_SRC_REG_TYPE_MASK (0x03 << 0) +#define NV20_VP_SRC_REG_TYPE_TEMP 1 +#define NV20_VP_SRC_REG_TYPE_INPUT 2 +#define NV20_VP_SRC_REG_TYPE_CONST 3 /* guess */ + diff --git a/src/mesa/drivers/dri/nouveau/nv20_vertprog.c b/src/mesa/drivers/dri/nouveau/nv20_vertprog.c new file mode 100644 index 00000000000..60cfcd70561 --- /dev/null +++ b/src/mesa/drivers/dri/nouveau/nv20_vertprog.c @@ -0,0 +1,447 @@ +#include "nouveau_context.h" +#include "nouveau_object.h" +#include "nouveau_fifo.h" +#include "nouveau_reg.h" + +#include "nouveau_shader.h" +#include "nv20_shader.h" + +unsigned int NVVP_TX_VOP_COUNT = 16; +unsigned int NVVP_TX_NVS_OP_COUNT = 16; +struct _op_xlat NVVP_TX_VOP[32]; +struct _op_xlat NVVP_TX_SOP[32]; + +nvsSwzComp NV20VP_TX_SWIZZLE[4] = { NVS_SWZ_X, NVS_SWZ_Y, NVS_SWZ_Z, NVS_SWZ_W }; + +/***************************************************************************** + * Support routines + */ +static void +NV20VPUploadToHW(GLcontext *ctx, nouveauShader *nvs) +{ + nouveauContextPtr nmesa = NOUVEAU_CONTEXT(ctx); + int i; + + /* XXX: missing a way to say what insn we're uploading from, and possible + * the program start position (if NV20 has one) */ + for (i=0; iprogram_size; i+=4) { + BEGIN_RING_SIZE(NvSub3D, NV20_TCL_PRIMITIVE_3D_VP_UPLOAD_INST0, 4); + OUT_RING(nvs->program[i + 0]); + OUT_RING(nvs->program[i + 1]); + OUT_RING(nvs->program[i + 2]); + OUT_RING(nvs->program[i + 3]); + } +} + +static void +NV20VPUpdateConst(GLcontext *ctx, nouveauShader *nvs, int id) +{ + nouveauContextPtr nmesa = NOUVEAU_CONTEXT(ctx); + + /* Worth checking if the value *actually* changed? Mesa doesn't tell us this + * as far as I know.. + */ + BEGIN_RING_SIZE(NvSub3D, NV30_TCL_PRIMITIVE_3D_VP_UPLOAD_CONST_ID, 1); + OUT_RING (id); + BEGIN_RING_SIZE(NvSub3D, NV20_TCL_PRIMITIVE_3D_VP_UPLOAD_CONST_X, 4); + OUT_RINGf(nvs->params[id].source_val[0]); + OUT_RINGf(nvs->params[id].source_val[1]); + OUT_RINGf(nvs->params[id].source_val[2]); + OUT_RINGf(nvs->params[id].source_val[3]); +} + +/***************************************************************************** + * Assembly routines + */ + +/***************************************************************************** + * Disassembly routines + */ +void +NV20VPTXSwizzle(int hwswz, nvsSwzComp *swz) +{ + swz[NVS_SWZ_X] = NV20VP_TX_SWIZZLE[(hwswz & 0xC0) >> 6]; + swz[NVS_SWZ_Y] = NV20VP_TX_SWIZZLE[(hwswz & 0x30) >> 4]; + swz[NVS_SWZ_Z] = NV20VP_TX_SWIZZLE[(hwswz & 0x0C) >> 2]; + swz[NVS_SWZ_W] = NV20VP_TX_SWIZZLE[(hwswz & 0x03) >> 0]; +} + +static int +NV20VPHasMergedInst(nvsFunc * shader) +{ + if (shader->GetOpcodeHW(shader, 0) != NV20_VP_INST_OPCODE_NOP && + shader->GetOpcodeHW(shader, 1) != NV20_VP_INST_OPCODE_NOP) + printf + ("\n\n*****both opcode fields have values - PLEASE REPORT*****\n"); + return 0; +} + +static int +NV20VPIsLastInst(nvsFunc * shader) +{ + return ((shader->inst[3] & (1 << 0)) ? 1 : 0); +} + +static int +NV20VPGetOffsetNext(nvsFunc * shader) +{ + return 4; +} + +static struct _op_xlat * +NV20VPGetOPTXRec(nvsFunc * shader, int merged) +{ + struct _op_xlat *opr; + int op; + + if (shader->GetOpcodeSlot(shader, merged)) { + opr = NVVP_TX_SOP; + op = shader->GetOpcodeHW(shader, 1); + if (op >= NVVP_TX_NVS_OP_COUNT) + return NULL; + } + else { + opr = NVVP_TX_VOP; + op = shader->GetOpcodeHW(shader, 0); + if (op >= NVVP_TX_VOP_COUNT) + return NULL; + } + + if (opr[op].SOP == NVS_OP_UNKNOWN) + return NULL; + return &opr[op]; +} + +static struct _op_xlat * +NV20VPGetOPTXFromSOP(nvsOpcode sop, int *id) +{ + int i; + + for (i=0;iHasMergedInst(shader)) + return merged; + if (shader->GetOpcodeHW(shader, 0) == NV20_VP_INST_OPCODE_NOP) + return 1; + return 0; +} + +static nvsOpcode +NV20VPGetOpcode(nvsFunc * shader, int merged) +{ + struct _op_xlat *opr; + + opr = shader->GetOPTXRec(shader, merged); + if (!opr) + return NVS_OP_UNKNOWN; + + return opr->SOP; +} + +static nvsOpcode +NV20VPGetOpcodeHW(nvsFunc * shader, int slot) +{ + if (slot) + return (shader->inst[1] & NV20_VP_INST_SCA_OPCODE_MASK) + >> NV20_VP_INST_SCA_OPCODE_SHIFT; + return (shader->inst[1] & NV20_VP_INST_VEC_OPCODE_MASK) + >> NV20_VP_INST_VEC_OPCODE_SHIFT; +} + +static nvsRegFile +NV20VPGetDestFile(nvsFunc * shader, int merged) +{ + switch (shader->GetOpcode(shader, merged)) { + case NVS_OP_ARL: + return NVS_FILE_ADDRESS; + default: + /*FIXME: This probably isn't correct.. */ + if ((shader->inst[3] & NV20_VP_INST_DEST_WRITEMASK_MASK) == 0) + return NVS_FILE_TEMP; + return NVS_FILE_RESULT; + } +} + +static unsigned int +NV20VPGetDestID(nvsFunc * shader, int merged) +{ + int id; + + switch (shader->GetDestFile(shader, merged)) { + case NVS_FILE_RESULT: + id = ((shader->inst[3] & NV20_VP_INST_DEST_MASK) + >> NV20_VP_INST_DEST_SHIFT); + switch (id) { + case NV20_VP_INST_DEST_POS : return NVS_FR_POSITION; + case NV20_VP_INST_DEST_COL0 : return NVS_FR_COL0; + case NV20_VP_INST_DEST_COL1 : return NVS_FR_COL1; + case NV20_VP_INST_DEST_TC(0): return NVS_FR_TEXCOORD0; + case NV20_VP_INST_DEST_TC(1): return NVS_FR_TEXCOORD1; + case NV20_VP_INST_DEST_TC(2): return NVS_FR_TEXCOORD2; + case NV20_VP_INST_DEST_TC(3): return NVS_FR_TEXCOORD3; + default: + return -1; + } + case NVS_FILE_ADDRESS: + return 0; + case NVS_FILE_TEMP: + id = ((shader->inst[3] & NV20_VP_INST_DEST_TEMP_ID_MASK) + >> NV20_VP_INST_DEST_TEMP_ID_SHIFT); + return id; + default: + return -1; + } +} + +static unsigned int +NV20VPGetDestMask(nvsFunc * shader, int merged) +{ + int hwmask, mask = 0; + + /* Special handling for ARL - hardware only supports a + * 1-component address reg + */ + if (shader->GetOpcode(shader, merged) == NVS_OP_ARL) + return SMASK_X; + + if (shader->GetDestFile(shader, merged) == NVS_FILE_RESULT) + hwmask = (shader->inst[3] & NV20_VP_INST_DEST_WRITEMASK_MASK) + >> NV20_VP_INST_DEST_WRITEMASK_SHIFT; + else if (shader->GetOpcodeSlot(shader, merged)) + hwmask = (shader->inst[3] & NV20_VP_INST_STEMP_WRITEMASK_MASK) + >> NV20_VP_INST_STEMP_WRITEMASK_SHIFT; + else + hwmask = (shader->inst[3] & NV20_VP_INST_VTEMP_WRITEMASK_MASK) + >> NV20_VP_INST_VTEMP_WRITEMASK_SHIFT; + + if (hwmask & (1 << 3)) mask |= SMASK_X; + if (hwmask & (1 << 2)) mask |= SMASK_Y; + if (hwmask & (1 << 1)) mask |= SMASK_Z; + if (hwmask & (1 << 0)) mask |= SMASK_W; + + return mask; +} + +static unsigned int +NV20VPGetSourceHW(nvsFunc * shader, int merged, int pos) +{ + struct _op_xlat *opr; + unsigned int src; + + opr = shader->GetOPTXRec(shader, merged); + if (!opr) + return -1; + + switch (opr->srcpos[pos]) { + case 0: + src = ((shader->inst[1] & NV20_VP_INST_SRC0H_MASK) + >> NV20_VP_INST_SRC0H_SHIFT) + << NV20_VP_SRC0_HIGH_SHIFT; + src |= ((shader->inst[2] & NV20_VP_INST_SRC0L_MASK) + >> NV20_VP_INST_SRC0L_SHIFT); + break; + case 1: + src = ((shader->inst[2] & NV20_VP_INST_SRC1_MASK) + >> NV20_VP_INST_SRC1_SHIFT); + break; + case 2: + src = ((shader->inst[2] & NV20_VP_INST_SRC2H_MASK) + >> NV20_VP_INST_SRC2H_SHIFT) + << NV20_VP_SRC2_HIGH_SHIFT; + src |= ((shader->inst[3] & NV20_VP_INST_SRC2L_MASK) + >> NV20_VP_INST_SRC2L_SHIFT); + break; + default: + src = -1; + } + + return src; +} + +static nvsRegFile +NV20VPGetSourceFile(nvsFunc * shader, int merged, int pos) +{ + unsigned int src; + struct _op_xlat *opr; + int file; + + opr = shader->GetOPTXRec(shader, merged); + if (!opr || opr->srcpos[pos] == -1) + return -1; + + switch (opr->srcpos[pos]) { + case SPOS_ADDRESS: + return NVS_FILE_ADDRESS; + default: + src = NV20VPGetSourceHW(shader, merged, pos); + file = (src & NV20_VP_SRC_REG_TYPE_MASK) >> NV20_VP_SRC_REG_TYPE_SHIFT; + + switch (file) { + case NV20_VP_SRC_REG_TYPE_TEMP : return NVS_FILE_TEMP; + case NV20_VP_SRC_REG_TYPE_INPUT: return NVS_FILE_ATTRIB; + case NV20_VP_SRC_REG_TYPE_CONST: return NVS_FILE_CONST; + default: + return NVS_FILE_UNKNOWN; + } + } +} + +static int +NV20VPGetSourceID(nvsFunc * shader, int merged, int pos) +{ + unsigned int src; + + switch (shader->GetSourceFile(shader, merged, pos)) { + case NVS_FILE_TEMP: + src = shader->GetSourceHW(shader, merged, pos); + return ((src & NV20_VP_SRC_REG_TEMP_ID_MASK) >> + NV20_VP_SRC_REG_TEMP_ID_SHIFT); + case NVS_FILE_CONST: + return ((shader->inst[1] & NV20_VP_INST_CONST_SRC_MASK) + >> NV20_VP_INST_CONST_SRC_SHIFT); + case NVS_FILE_ATTRIB: + src = ((shader->inst[1] & NV20_VP_INST_INPUT_SRC_MASK) + >> NV20_VP_INST_INPUT_SRC_SHIFT); + switch (src) { + case NV20_VP_INST_INPUT_SRC_POS : return NVS_FR_POSITION; + case NV20_VP_INST_INPUT_SRC_COL0 : return NVS_FR_COL0; + case NV20_VP_INST_INPUT_SRC_COL1 : return NVS_FR_COL1; + case NV20_VP_INST_INPUT_SRC_TC(0): return NVS_FR_TEXCOORD0; + case NV20_VP_INST_INPUT_SRC_TC(1): return NVS_FR_TEXCOORD1; + case NV20_VP_INST_INPUT_SRC_TC(2): return NVS_FR_TEXCOORD2; + case NV20_VP_INST_INPUT_SRC_TC(3): return NVS_FR_TEXCOORD3; + default: + return NVS_FR_UNKNOWN; + } + default: + return -1; + } +} + +static int +NV20VPGetSourceNegate(nvsFunc * shader, int merged, int pos) +{ + unsigned int src; + + src = shader->GetSourceHW(shader, merged, pos); + + return ((src & NV20_VP_SRC_REG_NEGATE) ? 1 : 0); +} + +static int +NV20VPGetSourceAbs(nvsFunc * shader, int merged, int pos) +{ + /* NV20 can't do ABS on sources? Appears to be emulated with + * MAX reg, reg, -reg + */ + return 0; +} + +static void +NV20VPGetSourceSwizzle(nvsFunc * shader, int merged, int pos, nvsSwzComp *swz) +{ + unsigned int src; + int swzbits; + + src = shader->GetSourceHW(shader, merged, pos); + swzbits = + (src & NV20_VP_SRC_REG_SWZ_ALL_MASK) >> NV20_VP_SRC_REG_SWZ_ALL_SHIFT; + return NV20VPTXSwizzle(swzbits, swz); +} + +static int +NV20VPGetSourceIndexed(nvsFunc * shader, int merged, int pos) +{ + /* I don't think NV20 can index into attribs, at least no GL + * extension is exposed that will allow it. + */ + if (shader->GetSourceFile(shader, merged, pos) != NVS_FILE_CONST) + return 0; + if (shader->inst[3] & NV20_VP_INST_INDEX_CONST) + return 1; + return 0; +} + +static int +NV20VPGetAddressRegID(nvsFunc * shader) +{ + /* Only 1 address reg */ + return 0; +} + +static nvsSwzComp +NV20VPGetAddressRegSwizzle(nvsFunc * shader) +{ + /* Only A0.x available */ + return NVS_SWZ_X; +} + +void +NV20VPInitShaderFuncs(nvsFunc * shader) +{ + MOD_OPCODE(NVVP_TX_VOP, NV20_VP_INST_OPCODE_NOP, NVS_OP_NOP, -1, -1, -1); + MOD_OPCODE(NVVP_TX_VOP, NV20_VP_INST_OPCODE_MOV, NVS_OP_MOV, 0, -1, -1); + MOD_OPCODE(NVVP_TX_VOP, NV20_VP_INST_OPCODE_MUL, NVS_OP_MUL, 0, 1, -1); + MOD_OPCODE(NVVP_TX_VOP, NV20_VP_INST_OPCODE_ADD, NVS_OP_ADD, 0, 2, -1); + MOD_OPCODE(NVVP_TX_VOP, NV20_VP_INST_OPCODE_MAD, NVS_OP_MAD, 0, 1, 2); + MOD_OPCODE(NVVP_TX_VOP, NV20_VP_INST_OPCODE_DP3, NVS_OP_DP3, 0, 1, -1); + MOD_OPCODE(NVVP_TX_VOP, NV20_VP_INST_OPCODE_DPH, NVS_OP_DPH, 0, 1, -1); + MOD_OPCODE(NVVP_TX_VOP, NV20_VP_INST_OPCODE_DP4, NVS_OP_DP4, 0, 1, -1); + MOD_OPCODE(NVVP_TX_VOP, NV20_VP_INST_OPCODE_DST, NVS_OP_DST, 0, 1, -1); + MOD_OPCODE(NVVP_TX_VOP, NV20_VP_INST_OPCODE_MIN, NVS_OP_MIN, 0, 1, -1); + MOD_OPCODE(NVVP_TX_VOP, NV20_VP_INST_OPCODE_MAX, NVS_OP_MAX, 0, 1, -1); + MOD_OPCODE(NVVP_TX_VOP, NV20_VP_INST_OPCODE_SLT, NVS_OP_SLT, 0, 1, -1); + MOD_OPCODE(NVVP_TX_VOP, NV20_VP_INST_OPCODE_SGE, NVS_OP_SGE, 0, 1, -1); + MOD_OPCODE(NVVP_TX_VOP, NV20_VP_INST_OPCODE_ARL, NVS_OP_ARL, 0, -1, -1); + + MOD_OPCODE(NVVP_TX_SOP, NV20_VP_INST_OPCODE_NOP, NVS_OP_NOP, -1, -1, -1); + MOD_OPCODE(NVVP_TX_SOP, NV20_VP_INST_OPCODE_RCP, NVS_OP_RCP, 2, -1, -1); + MOD_OPCODE(NVVP_TX_SOP, NV20_VP_INST_OPCODE_RCC, NVS_OP_RCC, 2, -1, -1); + MOD_OPCODE(NVVP_TX_SOP, NV20_VP_INST_OPCODE_RSQ, NVS_OP_RSQ, 2, -1, -1); + MOD_OPCODE(NVVP_TX_SOP, NV20_VP_INST_OPCODE_EXP, NVS_OP_EXP, 2, -1, -1); + MOD_OPCODE(NVVP_TX_SOP, NV20_VP_INST_OPCODE_LOG, NVS_OP_LOG, 2, -1, -1); + MOD_OPCODE(NVVP_TX_SOP, NV20_VP_INST_OPCODE_LIT, NVS_OP_LIT, 2, -1, -1); + + shader->UploadToHW = NV20VPUploadToHW; + shader->UpdateConst = NV20VPUpdateConst; + + shader->GetOPTXRec = NV20VPGetOPTXRec; + shader->GetOPTXFromSOP = NV20VPGetOPTXFromSOP; + + shader->HasMergedInst = NV20VPHasMergedInst; + shader->IsLastInst = NV20VPIsLastInst; + shader->GetOffsetNext = NV20VPGetOffsetNext; + shader->GetOpcodeSlot = NV20VPGetOpcodeSlot; + shader->GetOpcode = NV20VPGetOpcode; + shader->GetOpcodeHW = NV20VPGetOpcodeHW; + shader->GetDestFile = NV20VPGetDestFile; + shader->GetDestID = NV20VPGetDestID; + shader->GetDestMask = NV20VPGetDestMask; + shader->GetSourceHW = NV20VPGetSourceHW; + shader->GetSourceFile = NV20VPGetSourceFile; + shader->GetSourceID = NV20VPGetSourceID; + shader->GetSourceNegate = NV20VPGetSourceNegate; + shader->GetSourceAbs = NV20VPGetSourceAbs; + shader->GetSourceSwizzle = NV20VPGetSourceSwizzle; + shader->GetSourceIndexed = NV20VPGetSourceIndexed; + shader->GetRelAddressRegID = NV20VPGetAddressRegID; + shader->GetRelAddressSwizzle = NV20VPGetAddressRegSwizzle; +} diff --git a/src/mesa/drivers/dri/nouveau/nv30_fragprog.c b/src/mesa/drivers/dri/nouveau/nv30_fragprog.c new file mode 100644 index 00000000000..2e35d08c07c --- /dev/null +++ b/src/mesa/drivers/dri/nouveau/nv30_fragprog.c @@ -0,0 +1,707 @@ +#include + +#include "glheader.h" +#include "macros.h" + +#include "nouveau_context.h" +#include "nouveau_fifo.h" +#include "nouveau_reg.h" +#include "nouveau_drm.h" +#include "nouveau_shader.h" +#include "nouveau_object.h" +#include "nouveau_msg.h" +#include "nv30_shader.h" + +unsigned int NVFP_TX_AOP_COUNT = 64; +struct _op_xlat NVFP_TX_AOP[64]; + +/******************************************************************************* + * Support routines + */ + +/*XXX: bad bad bad bad */ +static uint64_t fragprog_ofs; +static uint32_t *fragprog_buf = NULL; + +static void +NV30FPUploadToHW(GLcontext *ctx, nouveauShader *nvs) +{ + nouveauContextPtr nmesa = NOUVEAU_CONTEXT(ctx); + drm_nouveau_mem_alloc_t mem; + + if (!fragprog_buf) { + mem.flags = NOUVEAU_MEM_FB|NOUVEAU_MEM_MAPPED; + mem.size = nvs->program_size * sizeof(uint32_t); + mem.alignment = 0; + mem.region_offset = &fragprog_ofs; + if (drmCommandWriteRead(nmesa->driFd, DRM_NOUVEAU_MEM_ALLOC, &mem, + sizeof(mem))) { + fprintf(stderr, "MEM_ALLOC fail\n"); + return; + } + + if (drmMap(nmesa->driFd, fragprog_ofs, mem.size, &fragprog_buf)) { + fprintf(stderr, "MEM_MAP fail\n"); + return; + } + } + + /*XXX: should do a DMA.. and not copy over a possibly in-use program.. */ + /* not using state cache here, updated programs at the same address + * seem to not take effect unless ACTIVE_PROGRAM is called again. hw + * caches the program somewhere? so, maybe not so bad to just clobber the + * old program in vram.. + */ + memcpy(fragprog_buf, nvs->program, nvs->program_size * sizeof(uint32_t)); + BEGIN_RING_SIZE(NvSub3D, NV30_TCL_PRIMITIVE_3D_FP_ACTIVE_PROGRAM, 1); + OUT_RING(((uint32_t)fragprog_ofs-0xE0000000)|1); +} + +static void +NV30FPUpdateConst(GLcontext *ctx, nouveauShader *nvs, int id) +{ + uint32_t *current = nvs->program + nvs->params[id].hw_index; + uint32_t *new = nvs->params[id].source_val ? + nvs->params[id].source_val : nvs->params[id].val; + + COPY_4V(current, new); + nvs->on_hardware = 0; +} + +/******************************************************************************* + * Assembly helpers + */ +static struct _op_xlat * +NV30FPGetOPTXFromSOP(nvsOpcode op, int *id) +{ + int i; + + for (i=0; iGetOPTXFromSOP(op, NULL)) + return 1; + return 0; +} + +static void +NV30FPSetOpcode(nvsFunc *shader, unsigned int opcode, int slot) +{ + shader->inst[0] |= (opcode << NV30_FP_OP_OPCODE_SHIFT); +} + +static void +NV30FPSetCCUpdate(nvsFunc *shader) +{ + shader->inst[0] |= NV30_FP_OP_COND_WRITE_ENABLE; +} + +static void +NV30FPSetCondition(nvsFunc *shader, int on, nvsCond cond, int reg, + nvsSwzComp *swz) +{ + nvsSwzComp default_swz[4] = { NVS_SWZ_X, NVS_SWZ_Y, NVS_SWZ_Z, NVS_SWZ_W }; + unsigned int hwcond; + + /* cond masking is always enabled */ + if (!on) { + cond = NVS_COND_TR; + reg = 0; + swz = default_swz; + } + + switch (cond) { + case NVS_COND_TR: hwcond = NV30_FP_OP_COND_TR; break; + case NVS_COND_FL: hwcond = NV30_FP_OP_COND_FL; break; + case NVS_COND_LT: hwcond = NV30_FP_OP_COND_LT; break; + case NVS_COND_GT: hwcond = NV30_FP_OP_COND_GT; break; + case NVS_COND_LE: hwcond = NV30_FP_OP_COND_LE; break; + case NVS_COND_GE: hwcond = NV30_FP_OP_COND_GE; break; + case NVS_COND_EQ: hwcond = NV30_FP_OP_COND_EQ; break; + case NVS_COND_NE: hwcond = NV30_FP_OP_COND_NE; break; + default: + WARN_ONCE("unknown fp condmask=%d\n", cond); + hwcond = NV30_FP_OP_COND_TR; + break; + } + + shader->inst[1] |= (hwcond << NV30_FP_OP_COND_SHIFT); + shader->inst[1] |= (swz[NVS_SWZ_X] << NV30_FP_OP_COND_SWZ_X_SHIFT); + shader->inst[1] |= (swz[NVS_SWZ_Y] << NV30_FP_OP_COND_SWZ_Y_SHIFT); + shader->inst[1] |= (swz[NVS_SWZ_Z] << NV30_FP_OP_COND_SWZ_Z_SHIFT); + shader->inst[1] |= (swz[NVS_SWZ_W] << NV30_FP_OP_COND_SWZ_W_SHIFT); +} + +static void +NV30FPSetResult(nvsFunc *shader, nvsRegister *reg, unsigned int mask, int slot) +{ + unsigned int hwreg, hwmask = 0; + + if (mask & SMASK_X) shader->inst[0] |= NV30_FP_OP_OUT_X; + if (mask & SMASK_Y) shader->inst[0] |= NV30_FP_OP_OUT_Y; + if (mask & SMASK_Z) shader->inst[0] |= NV30_FP_OP_OUT_Z; + if (mask & SMASK_W) shader->inst[0] |= NV30_FP_OP_OUT_W; + + if (reg->file == NVS_FILE_RESULT) { + hwreg = 0; /* FIXME: this is only fragment.color */ + /* This is *not* correct, I have no idea what it is either */ + shader->inst[0] |= NV30_FP_OP_UNK0_7; + } else + hwreg = reg->index; + shader->inst[0] |= (hwreg << NV30_FP_OP_OUT_REG_SHIFT); +} + +static void +NV30FPSetSource(nvsFunc *shader, nvsRegister *reg, int pos) +{ + unsigned int hwsrc = 0; + + switch (reg->file) { + case NVS_FILE_TEMP: + hwsrc |= (NV30_FP_REG_TYPE_TEMP << NV30_FP_REG_TYPE_SHIFT); + hwsrc |= (reg->index << NV30_FP_REG_SRC_SHIFT); + break; + case NVS_FILE_ATTRIB: + { + unsigned int hwin; + + switch (reg->index) { + case NVS_FR_POSITION : hwin = NV30_FP_OP_INPUT_SRC_POSITION; break; + case NVS_FR_COL0 : hwin = NV30_FP_OP_INPUT_SRC_COL0; break; + case NVS_FR_COL1 : hwin = NV30_FP_OP_INPUT_SRC_COL1; break; + case NVS_FR_FOGCOORD : hwin = NV30_FP_OP_INPUT_SRC_FOGC; break; + case NVS_FR_TEXCOORD0: hwin = NV30_FP_OP_INPUT_SRC_TC(0); break; + case NVS_FR_TEXCOORD1: hwin = NV30_FP_OP_INPUT_SRC_TC(1); break; + case NVS_FR_TEXCOORD2: hwin = NV30_FP_OP_INPUT_SRC_TC(2); break; + case NVS_FR_TEXCOORD3: hwin = NV30_FP_OP_INPUT_SRC_TC(3); break; + case NVS_FR_TEXCOORD4: hwin = NV30_FP_OP_INPUT_SRC_TC(4); break; + case NVS_FR_TEXCOORD5: hwin = NV30_FP_OP_INPUT_SRC_TC(5); break; + case NVS_FR_TEXCOORD6: hwin = NV30_FP_OP_INPUT_SRC_TC(6); break; + case NVS_FR_TEXCOORD7: hwin = NV30_FP_OP_INPUT_SRC_TC(7); break; + default: + WARN_ONCE("unknown fp input %d\n", reg->index); + hwin = NV30_FP_OP_INPUT_SRC_COL0; + break; + } + shader->inst[0] |= (hwin << NV30_FP_OP_INPUT_SRC_SHIFT); + hwsrc |= (hwin << NV30_FP_REG_SRC_SHIFT); + } + hwsrc |= (NV30_FP_REG_TYPE_INPUT << NV30_FP_REG_TYPE_SHIFT); + break; + case NVS_FILE_CONST: + /* consts are inlined after the inst */ + hwsrc |= (NV30_FP_REG_TYPE_CONST << NV30_FP_REG_TYPE_SHIFT); + break; + default: + assert(0); + break; + } + + if (reg->negate) + hwsrc |= NV30_FP_REG_NEGATE; + if (reg->abs) + shader->inst[1] |= (1 << (29+pos)); + hwsrc |= (reg->swizzle[NVS_SWZ_X] << NV30_FP_REG_SWZ_X_SHIFT); + hwsrc |= (reg->swizzle[NVS_SWZ_Y] << NV30_FP_REG_SWZ_Y_SHIFT); + hwsrc |= (reg->swizzle[NVS_SWZ_Z] << NV30_FP_REG_SWZ_Z_SHIFT); + hwsrc |= (reg->swizzle[NVS_SWZ_W] << NV30_FP_REG_SWZ_W_SHIFT); + + shader->inst[pos+1] |= hwsrc; +} + +static void +NV30FPSetUnusedSource(nvsFunc *shader, int pos) +{ + shader->inst[pos+1] |= ( + (NV30_FP_REG_TYPE_INPUT << NV30_FP_REG_TYPE_SHIFT) | + (NVS_SWZ_X << NV30_FP_REG_SWZ_X_SHIFT) | + (NVS_SWZ_Y << NV30_FP_REG_SWZ_Y_SHIFT) | + (NVS_SWZ_Z << NV30_FP_REG_SWZ_Z_SHIFT) | + (NVS_SWZ_W << NV30_FP_REG_SWZ_W_SHIFT) + ); +} + +static void +NV30FPSetTexImageUnit(nvsFunc *shader, int unit) +{ + shader->inst[0] |= (unit << NV30_FP_OP_TEX_UNIT_SHIFT); +} + +static void +NV30FPSetSaturate(nvsFunc *shader) +{ + shader->inst[0] |= NV30_FP_OP_OUT_SAT; +} + +static void +NV30FPSetLastInst(nvsFunc *shader) +{ + shader->inst[0] |= 1; + +} + +/******************************************************************************* + * Disassembly helpers + */ +static struct _op_xlat * +NV30FPGetOPTXRec(nvsFunc * shader, int merged) +{ + int op; + + op = shader->GetOpcodeHW(shader, 0); + if (op > NVFP_TX_AOP_COUNT) + return NULL; + if (NVFP_TX_AOP[op].SOP == NVS_OP_UNKNOWN) + return NULL; + return &NVFP_TX_AOP[op]; +} + +static int +NV30FPHasMergedInst(nvsFunc * shader) +{ + return 0; +} + +static int +NV30FPIsLastInst(nvsFunc * shader) +{ + return ((shader->inst[0] & NV30_FP_OP_PROGRAM_END) ? 1 : 0); +} + +static int +NV30FPGetOffsetNext(nvsFunc * shader) +{ + int i; + + for (i = 0; i < 3; i++) + if (shader->GetSourceFile(shader, 0, i) == NVS_FILE_CONST) + return 8; + return 4; +} + +static nvsOpcode +NV30FPGetOpcode(nvsFunc * shader, int merged) +{ + struct _op_xlat *opr; + + opr = shader->GetOPTXRec(shader, merged); + if (!opr) + return NVS_OP_UNKNOWN; + + return opr->SOP; +} + +static unsigned int +NV30FPGetOpcodeHW(nvsFunc * shader, int slot) +{ + int op; + + op = (shader->inst[0] & NV30_FP_OP_OPCODE_MASK) >> NV30_FP_OP_OPCODE_SHIFT; + + return op; +} + +static nvsRegFile +NV30FPGetDestFile(nvsFunc * shader, int merged) +{ + /* Result regs overlap temporary regs */ + return NVS_FILE_TEMP; +} + +static unsigned int +NV30FPGetDestID(nvsFunc * shader, int merged) +{ + int id; + + switch (shader->GetDestFile(shader, merged)) { + case NVS_FILE_TEMP: + id = ((shader->inst[0] & NV30_FP_OP_OUT_REG_MASK) + >> NV30_FP_OP_OUT_REG_SHIFT); + return id; + default: + return -1; + } +} + +static unsigned int +NV30FPGetDestMask(nvsFunc * shader, int merged) +{ + unsigned int mask = 0; + + if (shader->inst[0] & NV30_FP_OP_OUT_X) mask |= SMASK_X; + if (shader->inst[0] & NV30_FP_OP_OUT_Y) mask |= SMASK_Y; + if (shader->inst[0] & NV30_FP_OP_OUT_Z) mask |= SMASK_Z; + if (shader->inst[0] & NV30_FP_OP_OUT_W) mask |= SMASK_W; + + return mask; +} + +static unsigned int +NV30FPGetSourceHW(nvsFunc * shader, int merged, int pos) +{ + struct _op_xlat *opr; + + opr = shader->GetOPTXRec(shader, merged); + if (!opr || opr->srcpos[pos] == -1) + return -1; + + return shader->inst[opr->srcpos[pos] + 1]; +} + +static nvsRegFile +NV30FPGetSourceFile(nvsFunc * shader, int merged, int pos) +{ + unsigned int src; + struct _op_xlat *opr; + int file; + + opr = shader->GetOPTXRec(shader, merged); + if (!opr || opr->srcpos[pos] == -1) + return NVS_FILE_UNKNOWN; + + switch (opr->srcpos[pos]) { + case SPOS_ADDRESS: return NVS_FILE_ADDRESS; + default: + src = shader->GetSourceHW(shader, merged, pos); + file = (src & NV30_FP_REG_TYPE_MASK) >> NV30_FP_REG_TYPE_SHIFT; + + switch (file) { + case NV30_FP_REG_TYPE_TEMP : return NVS_FILE_TEMP; + case NV30_FP_REG_TYPE_INPUT: return NVS_FILE_ATTRIB; + case NV30_FP_REG_TYPE_CONST: return NVS_FILE_CONST; + default: + return NVS_FILE_UNKNOWN; + } + } +} + +static int +NV30FPGetSourceID(nvsFunc * shader, int merged, int pos) +{ + switch (shader->GetSourceFile(shader, merged, pos)) { + case NVS_FILE_ATTRIB: + switch ((shader->inst[0] & NV30_FP_OP_INPUT_SRC_MASK) + >> NV30_FP_OP_INPUT_SRC_SHIFT) { + case NV30_FP_OP_INPUT_SRC_POSITION: return NVS_FR_POSITION; + case NV30_FP_OP_INPUT_SRC_COL0 : return NVS_FR_COL0; + case NV30_FP_OP_INPUT_SRC_COL1 : return NVS_FR_COL1; + case NV30_FP_OP_INPUT_SRC_FOGC : return NVS_FR_FOGCOORD; + case NV30_FP_OP_INPUT_SRC_TC(0) : return NVS_FR_TEXCOORD0; + case NV30_FP_OP_INPUT_SRC_TC(1) : return NVS_FR_TEXCOORD1; + case NV30_FP_OP_INPUT_SRC_TC(2) : return NVS_FR_TEXCOORD2; + case NV30_FP_OP_INPUT_SRC_TC(3) : return NVS_FR_TEXCOORD3; + case NV30_FP_OP_INPUT_SRC_TC(4) : return NVS_FR_TEXCOORD4; + case NV30_FP_OP_INPUT_SRC_TC(5) : return NVS_FR_TEXCOORD5; + case NV30_FP_OP_INPUT_SRC_TC(6) : return NVS_FR_TEXCOORD6; + case NV30_FP_OP_INPUT_SRC_TC(7) : return NVS_FR_TEXCOORD7; + default: + return -1; + } + break; + case NVS_FILE_TEMP: + { + unsigned int src; + + src = shader->GetSourceHW(shader, merged, pos); + return ((src & NV30_FP_REG_SRC_MASK) >> NV30_FP_REG_SRC_SHIFT); + } + case NVS_FILE_CONST: /* inlined into fragprog */ + default: + return -1; + } +} + +static int +NV30FPGetTexImageUnit(nvsFunc *shader) +{ + return ((shader->inst[0] & NV30_FP_OP_TEX_UNIT_MASK) + >> NV30_FP_OP_TEX_UNIT_SHIFT); +} + +static int +NV30FPGetSourceNegate(nvsFunc * shader, int merged, int pos) +{ + unsigned int src; + + src = shader->GetSourceHW(shader, merged, pos); + + if (src == -1) + return -1; + return ((src & NV30_FP_REG_NEGATE) ? 1 : 0); +} + +static int +NV30FPGetSourceAbs(nvsFunc * shader, int merged, int pos) +{ + struct _op_xlat *opr; + static unsigned int abspos[3] = { + NV30_FP_OP_OUT_ABS, + (1 << 30), /* guess */ + (1 << 31) /* guess */ + }; + + opr = shader->GetOPTXRec(shader, merged); + if (!opr || opr->srcpos[pos] == -1) + return -1; + + return ((shader->inst[1] & abspos[opr->srcpos[pos]]) ? 1 : 0); +} + +nvsSwzComp NV30FP_TX_SWIZZLE[4] = {NVS_SWZ_X, NVS_SWZ_Y, NVS_SWZ_Z, NVS_SWZ_W }; + +static void +NV30FPTXSwizzle(int hwswz, nvsSwzComp *swz) +{ + swz[NVS_SWZ_W] = NV30FP_TX_SWIZZLE[(hwswz & 0xC0) >> 6]; + swz[NVS_SWZ_Z] = NV30FP_TX_SWIZZLE[(hwswz & 0x30) >> 4]; + swz[NVS_SWZ_Y] = NV30FP_TX_SWIZZLE[(hwswz & 0x0C) >> 2]; + swz[NVS_SWZ_X] = NV30FP_TX_SWIZZLE[(hwswz & 0x03) >> 0]; +} + +static void +NV30FPGetSourceSwizzle(nvsFunc * shader, int merged, int pos, nvsSwzComp *swz) +{ + unsigned int src; + int swzbits; + + src = shader->GetSourceHW(shader, merged, pos); + swzbits = (src & NV30_FP_REG_SWZ_ALL_MASK) >> NV30_FP_REG_SWZ_ALL_SHIFT; + NV30FPTXSwizzle(swzbits, swz); +} + +static int +NV30FPGetSourceIndexed(nvsFunc * shader, int merged, int pos) +{ + switch (shader->GetSourceFile(shader, merged, pos)) { + case NVS_FILE_ATTRIB: + return ((shader->inst[3] & NV30_FP_OP_INDEX_INPUT) ? 1 : 0); + default: + return 0; + } +} + +static void +NV30FPGetSourceConstVal(nvsFunc * shader, int merged, int pos, float *val) +{ + val[0] = *(float *) &(shader->inst[4]); + val[1] = *(float *) &(shader->inst[5]); + val[2] = *(float *) &(shader->inst[6]); + val[3] = *(float *) &(shader->inst[7]); +} + +static int +NV30FPGetSourceScale(nvsFunc * shader, int merged, int pos) +{ +/*FIXME: is this per-source, only for a specific source, or all sources??*/ + return (1 << ((shader->inst[2] & NV30_FP_OP_SRC_SCALE_MASK) + >> NV30_FP_OP_SRC_SCALE_SHIFT)); +} + +static int +NV30FPGetAddressRegID(nvsFunc * shader) +{ + return 0; +} + +static nvsSwzComp +NV30FPGetAddressRegSwizzle(nvsFunc * shader) +{ + return NVS_SWZ_X; +} + +static int +NV30FPSupportsConditional(nvsFunc * shader) +{ + /*FIXME: Is this true of all ops? */ + return 1; +} + +static int +NV30FPGetConditionUpdate(nvsFunc * shader) +{ + return ((shader->inst[0] & NV30_FP_OP_COND_WRITE_ENABLE) ? 1 : 0); +} + +static int +NV30FPGetConditionTest(nvsFunc * shader) +{ + /*FIXME: always? */ + return 1; +} + +static nvsCond +NV30FPGetCondition(nvsFunc * shader) +{ + int cond; + + cond = ((shader->inst[1] & NV30_FP_OP_COND_MASK) + >> NV30_FP_OP_COND_SHIFT); + + switch (cond) { + case NV30_FP_OP_COND_FL: return NVS_COND_FL; + case NV30_FP_OP_COND_LT: return NVS_COND_LT; + case NV30_FP_OP_COND_EQ: return NVS_COND_EQ; + case NV30_FP_OP_COND_LE: return NVS_COND_LE; + case NV30_FP_OP_COND_GT: return NVS_COND_GT; + case NV30_FP_OP_COND_NE: return NVS_COND_NE; + case NV30_FP_OP_COND_GE: return NVS_COND_GE; + case NV30_FP_OP_COND_TR: return NVS_COND_TR; + default: + return NVS_COND_UNKNOWN; + } +} + +static void +NV30FPGetCondRegSwizzle(nvsFunc * shader, nvsSwzComp *swz) +{ + int swzbits; + + swzbits = (shader->inst[1] & NV30_FP_OP_COND_SWZ_ALL_MASK) + >> NV30_FP_OP_COND_SWZ_ALL_SHIFT; + NV30FPTXSwizzle(swzbits, swz); +} + +static int +NV30FPGetCondRegID(nvsFunc * shader) +{ + return 0; +} + +static nvsPrecision +NV30FPGetPrecision(nvsFunc * shader) +{ + int p; + + p = (shader->inst[0] & NV30_FP_OP_PRECISION_MASK) + >> NV30_FP_OP_PRECISION_SHIFT; + + switch (p) { + case NV30_FP_PRECISION_FP32: return NVS_PREC_FLOAT32; + case NV30_FP_PRECISION_FP16: return NVS_PREC_FLOAT16; + case NV30_FP_PRECISION_FX12: return NVS_PREC_FIXED12; + default: + return NVS_PREC_UNKNOWN; + } +} + +static int +NV30FPGetSaturate(nvsFunc * shader) +{ + return ((shader->inst[0] & NV30_FP_OP_OUT_SAT) ? 1 : 0); +} + +/******************************************************************************* + * Init + */ +void +NV30FPInitShaderFuncs(nvsFunc * shader) +{ + /* These are probably bogus, I made them up... */ + shader->MaxInst = 1024; + shader->MaxAttrib = 16; + shader->MaxTemp = 32; + shader->MaxAddress = 1; + shader->MaxConst = 256; + shader->caps = SCAP_SRC_ABS; + + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_MOV, NVS_OP_MOV, 0, -1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_MUL, NVS_OP_MUL, 0, 1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_ADD, NVS_OP_ADD, 0, 1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_MAD, NVS_OP_MAD, 0, 1, 2); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_DP3, NVS_OP_DP3, 0, 1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_DP4, NVS_OP_DP4, 0, 1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_DST, NVS_OP_DST, 0, 1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_MIN, NVS_OP_MIN, 0, 1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_MAX, NVS_OP_MAX, 0, 1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_SLT, NVS_OP_SLT, 0, 1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_SGE, NVS_OP_SGE, 0, 1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_FRC, NVS_OP_FRC, 0, -1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_FLR, NVS_OP_FLR, 0, -1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_TEX, NVS_OP_TEX, 0, -1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_TXD, NVS_OP_TXD, 0, 1, 2); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_TXP, NVS_OP_TXP, 0, -1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_TXB, NVS_OP_TXB, 0, -1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_SEQ, NVS_OP_SEQ, 0, 1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_SGT, NVS_OP_SGT, 0, 1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_SLE, NVS_OP_SLE, 0, 1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_SNE, NVS_OP_SNE, 0, 1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_RCP, NVS_OP_RCP, 0, -1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_LG2, NVS_OP_LG2, 0, -1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_EX2, NVS_OP_EX2, 0, -1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_COS, NVS_OP_COS, 0, -1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_SIN, NVS_OP_SIN, 0, -1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_NOP, NVS_OP_NOP, -1, -1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_DDX, NVS_OP_DDX, 0, -1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_DDY, NVS_OP_DDY, 0, -1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_KIL, NVS_OP_KIL, -1, -1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_PK4B, NVS_OP_PK4B, 0, -1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_UP4B, NVS_OP_UP4B, 0, -1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_PK2H, NVS_OP_PK2H, 0, -1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_UP2H, NVS_OP_UP2H, 0, -1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_PK4UB, NVS_OP_PK4UB, 0, -1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_UP4UB, NVS_OP_UP4UB, 0, -1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_PK2US, NVS_OP_PK2US, 0, -1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_UP2US, NVS_OP_UP2US, 0, -1, -1); + /*FIXME: Haven't confirmed the source positions for the below opcodes */ + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_LIT, NVS_OP_LIT, 0, -1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_LRP, NVS_OP_LRP, 0, 1, 2); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_POW, NVS_OP_POW, 0, 1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_RSQ, NVS_OP_RSQ, 0, -1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV30_FP_OP_OPCODE_RFL, NVS_OP_RFL, 0, 1, -1); + + shader->GetOPTXRec = NV30FPGetOPTXRec; + shader->GetOPTXFromSOP = NV30FPGetOPTXFromSOP; + + shader->UploadToHW = NV30FPUploadToHW; + shader->UpdateConst = NV30FPUpdateConst; + + shader->SupportsOpcode = NV30FPSupportsOpcode; + shader->SetOpcode = NV30FPSetOpcode; + shader->SetCCUpdate = NV30FPSetCCUpdate; + shader->SetCondition = NV30FPSetCondition; + shader->SetResult = NV30FPSetResult; + shader->SetSource = NV30FPSetSource; + shader->SetUnusedSource = NV30FPSetUnusedSource; + shader->SetTexImageUnit = NV30FPSetTexImageUnit; + shader->SetSaturate = NV30FPSetSaturate; + shader->SetLastInst = NV30FPSetLastInst; + + shader->HasMergedInst = NV30FPHasMergedInst; + shader->IsLastInst = NV30FPIsLastInst; + shader->GetOffsetNext = NV30FPGetOffsetNext; + shader->GetOpcode = NV30FPGetOpcode; + shader->GetOpcodeHW = NV30FPGetOpcodeHW; + shader->GetDestFile = NV30FPGetDestFile; + shader->GetDestID = NV30FPGetDestID; + shader->GetDestMask = NV30FPGetDestMask; + shader->GetSourceHW = NV30FPGetSourceHW; + shader->GetSourceFile = NV30FPGetSourceFile; + shader->GetSourceID = NV30FPGetSourceID; + shader->GetTexImageUnit = NV30FPGetTexImageUnit; + shader->GetSourceNegate = NV30FPGetSourceNegate; + shader->GetSourceAbs = NV30FPGetSourceAbs; + shader->GetSourceSwizzle = NV30FPGetSourceSwizzle; + shader->GetSourceIndexed = NV30FPGetSourceIndexed; + shader->GetSourceConstVal = NV30FPGetSourceConstVal; + shader->GetSourceScale = NV30FPGetSourceScale; + shader->GetRelAddressRegID = NV30FPGetAddressRegID; + shader->GetRelAddressSwizzle = NV30FPGetAddressRegSwizzle; + shader->GetPrecision = NV30FPGetPrecision; + shader->GetSaturate = NV30FPGetSaturate; + shader->SupportsConditional = NV30FPSupportsConditional; + shader->GetConditionUpdate = NV30FPGetConditionUpdate; + shader->GetConditionTest = NV30FPGetConditionTest; + shader->GetCondition = NV30FPGetCondition; + shader->GetCondRegSwizzle = NV30FPGetCondRegSwizzle; + shader->GetCondRegID = NV30FPGetCondRegID; +} diff --git a/src/mesa/drivers/dri/nouveau/nv30_shader.h b/src/mesa/drivers/dri/nouveau/nv30_shader.h new file mode 100644 index 00000000000..d0bf6399302 --- /dev/null +++ b/src/mesa/drivers/dri/nouveau/nv30_shader.h @@ -0,0 +1,378 @@ +#ifndef __NV30_SHADER_H__ +#define __NV30_SHADER_H__ + +/* Vertex programs instruction set + * + * 128bit opcodes, split into 4 32-bit ones for ease of use. + * + * Non-native instructions + * ABS - MOV + NV40_VP_INST0_DEST_ABS + * POW - EX2 + MUL + LG2 + * SUB - ADD, second source negated + * SWZ - MOV + * XPD - + * + * Register access + * - Only one INPUT can be accessed per-instruction (move extras into TEMPs) + * - Only one CONST can be accessed per-instruction (move extras into TEMPs) + * + * Relative Addressing + * According to the value returned for MAX_PROGRAM_NATIVE_ADDRESS_REGISTERS_ARB + * there are only two address registers available. The destination in the ARL + * instruction is set to TEMP (The temp isn't actually written). + * + * When using vanilla ARB_v_p, the proprietary driver will squish both the available + * ADDRESS regs into the first hardware reg in the X and Y components. + * + * To use an address reg as an index into consts, the CONST_SRC is set to + * (const_base + offset) and INDEX_CONST is set. + * + * To access the second address reg use ADDR_REG_SELECT_1. A particular component + * of the address regs is selected with ADDR_SWZ. + * + * Only one address register can be accessed per instruction. + * + * Conditional execution (see NV_vertex_program{2,3} for details) + * Conditional execution of an instruction is enabled by setting COND_TEST_ENABLE, and + * selecting the condition which will allow the test to pass with COND_{FL,LT,...}. + * It is possible to swizzle the values in the condition register, which allows for + * testing against an individual component. + * + * Branching + * The BRA/CAL instructions seem to follow a slightly different opcode layout. The + * destination instruction ID (IADDR) overlaps a source field. Instruction ID's seem to + * be numbered based on the UPLOAD_FROM_ID FIFO command, and is incremented automatically + * on each UPLOAD_INST FIFO command. + * + * Conditional branching is achieved by using the condition tests described above. + * There doesn't appear to be dedicated looping instructions, but this can be done + * using a temp reg + conditional branching. + * + * Subroutines may be uploaded before the main program itself, but the first executed + * instruction is determined by the PROGRAM_START_ID FIFO command. + * + */ + +/* DWORD 0 */ +#define NV30_VP_INST_ADDR_REG_SELECT_1 (1 << 24) +#define NV30_VP_INST_SRC2_ABS (1 << 23) /* guess */ +#define NV30_VP_INST_SRC1_ABS (1 << 22) /* guess */ +#define NV30_VP_INST_SRC0_ABS (1 << 21) /* guess */ +#define NV30_VP_INST_OUT_RESULT (1 << 20) +#define NV30_VP_INST_DEST_TEMP_ID_SHIFT 16 +#define NV30_VP_INST_DEST_TEMP_ID_MASK (0x0F << 16) +#define NV30_VP_INST_COND_UPDATE_ENABLE (1<<15) +#define NV30_VP_INST_COND_TEST_ENABLE (1<<14) +#define NV30_VP_INST_COND_SHIFT 11 +#define NV30_VP_INST_COND_MASK (0x07 << 11) +# define NV30_VP_INST_COND_FL 0 /* guess */ +# define NV30_VP_INST_COND_LT 1 +# define NV30_VP_INST_COND_EQ 2 +# define NV30_VP_INST_COND_LE 3 +# define NV30_VP_INST_COND_GT 4 +# define NV30_VP_INST_COND_NE 5 +# define NV30_VP_INST_COND_GE 6 +# define NV30_VP_INST_COND_TR 7 /* guess */ +#define NV30_VP_INST_COND_SWZ_X_SHIFT 9 +#define NV30_VP_INST_COND_SWZ_X_MASK (0x03 << 9) +#define NV30_VP_INST_COND_SWZ_Y_SHIFT 7 +#define NV30_VP_INST_COND_SWZ_Y_MASK (0x03 << 7) +#define NV30_VP_INST_COND_SWZ_Z_SHIFT 5 +#define NV30_VP_INST_COND_SWZ_Z_MASK (0x03 << 5) +#define NV30_VP_INST_COND_SWZ_W_SHIFT 3 +#define NV30_VP_INST_COND_SWZ_W_MASK (0x03 << 3) +#define NV30_VP_INST_COND_SWZ_ALL_SHIFT 3 +#define NV30_VP_INST_COND_SWZ_ALL_MASK (0xFF << 3) +#define NV30_VP_INST_ADDR_SWZ_SHIFT 1 +#define NV30_VP_INST_ADDR_SWZ_MASK (0x03 << 1) +#define NV30_VP_INST_SCA_OPCODEH_SHIFT 0 +#define NV30_VP_INST_SCA_OPCODEH_MASK (0x01 << 0) + +/* DWORD 1 */ +#define NV30_VP_INST_SCA_OPCODEL_SHIFT 28 +#define NV30_VP_INST_SCA_OPCODEL_MASK (0x0F << 28) +# define NV30_VP_INST_OP_NOP 0x00 +# define NV30_VP_INST_OP_RCP 0x02 +# define NV30_VP_INST_OP_RCC 0x03 +# define NV30_VP_INST_OP_RSQ 0x04 +# define NV30_VP_INST_OP_EXP 0x05 +# define NV30_VP_INST_OP_LOG 0x06 +# define NV30_VP_INST_OP_LIT 0x07 +# define NV30_VP_INST_OP_BRA 0x09 +# define NV30_VP_INST_OP_CAL 0x0B +# define NV30_VP_INST_OP_RET 0x0C +# define NV30_VP_INST_OP_LG2 0x0D +# define NV30_VP_INST_OP_EX2 0x0E +# define NV30_VP_INST_OP_SIN 0x0F +# define NV30_VP_INST_OP_COS 0x10 +#define NV30_VP_INST_VEC_OPCODE_SHIFT 23 +#define NV30_VP_INST_VEC_OPCODE_MASK (0x1F << 23) +# define NV30_VP_INST_OP_NOPV 0x00 +# define NV30_VP_INST_OP_MOV 0x01 +# define NV30_VP_INST_OP_MUL 0x02 +# define NV30_VP_INST_OP_ADD 0x03 +# define NV30_VP_INST_OP_MAD 0x04 +# define NV30_VP_INST_OP_DP3 0x05 +# define NV30_VP_INST_OP_DP4 0x07 +# define NV30_VP_INST_OP_DPH 0x06 +# define NV30_VP_INST_OP_DST 0x08 +# define NV30_VP_INST_OP_MIN 0x09 +# define NV30_VP_INST_OP_MAX 0x0A +# define NV30_VP_INST_OP_SLT 0x0B +# define NV30_VP_INST_OP_SGE 0x0C +# define NV30_VP_INST_OP_ARL 0x0D +# define NV30_VP_INST_OP_FRC 0x0E +# define NV30_VP_INST_OP_FLR 0x0F +# define NV30_VP_INST_OP_SEQ 0x10 +# define NV30_VP_INST_OP_SFL 0x11 +# define NV30_VP_INST_OP_SGT 0x12 +# define NV30_VP_INST_OP_SLE 0x13 +# define NV30_VP_INST_OP_SNE 0x14 +# define NV30_VP_INST_OP_STR 0x15 +# define NV30_VP_INST_OP_SSG 0x16 +# define NV30_VP_INST_OP_ARR 0x17 +# define NV30_VP_INST_OP_ARA 0x18 +#define NV30_VP_INST_CONST_SRC_SHIFT 14 +#define NV30_VP_INST_CONST_SRC_MASK (0xFF << 14) +#define NV30_VP_INST_INPUT_SRC_SHIFT 9 /*NV20*/ +#define NV30_VP_INST_INPUT_SRC_MASK (0x0F << 9) /*NV20*/ +# define NV30_VP_INST_IN_POS 0 /* These seem to match the bindings specified in */ +# define NV30_VP_INST_IN_WEIGHT 1 /* the ARB_v_p spec (2.14.3.1) */ +# define NV30_VP_INST_IN_NORMAL 2 +# define NV30_VP_INST_IN_COL0 3 /* Should probably confirm them all though */ +# define NV30_VP_INST_IN_COL1 4 +# define NV30_VP_INST_IN_FOGC 5 +# define NV30_VP_INST_IN_TC0 8 +# define NV30_VP_INST_IN_TC(n) (8+n) +#define NV30_VP_INST_SRC0H_SHIFT 0 /*NV20*/ +#define NV30_VP_INST_SRC0H_MASK (0x1FF << 0) /*NV20*/ + +/* DWORD 2 */ +#define NV30_VP_INST_SRC0L_SHIFT 26 /*NV20*/ +#define NV30_VP_INST_SRC0L_MASK (0x3F <<26) /*NV20*/ +#define NV30_VP_INST_SRC1_SHIFT 11 /*NV20*/ +#define NV30_VP_INST_SRC1_MASK (0x7FFF<<11) /*NV20*/ +#define NV30_VP_INST_SRC2H_SHIFT 0 /*NV20*/ +#define NV30_VP_INST_SRC2H_MASK (0x7FF << 0) /*NV20*/ +#define NV30_VP_INST_IADDR_SHIFT 2 +#define NV30_VP_INST_IADDR_MASK (0xFF << 2) /* guess */ + +/* DWORD 3 */ +#define NV30_VP_INST_SRC2L_SHIFT 28 /*NV20*/ +#define NV30_VP_INST_SRC2L_MASK (0x0F <<28) /*NV20*/ +#define NV30_VP_INST_STEMP_WRITEMASK_SHIFT 24 +#define NV30_VP_INST_STEMP_WRITEMASK_MASK (0x0F << 24) +#define NV30_VP_INST_VTEMP_WRITEMASK_SHIFT 20 +#define NV30_VP_INST_VTEMP_WRITEMASK_MASK (0x0F << 20) +#define NV30_VP_INST_SDEST_WRITEMASK_SHIFT 16 +#define NV30_VP_INST_SDEST_WRITEMASK_MASK (0x0F << 16) +#define NV30_VP_INST_VDEST_WRITEMASK_SHIFT 12 /*NV20*/ +#define NV30_VP_INST_VDEST_WRITEMASK_MASK (0x0F << 12) /*NV20*/ +#define NV30_VP_INST_DEST_ID_SHIFT 2 +#define NV30_VP_INST_DEST_ID_MASK (0x0F << 2) +# define NV30_VP_INST_DEST_POS 0 +# define NV30_VP_INST_DEST_COL0 3 +# define NV30_VP_INST_DEST_COL1 4 +# define NV30_VP_INST_DEST_TC(n) (8+n) + +/* Source-register definition - matches NV20 exactly */ +#define NV30_VP_SRC_REG_NEGATE (1<<14) +#define NV30_VP_SRC_REG_SWZ_X_SHIFT 12 +#define NV30_VP_SRC_REG_SWZ_X_MASK (0x03 <<12) +#define NV30_VP_SRC_REG_SWZ_Y_SHIFT 10 +#define NV30_VP_SRC_REG_SWZ_Y_MASK (0x03 <<10) +#define NV30_VP_SRC_REG_SWZ_Z_SHIFT 8 +#define NV30_VP_SRC_REG_SWZ_Z_MASK (0x03 << 8) +#define NV30_VP_SRC_REG_SWZ_W_SHIFT 6 +#define NV30_VP_SRC_REG_SWZ_W_MASK (0x03 << 6) +#define NV30_VP_SRC_REG_SWZ_ALL_SHIFT 6 +#define NV30_VP_SRC_REG_SWZ_ALL_MASK (0xFF << 6) +#define NV30_VP_SRC_REG_TEMP_ID_SHIFT 2 +#define NV30_VP_SRC_REG_TEMP_ID_MASK (0x0F << 0) +#define NV30_VP_SRC_REG_TYPE_SHIFT 0 +#define NV30_VP_SRC_REG_TYPE_MASK (0x03 << 0) +#define NV30_VP_SRC_REG_TYPE_TEMP 1 +#define NV30_VP_SRC_REG_TYPE_INPUT 2 +#define NV30_VP_SRC_REG_TYPE_CONST 3 /* guess */ + +/* + * Each fragment program opcode appears to be comprised of 4 32-bit values. + * + * 0 - Opcode, output reg/mask, ATTRIB source + * 1 - Source 0 + * 2 - Source 1 + * 3 - Source 2 + * + * There appears to be no special difference between result regs and temp regs. + * result.color == R0.xyzw + * result.depth == R1.z + * When the fragprog contains instructions to write depth, NV30_TCL_PRIMITIVE_3D_UNK1D78=0 + * otherwise it is set to 1. + * + * Constants are inserted directly after the instruction that uses them. + * + * It appears that it's not possible to use two input registers in one + * instruction as the input sourcing is done in the instruction dword + * and not the source selection dwords. As such instructions such as: + * + * ADD result.color, fragment.color, fragment.texcoord[0]; + * + * must be split into two MOV's and then an ADD (nvidia does this) but + * I'm not sure why it's not just one MOV and then source the second input + * in the ADD instruction.. + * + * Negation of the full source is done with NV30_FP_REG_NEGATE, arbitrary + * negation requires multiplication with a const. + * + * Arbitrary swizzling is supported with the exception of SWIZZLE_ZERO/SWIZZLE_ONE + * The temp/result regs appear to be initialised to (0.0, 0.0, 0.0, 0.0) as SWIZZLE_ZERO + * is implemented simply by not writing to the relevant components of the destination. + * + * Conditional execution + * TODO + * + * Non-native instructions: + * LIT + * LRP - MAD+MAD + * SUB - ADD, negate second source + * RSQ - LG2 + EX2 + * POW - LG2 + MUL + EX2 + * SCS - COS + SIN + * XPD + */ + +//== Opcode / Destination selection == +#define NV30_FP_OP_PROGRAM_END (1 << 0) +#define NV30_FP_OP_OUT_REG_SHIFT 1 +#define NV30_FP_OP_OUT_REG_MASK (31 << 1) /* uncertain */ +/* Needs to be set when writing outputs to get expected result.. */ +#define NV30_FP_OP_UNK0_7 (1 << 7) +#define NV30_FP_OP_COND_WRITE_ENABLE (1 << 8) +#define NV30_FP_OP_OUTMASK_SHIFT 9 +#define NV30_FP_OP_OUTMASK_MASK (0xF << 9) +# define NV30_FP_OP_OUT_X (1<<9) +# define NV30_FP_OP_OUT_Y (1<<10) +# define NV30_FP_OP_OUT_Z (1<<11) +# define NV30_FP_OP_OUT_W (1<<12) +/* Uncertain about these, especially the input_src values.. it's possible that + * they can be dynamically changed. + */ +#define NV30_FP_OP_INPUT_SRC_SHIFT 13 +#define NV30_FP_OP_INPUT_SRC_MASK (15 << 13) +# define NV30_FP_OP_INPUT_SRC_POSITION 0x0 +# define NV30_FP_OP_INPUT_SRC_COL0 0x1 +# define NV30_FP_OP_INPUT_SRC_COL1 0x2 +# define NV30_FP_OP_INPUT_SRC_FOGC 0x3 +# define NV30_FP_OP_INPUT_SRC_TC0 0x4 +# define NV30_FP_OP_INPUT_SRC_TC(n) (0x4 + n) +#define NV30_FP_OP_TEX_UNIT_SHIFT 17 +#define NV30_FP_OP_TEX_UNIT_MASK (0xF << 17) /* guess */ +#define NV30_FP_OP_PRECISION_SHIFT 22 +#define NV30_FP_OP_PRECISION_MASK (3 << 22) +# define NV30_FP_PRECISION_FP32 0 +# define NV30_FP_PRECISION_FP16 1 +# define NV30_FP_PRECISION_FX12 2 +#define NV30_FP_OP_OPCODE_SHIFT 24 +#define NV30_FP_OP_OPCODE_MASK (0x3F << 24) +# define NV30_FP_OP_OPCODE_NOP 0x00 +# define NV30_FP_OP_OPCODE_MOV 0x01 +# define NV30_FP_OP_OPCODE_MUL 0x02 +# define NV30_FP_OP_OPCODE_ADD 0x03 +# define NV30_FP_OP_OPCODE_MAD 0x04 +# define NV30_FP_OP_OPCODE_DP3 0x05 +# define NV30_FP_OP_OPCODE_DP4 0x06 +# define NV30_FP_OP_OPCODE_DST 0x07 +# define NV30_FP_OP_OPCODE_MIN 0x08 +# define NV30_FP_OP_OPCODE_MAX 0x09 +# define NV30_FP_OP_OPCODE_SLT 0x0A +# define NV30_FP_OP_OPCODE_SGE 0x0B +# define NV30_FP_OP_OPCODE_SLE 0x0C +# define NV30_FP_OP_OPCODE_SGT 0x0D +# define NV30_FP_OP_OPCODE_SNE 0x0E +# define NV30_FP_OP_OPCODE_SEQ 0x0F +# define NV30_FP_OP_OPCODE_FRC 0x10 +# define NV30_FP_OP_OPCODE_FLR 0x11 +# define NV30_FP_OP_OPCODE_KIL 0x12 +# define NV30_FP_OP_OPCODE_PK4B 0x13 +# define NV30_FP_OP_OPCODE_UP4B 0x14 +# define NV30_FP_OP_OPCODE_DDX 0x15 /* can only write XY */ +# define NV30_FP_OP_OPCODE_DDY 0x16 /* can only write XY */ +# define NV30_FP_OP_OPCODE_TEX 0x17 +# define NV30_FP_OP_OPCODE_TXP 0x18 +# define NV30_FP_OP_OPCODE_TXD 0x19 +# define NV30_FP_OP_OPCODE_RCP 0x1A +# define NV30_FP_OP_OPCODE_RSQ 0x1B +# define NV30_FP_OP_OPCODE_EX2 0x1C +# define NV30_FP_OP_OPCODE_LG2 0x1D +# define NV30_FP_OP_OPCODE_LIT 0x1E +# define NV30_FP_OP_OPCODE_LRP 0x1F +# define NV30_FP_OP_OPCODE_COS 0x22 +# define NV30_FP_OP_OPCODE_SIN 0x23 +# define NV30_FP_OP_OPCODE_PK2H 0x24 +# define NV30_FP_OP_OPCODE_UP2H 0x25 +# define NV30_FP_OP_OPCODE_POW 0x26 +# define NV30_FP_OP_OPCODE_PK4UB 0x27 +# define NV30_FP_OP_OPCODE_UP4UB 0x28 +# define NV30_FP_OP_OPCODE_PK2US 0x29 +# define NV30_FP_OP_OPCODE_UP2US 0x2A +# define NV30_FP_OP_OPCODE_DP2A 0x2E +# define NV30_FP_OP_OPCODE_TXB 0x31 +# define NV30_FP_OP_OPCODE_RFL 0x36 +#define NV30_FP_OP_OUT_SAT (1 << 31) + +/* high order bits of SRC0 */ +#define NV30_FP_OP_OUT_ABS (1 << 29) +#define NV30_FP_OP_COND_SWZ_W_SHIFT 27 +#define NV30_FP_OP_COND_SWZ_W_MASK (3 << 27) +#define NV30_FP_OP_COND_SWZ_Z_SHIFT 25 +#define NV30_FP_OP_COND_SWZ_Z_MASK (3 << 25) +#define NV30_FP_OP_COND_SWZ_Y_SHIFT 23 +#define NV30_FP_OP_COND_SWZ_Y_MASK (3 << 23) +#define NV30_FP_OP_COND_SWZ_X_SHIFT 21 +#define NV30_FP_OP_COND_SWZ_X_MASK (3 << 21) +#define NV30_FP_OP_COND_SWZ_ALL_SHIFT 21 +#define NV30_FP_OP_COND_SWZ_ALL_MASK (0xFF << 21) +#define NV30_FP_OP_COND_SHIFT 18 +#define NV30_FP_OP_COND_MASK (0x07 << 18) +# define NV30_FP_OP_COND_FL 0 +# define NV30_FP_OP_COND_LT 1 +# define NV30_FP_OP_COND_EQ 2 +# define NV30_FP_OP_COND_LE 3 +# define NV30_FP_OP_COND_GT 4 +# define NV30_FP_OP_COND_NE 5 +# define NV30_FP_OP_COND_GE 6 +# define NV30_FP_OP_COND_TR 7 + +/* high order bits of SRC1 */ +#define NV30_FP_OP_SRC_SCALE_SHIFT 28 +#define NV30_FP_OP_SRC_SCALE_MASK (3 << 28) + +/* high order bits of SRC2 */ +#define NV30_FP_OP_INDEX_INPUT (1 << 30) + +//== Register selection == +#define NV30_FP_REG_TYPE_SHIFT 0 +#define NV30_FP_REG_TYPE_MASK (3 << 0) +# define NV30_FP_REG_TYPE_TEMP 0 +# define NV30_FP_REG_TYPE_INPUT 1 +# define NV30_FP_REG_TYPE_CONST 2 +#define NV30_FP_REG_SRC_SHIFT 2 /* uncertain */ +#define NV30_FP_REG_SRC_MASK (31 << 2) +#define NV30_FP_REG_UNK_0 (1 << 8) +#define NV30_FP_REG_SWZ_ALL_SHIFT 9 +#define NV30_FP_REG_SWZ_ALL_MASK (255 << 9) +#define NV30_FP_REG_SWZ_X_SHIFT 9 +#define NV30_FP_REG_SWZ_X_MASK (3 << 9) +#define NV30_FP_REG_SWZ_Y_SHIFT 11 +#define NV30_FP_REG_SWZ_Y_MASK (3 << 11) +#define NV30_FP_REG_SWZ_Z_SHIFT 13 +#define NV30_FP_REG_SWZ_Z_MASK (3 << 13) +#define NV30_FP_REG_SWZ_W_SHIFT 15 +#define NV30_FP_REG_SWZ_W_MASK (3 << 15) +# define NV30_FP_SWIZZLE_X 0 +# define NV30_FP_SWIZZLE_Y 1 +# define NV30_FP_SWIZZLE_Z 2 +# define NV30_FP_SWIZZLE_W 3 +#define NV30_FP_REG_NEGATE (1 << 17) + +#endif diff --git a/src/mesa/drivers/dri/nouveau/nv30_state.c b/src/mesa/drivers/dri/nouveau/nv30_state.c index 851641c0c96..3ffb5d3a41f 100644 --- a/src/mesa/drivers/dri/nouveau/nv30_state.c +++ b/src/mesa/drivers/dri/nouveau/nv30_state.c @@ -327,7 +327,7 @@ static void nv30Fogfv(GLcontext *ctx, GLenum pname, const GLfloat *params) switch(pname) { case GL_FOG_MODE: - BEGIN_RING_CACHE(NvSub3D, NV30_TCL_PRIMITIVE_3D_FOG_MODE, 1); + //BEGIN_RING_CACHE(NvSub3D, NV30_TCL_PRIMITIVE_3D_FOG_MODE, 1); //OUT_RING_CACHE (params); break; /* TODO: unsure about the rest.*/ diff --git a/src/mesa/drivers/dri/nouveau/nv30_vertprog.c b/src/mesa/drivers/dri/nouveau/nv30_vertprog.c new file mode 100644 index 00000000000..e60422dad10 --- /dev/null +++ b/src/mesa/drivers/dri/nouveau/nv30_vertprog.c @@ -0,0 +1,356 @@ +#include "nouveau_context.h" +#include "nouveau_object.h" +#include "nouveau_fifo.h" +#include "nouveau_reg.h" + +#include "nouveau_shader.h" +#include "nv30_shader.h" + +extern nvsSwzComp NV20VP_TX_SWIZZLE[4]; +extern void NV20VPTXSwizzle(int hwswz, nvsSwzComp *swz); + +/***************************************************************************** + * Support routines + */ +static void +NV30VPUploadToHW(GLcontext *ctx, nouveauShader *nvs) +{ + nouveauContextPtr nmesa = NOUVEAU_CONTEXT(ctx); + int i; + + /* We can do better here and keep more than one VP on the hardware, and + * switch between them with PROGRAM_START_ID.. + */ + BEGIN_RING_SIZE(NvSub3D, NV30_TCL_PRIMITIVE_3D_VP_UPLOAD_FROM_ID, 1); + OUT_RING(0); + for (i=0; iprogram_size; i+=4) { + BEGIN_RING_SIZE(NvSub3D, NV30_TCL_PRIMITIVE_3D_VP_UPLOAD_INST0, 4); + OUT_RING(nvs->program[i + 0]); + OUT_RING(nvs->program[i + 1]); + OUT_RING(nvs->program[i + 2]); + OUT_RING(nvs->program[i + 3]); + } + BEGIN_RING_SIZE(NvSub3D, NV30_TCL_PRIMITIVE_3D_VP_PROGRAM_START_ID, 1); + OUT_RING(0); +} + +static void +NV30VPUpdateConst(GLcontext *ctx, nouveauShader *nvs, int id) +{ + nouveauContextPtr nmesa = NOUVEAU_CONTEXT(ctx); + GLfloat *val; + + val = nvs->params[id].source_val ? + nvs->params[id].source_val : nvs->params[id].val; + + BEGIN_RING_SIZE(NvSub3D, NV30_TCL_PRIMITIVE_3D_VP_UPLOAD_CONST_ID, 5); + OUT_RING (id); + OUT_RINGp(val, 4); +} + +/***************************************************************************** + * Assembly routines + */ + +/***************************************************************************** + * Disassembly routines + */ +static unsigned int +NV30VPGetOpcodeHW(nvsFunc * shader, int slot) +{ + int op; + + if (slot) { + op = (shader->inst[1] & NV30_VP_INST_SCA_OPCODEL_MASK) + >> NV30_VP_INST_SCA_OPCODEL_SHIFT; + op |= ((shader->inst[0] & NV30_VP_INST_SCA_OPCODEH_MASK) + >> NV30_VP_INST_SCA_OPCODEH_SHIFT) << 4; + } + else { + op = (shader->inst[1] & NV30_VP_INST_VEC_OPCODE_MASK) + >> NV30_VP_INST_VEC_OPCODE_SHIFT; + } + + return op; +} + +static nvsRegFile +NV30VPGetDestFile(nvsFunc * shader, int merged) +{ + switch (shader->GetOpcode(shader, merged)) { + case NVS_OP_ARL: + case NVS_OP_ARR: + case NVS_OP_ARA: + return NVS_FILE_ADDRESS; + default: + /*FIXME: This probably isn't correct.. */ + if ((shader->inst[3] & NV30_VP_INST_VDEST_WRITEMASK_MASK) != 0) + return NVS_FILE_RESULT; + if ((shader->inst[3] & NV30_VP_INST_SDEST_WRITEMASK_MASK) != 0) + return NVS_FILE_RESULT; + return NVS_FILE_TEMP; + } +} + +static unsigned int +NV30VPGetDestID(nvsFunc * shader, int merged) +{ + int id; + + switch (shader->GetDestFile(shader, merged)) { + case NVS_FILE_RESULT: + id = ((shader->inst[3] & NV30_VP_INST_DEST_ID_MASK) + >> NV30_VP_INST_DEST_ID_SHIFT); + switch (id) { + case NV30_VP_INST_DEST_POS : return NVS_FR_POSITION; + case NV30_VP_INST_DEST_COL0 : return NVS_FR_COL0; + case NV30_VP_INST_DEST_COL1 : return NVS_FR_COL1; + case NV30_VP_INST_DEST_TC(0): return NVS_FR_TEXCOORD0; + case NV30_VP_INST_DEST_TC(1): return NVS_FR_TEXCOORD1; + case NV30_VP_INST_DEST_TC(2): return NVS_FR_TEXCOORD2; + case NV30_VP_INST_DEST_TC(3): return NVS_FR_TEXCOORD3; + case NV30_VP_INST_DEST_TC(4): return NVS_FR_TEXCOORD4; + case NV30_VP_INST_DEST_TC(5): return NVS_FR_TEXCOORD5; + case NV30_VP_INST_DEST_TC(6): return NVS_FR_TEXCOORD6; + case NV30_VP_INST_DEST_TC(7): return NVS_FR_TEXCOORD7; + default: + return -1; + } + case NVS_FILE_ADDRESS: + case NVS_FILE_TEMP: + return (shader->inst[0] & NV30_VP_INST_DEST_TEMP_ID_MASK) + >> NV30_VP_INST_DEST_TEMP_ID_SHIFT; + default: + return -1; + } +} + +static unsigned int +NV30VPGetDestMask(nvsFunc * shader, int merged) +{ + int hwmask, mask = 0; + + if (shader->GetDestFile(shader, merged) == NVS_FILE_RESULT) + if (shader->GetOpcodeSlot(shader, merged)) + hwmask = (shader->inst[3] & NV30_VP_INST_SDEST_WRITEMASK_MASK) + >> NV30_VP_INST_SDEST_WRITEMASK_SHIFT; + else + hwmask = (shader->inst[3] & NV30_VP_INST_VDEST_WRITEMASK_MASK) + >> NV30_VP_INST_VDEST_WRITEMASK_SHIFT; + else if (shader->GetOpcodeSlot(shader, merged)) + hwmask = (shader->inst[3] & NV30_VP_INST_STEMP_WRITEMASK_MASK) + >> NV30_VP_INST_STEMP_WRITEMASK_SHIFT; + else + hwmask = (shader->inst[3] & NV30_VP_INST_VTEMP_WRITEMASK_MASK) + >> NV30_VP_INST_VTEMP_WRITEMASK_SHIFT; + + if (hwmask & (1 << 3)) mask |= SMASK_X; + if (hwmask & (1 << 2)) mask |= SMASK_Y; + if (hwmask & (1 << 1)) mask |= SMASK_Z; + if (hwmask & (1 << 0)) mask |= SMASK_W; + + return mask; +} + +static int +NV30VPGetSourceID(nvsFunc * shader, int merged, int pos) +{ + unsigned int src; + + switch (shader->GetSourceFile(shader, merged, pos)) { + case NVS_FILE_TEMP: + src = shader->GetSourceHW(shader, merged, pos); + return ((src & NV30_VP_SRC_REG_TEMP_ID_MASK) >> + NV30_VP_SRC_REG_TEMP_ID_SHIFT); + case NVS_FILE_CONST: + return ((shader->inst[1] & NV30_VP_INST_CONST_SRC_MASK) + >> NV30_VP_INST_CONST_SRC_SHIFT); + case NVS_FILE_ATTRIB: + src = ((shader->inst[1] & NV30_VP_INST_INPUT_SRC_MASK) + >> NV30_VP_INST_INPUT_SRC_SHIFT); + switch (src) { + case NV30_VP_INST_IN_POS : return NVS_FR_POSITION; + case NV30_VP_INST_IN_COL0 : return NVS_FR_COL0; + case NV30_VP_INST_IN_COL1 : return NVS_FR_COL1; + case NV30_VP_INST_IN_TC(0): return NVS_FR_TEXCOORD0; + case NV30_VP_INST_IN_TC(1): return NVS_FR_TEXCOORD1; + case NV30_VP_INST_IN_TC(2): return NVS_FR_TEXCOORD2; + case NV30_VP_INST_IN_TC(3): return NVS_FR_TEXCOORD3; + case NV30_VP_INST_IN_TC(4): return NVS_FR_TEXCOORD4; + case NV30_VP_INST_IN_TC(5): return NVS_FR_TEXCOORD5; + case NV30_VP_INST_IN_TC(6): return NVS_FR_TEXCOORD6; + case NV30_VP_INST_IN_TC(7): return NVS_FR_TEXCOORD7; + default: + return NVS_FR_UNKNOWN; + } + default: + return -1; + } +} + +static int +NV30VPGetSourceAbs(nvsFunc * shader, int merged, int pos) +{ + struct _op_xlat *opr; + static unsigned int abspos[3] = { + NV30_VP_INST_SRC0_ABS, + NV30_VP_INST_SRC1_ABS, + NV30_VP_INST_SRC2_ABS, + }; + + opr = shader->GetOPTXRec(shader, merged); + if (!opr || opr->srcpos[pos] == -1 || opr->srcpos[pos] > 2) + return 0; + + return ((shader->inst[0] & abspos[opr->srcpos[pos]]) ? 1 : 0); +} + +static int +NV30VPGetRelAddressRegID(nvsFunc * shader) +{ + return ((shader->inst[0] & NV30_VP_INST_ADDR_REG_SELECT_1) ? 1 : 0); +} + +static nvsSwzComp +NV30VPGetRelAddressSwizzle(nvsFunc * shader) +{ + nvsSwzComp swz; + + swz = NV20VP_TX_SWIZZLE[(shader->inst[0] & NV30_VP_INST_ADDR_SWZ_MASK) + >> NV30_VP_INST_ADDR_SWZ_SHIFT]; + return swz; +} + +static int +NV30VPSupportsConditional(nvsFunc * shader) +{ + /*FIXME: Is this true of all ops? */ + return 1; +} + +static int +NV30VPGetConditionUpdate(nvsFunc * shader) +{ + return ((shader->inst[0] & NV30_VP_INST_COND_UPDATE_ENABLE) ? 1 : 0); +} + +static int +NV30VPGetConditionTest(nvsFunc * shader) +{ + int op; + + /* The condition test is unconditionally enabled on some + * instructions. ie: the condition test bit does *NOT* have + * to be set. + * + * FIXME: check other relevant ops for this situation. + */ + op = shader->GetOpcodeHW(shader, 1); + switch (op) { + case NV30_VP_INST_OP_BRA: + return 1; + default: + return ((shader->inst[0] & NV30_VP_INST_COND_TEST_ENABLE) ? 1 : 0); + } +} + +static nvsCond +NV30VPGetCondition(nvsFunc * shader) +{ + int cond; + + cond = ((shader->inst[0] & NV30_VP_INST_COND_MASK) + >> NV30_VP_INST_COND_SHIFT); + + switch (cond) { + case NV30_VP_INST_COND_FL: return NVS_COND_FL; + case NV30_VP_INST_COND_LT: return NVS_COND_LT; + case NV30_VP_INST_COND_EQ: return NVS_COND_EQ; + case NV30_VP_INST_COND_LE: return NVS_COND_LE; + case NV30_VP_INST_COND_GT: return NVS_COND_GT; + case NV30_VP_INST_COND_NE: return NVS_COND_NE; + case NV30_VP_INST_COND_GE: return NVS_COND_GE; + case NV30_VP_INST_COND_TR: return NVS_COND_TR; + default: + return NVS_COND_UNKNOWN; + } +} + +static void +NV30VPGetCondRegSwizzle(nvsFunc * shader, nvsSwzComp *swz) +{ + int swzbits; + + swzbits = (shader->inst[0] & NV30_VP_INST_COND_SWZ_ALL_MASK) + >> NV30_VP_INST_COND_SWZ_ALL_SHIFT; + NV20VPTXSwizzle(swzbits, swz); +} + +static int +NV30VPGetCondRegID(nvsFunc * shader) +{ + return 0; +} + + +static int +NV30VPGetBranch(nvsFunc * shader) +{ + return ((shader->inst[2] & NV30_VP_INST_IADDR_MASK) + >> NV30_VP_INST_IADDR_SHIFT); +} + +void +NV30VPInitShaderFuncs(nvsFunc * shader) +{ + /* Inherit NV20 code, a lot of it is the same */ + NV20VPInitShaderFuncs(shader); + + /* Increase max valid opcode ID, and add new instructions */ + NVVP_TX_VOP_COUNT = NVVP_TX_NVS_OP_COUNT = 32; + + MOD_OPCODE(NVVP_TX_VOP, NV30_VP_INST_OP_FRC, NVS_OP_FRC, 0, -1, -1); + MOD_OPCODE(NVVP_TX_VOP, NV30_VP_INST_OP_FLR, NVS_OP_FLR, 0, -1, -1); + MOD_OPCODE(NVVP_TX_VOP, NV30_VP_INST_OP_SEQ, NVS_OP_SEQ, 0, 1, -1); + MOD_OPCODE(NVVP_TX_VOP, NV30_VP_INST_OP_SFL, NVS_OP_SFL, 0, 1, -1); + MOD_OPCODE(NVVP_TX_VOP, NV30_VP_INST_OP_SGT, NVS_OP_SGT, 0, 1, -1); + MOD_OPCODE(NVVP_TX_VOP, NV30_VP_INST_OP_SLE, NVS_OP_SLE, 0, 1, -1); + MOD_OPCODE(NVVP_TX_VOP, NV30_VP_INST_OP_SNE, NVS_OP_SNE, 0, 1, -1); + MOD_OPCODE(NVVP_TX_VOP, NV30_VP_INST_OP_STR, NVS_OP_STR, 0, 1, -1); + MOD_OPCODE(NVVP_TX_VOP, NV30_VP_INST_OP_SSG, NVS_OP_SSG, 0, -1, -1); + MOD_OPCODE(NVVP_TX_VOP, NV30_VP_INST_OP_ARR, NVS_OP_ARR, 0, -1, -1); + MOD_OPCODE(NVVP_TX_VOP, NV30_VP_INST_OP_ARA, NVS_OP_ARA, 3, -1, -1); + + MOD_OPCODE(NVVP_TX_SOP, NV30_VP_INST_OP_BRA, NVS_OP_BRA, -1, -1, -1); + MOD_OPCODE(NVVP_TX_SOP, NV30_VP_INST_OP_CAL, NVS_OP_CAL, -1, -1, -1); + MOD_OPCODE(NVVP_TX_SOP, NV30_VP_INST_OP_RET, NVS_OP_RET, -1, -1, -1); + MOD_OPCODE(NVVP_TX_SOP, NV30_VP_INST_OP_LG2, NVS_OP_LG2, 2, -1, -1); + MOD_OPCODE(NVVP_TX_SOP, NV30_VP_INST_OP_EX2, NVS_OP_EX2, 2, -1, -1); + MOD_OPCODE(NVVP_TX_SOP, NV30_VP_INST_OP_SIN, NVS_OP_SIN, 2, -1, -1); + MOD_OPCODE(NVVP_TX_SOP, NV30_VP_INST_OP_COS, NVS_OP_COS, 2, -1, -1); + + shader->UploadToHW = NV30VPUploadToHW; + shader->UpdateConst = NV30VPUpdateConst; + + shader->GetOpcodeHW = NV30VPGetOpcodeHW; + + shader->GetDestFile = NV30VPGetDestFile; + shader->GetDestID = NV30VPGetDestID; + shader->GetDestMask = NV30VPGetDestMask; + + shader->GetSourceID = NV30VPGetSourceID; + shader->GetSourceAbs = NV30VPGetSourceAbs; + + shader->GetRelAddressRegID = NV30VPGetRelAddressRegID; + shader->GetRelAddressSwizzle = NV30VPGetRelAddressSwizzle; + + shader->SupportsConditional = NV30VPSupportsConditional; + shader->GetConditionUpdate = NV30VPGetConditionUpdate; + shader->GetConditionTest = NV30VPGetConditionTest; + shader->GetCondition = NV30VPGetCondition; + shader->GetCondRegSwizzle = NV30VPGetCondRegSwizzle; + shader->GetCondRegID = NV30VPGetCondRegID; + + shader->GetBranch = NV30VPGetBranch; +} + diff --git a/src/mesa/drivers/dri/nouveau/nv40_fragprog.c b/src/mesa/drivers/dri/nouveau/nv40_fragprog.c new file mode 100644 index 00000000000..3d58d6b6667 --- /dev/null +++ b/src/mesa/drivers/dri/nouveau/nv40_fragprog.c @@ -0,0 +1,152 @@ +#include "nouveau_shader.h" +#include "nv40_shader.h" + +/* branching ops */ +unsigned int NVFP_TX_BOP_COUNT = 5; +struct _op_xlat NVFP_TX_BOP[64]; + +static struct _op_xlat * +NV40FPGetOPTXRec(nvsFunc * shader, int merged) +{ + struct _op_xlat *opr; + int op; + + op = shader->GetOpcodeHW(shader, 0); + if (shader->inst[2] & NV40_FP_OP_OPCODE_IS_BRANCH) { + opr = NVFP_TX_BOP; + op &= ~NV40_FP_OP_OPCODE_IS_BRANCH; + if (op > NVFP_TX_BOP_COUNT) + return NULL; + } + else { + opr = NVFP_TX_AOP; + if (op > NVFP_TX_AOP_COUNT) + return NULL; + } + + if (opr[op].SOP == NVS_OP_UNKNOWN) + return NULL; + return &opr[op]; +} + +static int +NV40FPGetSourceID(nvsFunc * shader, int merged, int pos) +{ + switch (shader->GetSourceFile(shader, merged, pos)) { + case NVS_FILE_ATTRIB: + switch ((shader->inst[0] & NV40_FP_OP_INPUT_SRC_MASK) + >> NV40_FP_OP_INPUT_SRC_SHIFT) { + case NV40_FP_OP_INPUT_SRC_POSITION: return NVS_FR_POSITION; + case NV40_FP_OP_INPUT_SRC_COL0 : return NVS_FR_COL0; + case NV40_FP_OP_INPUT_SRC_COL1 : return NVS_FR_COL1; + case NV40_FP_OP_INPUT_SRC_FOGC : return NVS_FR_FOGCOORD; + case NV40_FP_OP_INPUT_SRC_TC(0) : return NVS_FR_TEXCOORD0; + case NV40_FP_OP_INPUT_SRC_TC(1) : return NVS_FR_TEXCOORD1; + case NV40_FP_OP_INPUT_SRC_TC(2) : return NVS_FR_TEXCOORD2; + case NV40_FP_OP_INPUT_SRC_TC(3) : return NVS_FR_TEXCOORD3; + case NV40_FP_OP_INPUT_SRC_TC(4) : return NVS_FR_TEXCOORD4; + case NV40_FP_OP_INPUT_SRC_TC(5) : return NVS_FR_TEXCOORD5; + case NV40_FP_OP_INPUT_SRC_TC(6) : return NVS_FR_TEXCOORD6; + case NV40_FP_OP_INPUT_SRC_TC(7) : return NVS_FR_TEXCOORD7; + case NV40_FP_OP_INPUT_SRC_FACING : return NVS_FR_FACING; + default: + return -1; + } + break; + case NVS_FILE_TEMP: + { + unsigned int src; + + src = shader->GetSourceHW(shader, merged, pos); + return ((src & NV40_FP_REG_SRC_MASK) >> NV40_FP_REG_SRC_SHIFT); + } + case NVS_FILE_CONST: /* inlined into fragprog */ + default: + return -1; + } +} + +static int +NV40FPGetBranch(nvsFunc * shader) +{ + return ((shader->inst[2] & NV40_FP_OP_IADDR_MASK) + >> NV40_FP_OP_IADDR_SHIFT);; +} + +static int +NV40FPGetBranchElse(nvsFunc * shader) +{ + return ((shader->inst[2] & NV40_FP_OP_ELSE_ID_MASK) + >> NV40_FP_OP_ELSE_ID_SHIFT); +} + +static int +NV40FPGetBranchEnd(nvsFunc * shader) +{ + return ((shader->inst[3] & NV40_FP_OP_END_ID_MASK) + >> NV40_FP_OP_END_ID_SHIFT); +} + +static int +NV40FPGetLoopCount(nvsFunc * shader) +{ + return ((shader->inst[2] & NV40_FP_OP_LOOP_COUNT_MASK) + >> NV40_FP_OP_LOOP_COUNT_SHIFT); +} + +static int +NV40FPGetLoopInitial(nvsFunc * shader) +{ + return ((shader->inst[2] & NV40_FP_OP_LOOP_INDEX_MASK) + >> NV40_FP_OP_LOOP_INDEX_SHIFT); +} + +static int +NV40FPGetLoopIncrement(nvsFunc * shader) +{ + return ((shader->inst[2] & NV40_FP_OP_LOOP_INCR_MASK) + >> NV40_FP_OP_LOOP_INCR_SHIFT); +} + +void +NV40FPInitShaderFuncs(nvsFunc * shader) +{ + /* Inherit NV30 FP code, it's mostly the same */ + NV30FPInitShaderFuncs(shader); + + /* Kill off opcodes seen on NV30, but not seen on NV40 - need to find + * out if these actually work or not. + * + * update: either LIT/RSQ don't work on nv40, or I generate bad code for + * them. haven't tested the others yet + */ + MOD_OPCODE(NVFP_TX_AOP, 0x1B, NVS_OP_UNKNOWN, -1, -1, -1); /* NV30 RSQ */ + MOD_OPCODE(NVFP_TX_AOP, 0x1E, NVS_OP_UNKNOWN, -1, -1, -1); /* NV30 LIT */ + MOD_OPCODE(NVFP_TX_AOP, 0x1F, NVS_OP_UNKNOWN, -1, -1, -1); /* NV30 LRP */ + MOD_OPCODE(NVFP_TX_AOP, 0x26, NVS_OP_UNKNOWN, -1, -1, -1); /* NV30 POW */ + MOD_OPCODE(NVFP_TX_AOP, 0x36, NVS_OP_UNKNOWN, -1, -1, -1); /* NV30 RFL */ + + /* Extra opcodes supported on NV40 */ + MOD_OPCODE(NVFP_TX_AOP, NV40_FP_OP_OPCODE_DIV , NVS_OP_DIV , 0, 1, -1); + MOD_OPCODE(NVFP_TX_AOP, NV40_FP_OP_OPCODE_DP2A , NVS_OP_DP2A, 0, 1, 2); + MOD_OPCODE(NVFP_TX_AOP, NV40_FP_OP_OPCODE_TXL , NVS_OP_TXL , 0, -1, -1); + + MOD_OPCODE(NVFP_TX_BOP, NV40_FP_OP_BRA_OPCODE_BRK , NVS_OP_BRK , -1, -1, -1); + MOD_OPCODE(NVFP_TX_BOP, NV40_FP_OP_BRA_OPCODE_CAL , NVS_OP_CAL , -1, -1, -1); + MOD_OPCODE(NVFP_TX_BOP, NV40_FP_OP_BRA_OPCODE_IF , NVS_OP_IF , -1, -1, -1); + MOD_OPCODE(NVFP_TX_BOP, NV40_FP_OP_BRA_OPCODE_LOOP, NVS_OP_LOOP, -1, -1, -1); + MOD_OPCODE(NVFP_TX_BOP, NV40_FP_OP_BRA_OPCODE_REP , NVS_OP_REP , -1, -1, -1); + MOD_OPCODE(NVFP_TX_BOP, NV40_FP_OP_BRA_OPCODE_RET , NVS_OP_RET , -1, -1, -1); + + /* fragment.facing */ + shader->GetSourceID = NV40FPGetSourceID; + + /* branching */ + shader->GetOPTXRec = NV40FPGetOPTXRec; + shader->GetBranch = NV40FPGetBranch; + shader->GetBranchElse = NV40FPGetBranchElse; + shader->GetBranchEnd = NV40FPGetBranchEnd; + shader->GetLoopCount = NV40FPGetLoopCount; + shader->GetLoopInitial = NV40FPGetLoopInitial; + shader->GetLoopIncrement = NV40FPGetLoopIncrement; +} diff --git a/src/mesa/drivers/dri/nouveau/nv40_shader.h b/src/mesa/drivers/dri/nouveau/nv40_shader.h new file mode 100644 index 00000000000..2a2b5639b6c --- /dev/null +++ b/src/mesa/drivers/dri/nouveau/nv40_shader.h @@ -0,0 +1,467 @@ +#ifndef __NV40_SHADER_H__ +#define __NV40_SHADER_H__ + +/* Vertex programs instruction set + * + * The NV40 instruction set is very similar to NV30. Most fields are in + * a slightly different position in the instruction however. + * + * Merged instructions + * In some cases it is possible to put two instructions into one opcode + * slot. The rules for when this is OK is not entirely clear to me yet. + * + * There are separate writemasks and dest temp register fields for each + * grouping of instructions. There is however only one field with the + * ID of a result register. Writing to temp/result regs is selected by + * setting VEC_RESULT/SCA_RESULT. + * + * Temporary registers + * The source/dest temp register fields have been extended by 1 bit, to + * give a total of 32 temporary registers. + * + * Relative Addressing + * NV40 can use an address register to index into vertex attribute regs. + * This is done by putting the offset value into INPUT_SRC and setting + * the INDEX_INPUT flag. + * + * Conditional execution (see NV_vertex_program{2,3} for details) + * There is a second condition code register on NV40, it's use is enabled + * by setting the COND_REG_SELECT_1 flag. + * + * Texture lookup + * TODO + */ + +/* ---- OPCODE BITS 127:96 / data DWORD 0 --- */ +#define NV40_VP_INST_VEC_RESULT (1 << 30) +/* uncertain.. */ +#define NV40_VP_INST_COND_UPDATE_ENABLE ((1 << 14)|1<<29) +/* use address reg as index into attribs */ +#define NV40_VP_INST_INDEX_INPUT (1 << 27) +#define NV40_VP_INST_COND_REG_SELECT_1 (1 << 25) +#define NV40_VP_INST_ADDR_REG_SELECT_1 (1 << 24) +#define NV40_VP_INST_SRC2_ABS (1 << 23) +#define NV40_VP_INST_SRC1_ABS (1 << 22) +#define NV40_VP_INST_SRC0_ABS (1 << 21) +#define NV40_VP_INST_VEC_DEST_TEMP_SHIFT 15 +#define NV40_VP_INST_VEC_DEST_TEMP_MASK (0x1F << 15) +#define NV40_VP_INST_COND_TEST_ENABLE (1 << 13) +#define NV40_VP_INST_COND_SHIFT 10 +#define NV40_VP_INST_COND_MASK (0x7 << 10) +# define NV40_VP_INST_COND_FL 0 +# define NV40_VP_INST_COND_LT 1 +# define NV40_VP_INST_COND_EQ 2 +# define NV40_VP_INST_COND_LE 3 +# define NV40_VP_INST_COND_GT 4 +# define NV40_VP_INST_COND_NE 5 +# define NV40_VP_INST_COND_GE 6 +# define NV40_VP_INST_COND_TR 7 +#define NV40_VP_INST_COND_SWZ_X_SHIFT 8 +#define NV40_VP_INST_COND_SWZ_X_MASK (3 << 8) +#define NV40_VP_INST_COND_SWZ_Y_SHIFT 6 +#define NV40_VP_INST_COND_SWZ_Y_MASK (3 << 6) +#define NV40_VP_INST_COND_SWZ_Z_SHIFT 4 +#define NV40_VP_INST_COND_SWZ_Z_MASK (3 << 4) +#define NV40_VP_INST_COND_SWZ_W_SHIFT 2 +#define NV40_VP_INST_COND_SWZ_W_MASK (3 << 2) +#define NV40_VP_INST_COND_SWZ_ALL_SHIFT 2 +#define NV40_VP_INST_COND_SWZ_ALL_MASK (0xFF << 2) +#define NV40_VP_INST_ADDR_SWZ_SHIFT 0 +#define NV40_VP_INST_ADDR_SWZ_MASK (0x03 << 0) +#define NV40_VP_INST0_KNOWN ( \ + NV40_VP_INST_INDEX_INPUT | \ + NV40_VP_INST_COND_REG_SELECT_1 | \ + NV40_VP_INST_ADDR_REG_SELECT_1 | \ + NV40_VP_INST_SRC2_ABS | \ + NV40_VP_INST_SRC1_ABS | \ + NV40_VP_INST_SRC0_ABS | \ + NV40_VP_INST_VEC_DEST_TEMP_MASK | \ + NV40_VP_INST_COND_TEST_ENABLE | \ + NV40_VP_INST_COND_MASK | \ + NV40_VP_INST_COND_SWZ_ALL_MASK | \ + NV40_VP_INST_ADDR_SWZ_MASK) + +/* ---- OPCODE BITS 95:64 / data DWORD 1 --- */ +#define NV40_VP_INST_VEC_OPCODE_SHIFT 22 +#define NV40_VP_INST_VEC_OPCODE_MASK (0x1F << 22) +# define NV40_VP_INST_OP_NOP 0x00 +# define NV40_VP_INST_OP_MOV 0x01 +# define NV40_VP_INST_OP_MUL 0x02 +# define NV40_VP_INST_OP_ADD 0x03 +# define NV40_VP_INST_OP_MAD 0x04 +# define NV40_VP_INST_OP_DP3 0x05 +# define NV40_VP_INST_OP_DP4 0x07 +# define NV40_VP_INST_OP_DPH 0x06 +# define NV40_VP_INST_OP_DST 0x08 +# define NV40_VP_INST_OP_MIN 0x09 +# define NV40_VP_INST_OP_MAX 0x0A +# define NV40_VP_INST_OP_SLT 0x0B +# define NV40_VP_INST_OP_SGE 0x0C +# define NV40_VP_INST_OP_ARL 0x0D +# define NV40_VP_INST_OP_FRC 0x0E +# define NV40_VP_INST_OP_FLR 0x0F +# define NV40_VP_INST_OP_SEQ 0x10 +# define NV40_VP_INST_OP_SFL 0x11 +# define NV40_VP_INST_OP_SGT 0x12 +# define NV40_VP_INST_OP_SLE 0x13 +# define NV40_VP_INST_OP_SNE 0x14 +# define NV40_VP_INST_OP_STR 0x15 +# define NV40_VP_INST_OP_SSG 0x16 +# define NV40_VP_INST_OP_ARR 0x17 +# define NV40_VP_INST_OP_ARA 0x18 +# define NV40_VP_INST_OP_TXWHAT 0x19 +#define NV40_VP_INST_SCA_OPCODE_SHIFT 27 +#define NV40_VP_INST_SCA_OPCODE_MASK (0x1F << 27) +# define NV40_VP_INST_OP_RCP 0x02 +# define NV40_VP_INST_OP_RCC 0x03 +# define NV40_VP_INST_OP_RSQ 0x04 +# define NV40_VP_INST_OP_EXP 0x05 +# define NV40_VP_INST_OP_LOG 0x06 +# define NV40_VP_INST_OP_LIT 0x07 +# define NV40_VP_INST_OP_BRA 0x09 +# define NV40_VP_INST_OP_CAL 0x0B +# define NV40_VP_INST_OP_RET 0x0C +# define NV40_VP_INST_OP_LG2 0x0D +# define NV40_VP_INST_OP_EX2 0x0E +# define NV40_VP_INST_OP_SIN 0x0F +# define NV40_VP_INST_OP_COS 0x10 +# define NV40_VP_INST_OP_PUSHA 0x13 +# define NV40_VP_INST_OP_POPA 0x14 +#define NV40_VP_INST_CONST_SRC_SHIFT 12 +#define NV40_VP_INST_CONST_SRC_MASK (0xFF << 12) +#define NV40_VP_INST_INPUT_SRC_SHIFT 8 +#define NV40_VP_INST_INPUT_SRC_MASK (0x0F << 8) +# define NV40_VP_INST_IN_POS 0 +# define NV40_VP_INST_IN_WEIGHT 1 +# define NV40_VP_INST_IN_NORMAL 2 +# define NV40_VP_INST_IN_COL0 3 +# define NV40_VP_INST_IN_COL1 4 +# define NV40_VP_INST_IN_FOGC 5 +# define NV40_VP_INST_IN_TC0 8 +# define NV40_VP_INST_IN_TC(n) (8+n) +#define NV40_VP_INST_SRC0H_SHIFT 0 +#define NV40_VP_INST_SRC0H_MASK (0xFF << 0) +#define NV40_VP_INST1_KNOWN ( \ + NV40_VP_INST_VEC_OPCODE_MASK | \ + NV40_VP_INST_SCA_OPCODE_MASK | \ + NV40_VP_INST_CONST_SRC_MASK | \ + NV40_VP_INST_INPUT_SRC_MASK | \ + NV40_VP_INST_SRC0H_MASK \ + ) + +/* ---- OPCODE BITS 63:32 / data DWORD 2 --- */ +#define NV40_VP_INST_SRC0L_SHIFT 23 +#define NV40_VP_INST_SRC0L_MASK (0x1FF << 23) +#define NV40_VP_INST_SRC1_SHIFT 6 +#define NV40_VP_INST_SRC1_MASK (0x1FFFF << 6) +#define NV40_VP_INST_SRC2H_SHIFT 0 +#define NV40_VP_INST_SRC2H_MASK (0x3F << 0) +#define NV40_VP_INST_IADDRH_SHIFT 0 +#define NV40_VP_INST_IADDRH_MASK (0x1F << 0) + +/* ---- OPCODE BITS 31:0 / data DWORD 3 --- */ +#define NV40_VP_INST_IADDRL_SHIFT 29 +#define NV40_VP_INST_IADDRL_MASK (7 << 29) +#define NV40_VP_INST_SRC2L_SHIFT 21 +#define NV40_VP_INST_SRC2L_MASK (0x7FF << 21) +#define NV40_VP_INST_SCA_WRITEMASK_SHIFT 17 +#define NV40_VP_INST_SCA_WRITEMASK_MASK (0xF << 17) +# define NV40_VP_INST_SCA_WRITEMASK_X (1 << 20) +# define NV40_VP_INST_SCA_WRITEMASK_Y (1 << 19) +# define NV40_VP_INST_SCA_WRITEMASK_Z (1 << 18) +# define NV40_VP_INST_SCA_WRITEMASK_W (1 << 17) +#define NV40_VP_INST_VEC_WRITEMASK_SHIFT 13 +#define NV40_VP_INST_VEC_WRITEMASK_MASK (0xF << 13) +# define NV40_VP_INST_VEC_WRITEMASK_X (1 << 16) +# define NV40_VP_INST_VEC_WRITEMASK_Y (1 << 15) +# define NV40_VP_INST_VEC_WRITEMASK_Z (1 << 14) +# define NV40_VP_INST_VEC_WRITEMASK_W (1 << 13) +#define NV40_VP_INST_SCA_RESULT (1 << 12) +#define NV40_VP_INST_SCA_DEST_TEMP_SHIFT 7 +#define NV40_VP_INST_SCA_DEST_TEMP_MASK (0x1F << 7) +#define NV40_VP_INST_DEST_SHIFT 2 +#define NV40_VP_INST_DEST_MASK (31 << 2) +# define NV40_VP_INST_DEST_POS 0 +# define NV40_VP_INST_DEST_COL0 1 +# define NV40_VP_INST_DEST_COL1 2 +# define NV40_VP_INST_DEST_BFC0 3 +# define NV40_VP_INST_DEST_BFC1 4 +# define NV40_VP_INST_DEST_FOGC 5 +# define NV40_VP_INST_DEST_PSZ 6 +# define NV40_VP_INST_DEST_TC0 7 +# define NV40_VP_INST_DEST_TC(n) (7+n) +# define NV40_VP_INST_DEST_TEMP 0x1F +#define NV40_VP_INST_INDEX_CONST (1 << 1) +#define NV40_VP_INST_LAST (1 << 0) +#define NV40_VP_INST3_KNOWN ( \ + NV40_VP_INST_SRC2L_MASK |\ + NV40_VP_INST_SCA_WRITEMASK_MASK |\ + NV40_VP_INST_VEC_WRITEMASK_MASK |\ + NV40_VP_INST_SCA_DEST_TEMP_MASK |\ + NV40_VP_INST_DEST_MASK |\ + NV40_VP_INST_INDEX_CONST) + +/* Useful to split the source selection regs into their pieces */ +#define NV40_VP_SRC0_HIGH_SHIFT 9 +#define NV40_VP_SRC0_HIGH_MASK 0x0001FE00 +#define NV40_VP_SRC0_LOW_MASK 0x000001FF +#define NV40_VP_SRC2_HIGH_SHIFT 11 +#define NV40_VP_SRC2_HIGH_MASK 0x0001F800 +#define NV40_VP_SRC2_LOW_MASK 0x000007FF + +/* Source selection - these are the bits you fill NV40_VP_INST_SRCn with */ +#define NV40_VP_SRC_NEGATE (1 << 16) +#define NV40_VP_SRC_SWZ_X_SHIFT 14 +#define NV40_VP_SRC_SWZ_X_MASK (3 << 14) +#define NV40_VP_SRC_SWZ_Y_SHIFT 12 +#define NV40_VP_SRC_SWZ_Y_MASK (3 << 12) +#define NV40_VP_SRC_SWZ_Z_SHIFT 10 +#define NV40_VP_SRC_SWZ_Z_MASK (3 << 10) +#define NV40_VP_SRC_SWZ_W_SHIFT 8 +#define NV40_VP_SRC_SWZ_W_MASK (3 << 8) +#define NV40_VP_SRC_SWZ_ALL_SHIFT 8 +#define NV40_VP_SRC_SWZ_ALL_MASK (0xFF << 8) +#define NV40_VP_SRC_TEMP_SRC_SHIFT 2 +#define NV40_VP_SRC_TEMP_SRC_MASK (0x1F << 2) +#define NV40_VP_SRC_REG_TYPE_SHIFT 0 +#define NV40_VP_SRC_REG_TYPE_MASK (3 << 0) +# define NV40_VP_SRC_REG_TYPE_UNK0 0 +# define NV40_VP_SRC_REG_TYPE_TEMP 1 +# define NV40_VP_SRC_REG_TYPE_INPUT 2 +# define NV40_VP_SRC_REG_TYPE_CONST 3 + + +/* + * Each fragment program opcode appears to be comprised of 4 32-bit values. + * + * 0 - Opcode, output reg/mask, ATTRIB source + * 1 - Source 0 + * 2 - Source 1 + * 3 - Source 2 + * + * There appears to be no special difference between result regs and temp regs. + * result.color == R0.xyzw + * result.depth == R1.z + * When the fragprog contains instructions to write depth, + * NV30_TCL_PRIMITIVE_3D_UNK1D78=0 otherwise it is set to 1. + * + * Constants are inserted directly after the instruction that uses them. + * + * It appears that it's not possible to use two input registers in one + * instruction as the input sourcing is done in the instruction dword + * and not the source selection dwords. As such instructions such as: + * + * ADD result.color, fragment.color, fragment.texcoord[0]; + * + * must be split into two MOV's and then an ADD (nvidia does this) but + * I'm not sure why it's not just one MOV and then source the second input + * in the ADD instruction.. + * + * Negation of the full source is done with NV30_FP_REG_NEGATE, arbitrary + * negation requires multiplication with a const. + * + * Arbitrary swizzling is supported with the exception of SWIZZLE_ZERO and + * SWIZZLE_ONE. + * + * The temp/result regs appear to be initialised to (0.0, 0.0, 0.0, 0.0) as + * SWIZZLE_ZERO is implemented simply by not writing to the relevant components + * of the destination. + * + * Looping + * Loops appear to be fairly expensive on NV40 at least, the proprietary + * driver goes to a lot of effort to avoid using the native looping + * instructions. If the total number of *executed* instructions between + * REP/ENDREP or LOOP/ENDLOOP is <=500, the driver will unroll the loop. + * The maximum loop count is 255. + * + * Conditional execution + * TODO + * + * Non-native instructions: + * LIT + * LRP - MAD+MAD + * SUB - ADD, negate second source + * RSQ - LG2 + EX2 + * POW - LG2 + MUL + EX2 + * SCS - COS + SIN + * XPD + * DP2 - MUL + ADD + * NRM + */ + +//== Opcode / Destination selection == +#define NV40_FP_OP_PROGRAM_END (1 << 0) +#define NV40_FP_OP_OUT_REG_SHIFT 1 +#define NV40_FP_OP_OUT_REG_MASK (31 << 1) +/* Needs to be set when writing outputs to get expected result.. */ +#define NV40_FP_OP_UNK0_7 (1 << 7) +#define NV40_FP_OP_COND_WRITE_ENABLE (1 << 8) +#define NV40_FP_OP_OUTMASK_SHIFT 9 +#define NV40_FP_OP_OUTMASK_MASK (0xF << 9) +# define NV40_FP_OP_OUT_X (1 << 9) +# define NV40_FP_OP_OUT_Y (1 <<10) +# define NV40_FP_OP_OUT_Z (1 <<11) +# define NV40_FP_OP_OUT_W (1 <<12) +/* Uncertain about these, especially the input_src values.. it's possible that + * they can be dynamically changed. + */ +#define NV40_FP_OP_INPUT_SRC_SHIFT 13 +#define NV40_FP_OP_INPUT_SRC_MASK (15 << 13) +# define NV40_FP_OP_INPUT_SRC_POSITION 0x0 +# define NV40_FP_OP_INPUT_SRC_COL0 0x1 +# define NV40_FP_OP_INPUT_SRC_COL1 0x2 +# define NV40_FP_OP_INPUT_SRC_FOGC 0x3 +# define NV40_FP_OP_INPUT_SRC_TC0 0x4 +# define NV40_FP_OP_INPUT_SRC_TC(n) (0x4 + n) +# define NV40_FP_OP_INPUT_SRC_FACING 0xE +#define NV40_FP_OP_TEX_UNIT_SHIFT 17 +#define NV40_FP_OP_TEX_UNIT_MASK (0xF << 17) +#define NV40_FP_OP_PRECISION_SHIFT 22 +#define NV40_FP_OP_PRECISION_MASK (3 << 22) +# define NV40_FP_PRECISION_FP32 0 +# define NV40_FP_PRECISION_FP16 1 +# define NV40_FP_PRECISION_FX12 2 +#define NV40_FP_OP_OPCODE_SHIFT 24 +#define NV40_FP_OP_OPCODE_MASK (0x3F << 24) +# define NV40_FP_OP_OPCODE_NOP 0x00 +# define NV40_FP_OP_OPCODE_MOV 0x01 +# define NV40_FP_OP_OPCODE_MUL 0x02 +# define NV40_FP_OP_OPCODE_ADD 0x03 +# define NV40_FP_OP_OPCODE_MAD 0x04 +# define NV40_FP_OP_OPCODE_DP3 0x05 +# define NV40_FP_OP_OPCODE_DP4 0x06 +# define NV40_FP_OP_OPCODE_DST 0x07 +# define NV40_FP_OP_OPCODE_MIN 0x08 +# define NV40_FP_OP_OPCODE_MAX 0x09 +# define NV40_FP_OP_OPCODE_SLT 0x0A +# define NV40_FP_OP_OPCODE_SGE 0x0B +# define NV40_FP_OP_OPCODE_SLE 0x0C +# define NV40_FP_OP_OPCODE_SGT 0x0D +# define NV40_FP_OP_OPCODE_SNE 0x0E +# define NV40_FP_OP_OPCODE_SEQ 0x0F +# define NV40_FP_OP_OPCODE_FRC 0x10 +# define NV40_FP_OP_OPCODE_FLR 0x11 +# define NV40_FP_OP_OPCODE_KIL 0x12 +# define NV40_FP_OP_OPCODE_PK4B 0x13 +# define NV40_FP_OP_OPCODE_UP4B 0x14 +/* DDX/DDY can only write to XY */ +# define NV40_FP_OP_OPCODE_DDX 0x15 +# define NV40_FP_OP_OPCODE_DDY 0x16 +# define NV40_FP_OP_OPCODE_TEX 0x17 +# define NV40_FP_OP_OPCODE_TXP 0x18 +# define NV40_FP_OP_OPCODE_TXD 0x19 +# define NV40_FP_OP_OPCODE_RCP 0x1A +# define NV40_FP_OP_OPCODE_EX2 0x1C +# define NV40_FP_OP_OPCODE_LG2 0x1D +# define NV40_FP_OP_OPCODE_COS 0x22 +# define NV40_FP_OP_OPCODE_SIN 0x23 +# define NV40_FP_OP_OPCODE_PK2H 0x24 +# define NV40_FP_OP_OPCODE_UP2H 0x25 +# define NV40_FP_OP_OPCODE_PK4UB 0x27 +# define NV40_FP_OP_OPCODE_UP4UB 0x28 +# define NV40_FP_OP_OPCODE_PK2US 0x29 +# define NV40_FP_OP_OPCODE_UP2US 0x2A +# define NV40_FP_OP_OPCODE_DP2A 0x2E +# define NV40_FP_OP_OPCODE_TXL 0x2F +# define NV40_FP_OP_OPCODE_TXB 0x31 +# define NV40_FP_OP_OPCODE_DIV 0x3A +/* The use of these instructions appears to be indicated by bit 31 of DWORD 2.*/ +# define NV40_FP_OP_BRA_OPCODE_BRK 0x0 +# define NV40_FP_OP_BRA_OPCODE_CAL 0x1 +# define NV40_FP_OP_BRA_OPCODE_IF 0x2 +# define NV40_FP_OP_BRA_OPCODE_LOOP 0x3 +# define NV40_FP_OP_BRA_OPCODE_REP 0x4 +# define NV40_FP_OP_BRA_OPCODE_RET 0x5 +#define NV40_FP_OP_OUT_SAT (1 << 31) + +/* high order bits of SRC0 */ +#define NV40_FP_OP_OUT_ABS (1 << 29) +#define NV40_FP_OP_COND_SWZ_W_SHIFT 27 +#define NV40_FP_OP_COND_SWZ_W_MASK (3 << 27) +#define NV40_FP_OP_COND_SWZ_Z_SHIFT 25 +#define NV40_FP_OP_COND_SWZ_Z_MASK (3 << 25) +#define NV40_FP_OP_COND_SWZ_Y_SHIFT 23 +#define NV40_FP_OP_COND_SWZ_Y_MASK (3 << 23) +#define NV40_FP_OP_COND_SWZ_X_SHIFT 21 +#define NV40_FP_OP_COND_SWZ_X_MASK (3 << 21) +#define NV40_FP_OP_COND_SWZ_ALL_SHIFT 21 +#define NV40_FP_OP_COND_SWZ_ALL_MASK (0xFF << 21) +#define NV40_FP_OP_COND_SHIFT 18 +#define NV40_FP_OP_COND_MASK (0x07 << 18) +# define NV40_FP_OP_COND_FL 0 +# define NV40_FP_OP_COND_LT 1 +# define NV40_FP_OP_COND_EQ 2 +# define NV40_FP_OP_COND_LE 3 +# define NV40_FP_OP_COND_GT 4 +# define NV40_FP_OP_COND_NE 5 +# define NV40_FP_OP_COND_GE 6 +# define NV40_FP_OP_COND_TR 7 + +/* high order bits of SRC1 */ +#define NV40_FP_OP_OPCODE_IS_BRANCH (1<<31) +#define NV40_FP_OP_SRC_SCALE_SHIFT 28 +#define NV40_FP_OP_SRC_SCALE_MASK (3 << 28) + +/* SRC1 LOOP */ +#define NV40_FP_OP_LOOP_INCR_SHIFT 19 +#define NV40_FP_OP_LOOP_INCR_MASK (0xFF << 19) +#define NV40_FP_OP_LOOP_INDEX_SHIFT 10 +#define NV40_FP_OP_LOOP_INDEX_MASK (0xFF << 10) +#define NV40_FP_OP_LOOP_COUNT_SHIFT 2 +#define NV40_FP_OP_LOOP_COUNT_MASK (0xFF << 2) + +/* SRC1 IF */ +#define NV40_FP_OP_ELSE_ID_SHIFT 2 +#define NV40_FP_OP_ELSE_ID_MASK (0xFF << 2) + +/* SRC1 CAL */ +#define NV40_FP_OP_IADDR_SHIFT 2 +#define NV40_FP_OP_IADDR_MASK (0xFF << 2) + +/* SRC1 REP + * I have no idea why there are 3 count values here.. but they + * have always been filled with the same value in my tests so + * far.. + */ +#define NV40_FP_OP_REP_COUNT1_SHIFT 2 +#define NV40_FP_OP_REP_COUNT1_MASK (0xFF << 2) +#define NV40_FP_OP_REP_COUNT2_SHIFT 10 +#define NV40_FP_OP_REP_COUNT2_MASK (0xFF << 10) +#define NV40_FP_OP_REP_COUNT3_SHIFT 19 +#define NV40_FP_OP_REP_COUNT3_MASK (0xFF << 19) + +/* SRC2 REP/IF */ +#define NV40_FP_OP_END_ID_SHIFT 2 +#define NV40_FP_OP_END_ID_MASK (0xFF << 2) + +// SRC2 high-order +#define NV40_FP_OP_INDEX_INPUT (1 << 30) +#define NV40_FP_OP_ADDR_INDEX_SHIFT 19 +#define NV40_FP_OP_ADDR_INDEX_MASK (0xF << 19) + +//== Register selection == +#define NV40_FP_REG_TYPE_SHIFT 0 +#define NV40_FP_REG_TYPE_MASK (3 << 0) +# define NV40_FP_REG_TYPE_TEMP 0 +# define NV40_FP_REG_TYPE_INPUT 1 +# define NV40_FP_REG_TYPE_CONST 2 +#define NV40_FP_REG_SRC_SHIFT 2 +#define NV40_FP_REG_SRC_MASK (31 << 2) +#define NV40_FP_REG_UNK_0 (1 << 8) +#define NV40_FP_REG_SWZ_ALL_SHIFT 9 +#define NV40_FP_REG_SWZ_ALL_MASK (255 << 9) +#define NV40_FP_REG_SWZ_X_SHIFT 9 +#define NV40_FP_REG_SWZ_X_MASK (3 << 9) +#define NV40_FP_REG_SWZ_Y_SHIFT 11 +#define NV40_FP_REG_SWZ_Y_MASK (3 << 11) +#define NV40_FP_REG_SWZ_Z_SHIFT 13 +#define NV40_FP_REG_SWZ_Z_MASK (3 << 13) +#define NV40_FP_REG_SWZ_W_SHIFT 15 +#define NV40_FP_REG_SWZ_W_MASK (3 << 15) +# define NV40_FP_SWIZZLE_X 0 +# define NV40_FP_SWIZZLE_Y 1 +# define NV40_FP_SWIZZLE_Z 2 +# define NV40_FP_SWIZZLE_W 3 +#define NV40_FP_REG_NEGATE (1 << 17) + +#endif diff --git a/src/mesa/drivers/dri/nouveau/nv40_vertprog.c b/src/mesa/drivers/dri/nouveau/nv40_vertprog.c new file mode 100644 index 00000000000..111c6de71b0 --- /dev/null +++ b/src/mesa/drivers/dri/nouveau/nv40_vertprog.c @@ -0,0 +1,647 @@ +#include "nouveau_shader.h" +#include "nouveau_msg.h" +#include "nv40_shader.h" + +extern nvsSwzComp NV20VP_TX_SWIZZLE[4]; +extern void NV20VPTXSwizzle(int hwswz, nvsSwzComp *swz); + +/***************************************************************************** + * Assembly routines + */ +static int +NV40VPSupportsOpcode(nvsFunc * shader, nvsOpcode op) +{ + if (shader->GetOPTXFromSOP(op, NULL)) + return 1; + return 0; +} + +static void +NV40VPSetOpcode(nvsFunc *shader, unsigned int opcode, int slot) +{ + if (slot) shader->inst[1] |= (opcode << NV40_VP_INST_SCA_OPCODE_SHIFT); + else shader->inst[1] |= (opcode << NV40_VP_INST_VEC_OPCODE_SHIFT); +} + +static void +NV40VPSetCCUpdate(nvsFunc *shader) +{ + shader->inst[0] |= NV40_VP_INST_COND_UPDATE_ENABLE; +} + +static void +NV40VPSetCondition(nvsFunc *shader, int on, nvsCond cond, int reg, + nvsSwzComp *swizzle) +{ + unsigned int hwcond; + + if (on ) shader->inst[0] |= NV40_VP_INST_COND_TEST_ENABLE; + if (reg) shader->inst[0] |= NV40_VP_INST_COND_REG_SELECT_1; + + switch (cond) { + case NVS_COND_TR: hwcond = NV40_VP_INST_COND_TR; break; + case NVS_COND_FL: hwcond = NV40_VP_INST_COND_FL; break; + case NVS_COND_LT: hwcond = NV40_VP_INST_COND_LT; break; + case NVS_COND_GT: hwcond = NV40_VP_INST_COND_GT; break; + case NVS_COND_NE: hwcond = NV40_VP_INST_COND_NE; break; + case NVS_COND_EQ: hwcond = NV40_VP_INST_COND_EQ; break; + case NVS_COND_GE: hwcond = NV40_VP_INST_COND_GE; break; + case NVS_COND_LE: hwcond = NV40_VP_INST_COND_LE; break; + default: + WARN_ONCE("unknown vp cond %d\n", cond); + hwcond = NV40_VP_INST_COND_TR; + break; + } + shader->inst[0] |= (hwcond << NV40_VP_INST_COND_SHIFT); + + shader->inst[0] |= (swizzle[NVS_SWZ_X] << NV40_VP_INST_COND_SWZ_X_SHIFT); + shader->inst[0] |= (swizzle[NVS_SWZ_Y] << NV40_VP_INST_COND_SWZ_Y_SHIFT); + shader->inst[0] |= (swizzle[NVS_SWZ_Z] << NV40_VP_INST_COND_SWZ_Z_SHIFT); + shader->inst[0] |= (swizzle[NVS_SWZ_W] << NV40_VP_INST_COND_SWZ_W_SHIFT); +} + +static void +NV40VPSetResult(nvsFunc *shader, nvsRegister * dest, unsigned int mask, + int slot) +{ + unsigned int hwmask = 0; + + if (mask & SMASK_X) hwmask |= (1 << 3); + if (mask & SMASK_Y) hwmask |= (1 << 2); + if (mask & SMASK_Z) hwmask |= (1 << 1); + if (mask & SMASK_W) hwmask |= (1 << 0); + + if (dest->file == NVS_FILE_RESULT) { + int hwidx; + + switch (dest->index) { + case NVS_FR_POSITION : hwidx = NV40_VP_INST_DEST_POS; break; + case NVS_FR_COL0 : hwidx = NV40_VP_INST_DEST_COL0; break; + case NVS_FR_COL1 : hwidx = NV40_VP_INST_DEST_COL1; break; + case NVS_FR_BFC0 : hwidx = NV40_VP_INST_DEST_BFC0; break; + case NVS_FR_BFC1 : hwidx = NV40_VP_INST_DEST_BFC1; break; + case NVS_FR_FOGCOORD : hwidx = NV40_VP_INST_DEST_FOGC; break; + case NVS_FR_POINTSZ : hwidx = NV40_VP_INST_DEST_PSZ; break; + case NVS_FR_TEXCOORD0: hwidx = NV40_VP_INST_DEST_TC(0); break; + case NVS_FR_TEXCOORD1: hwidx = NV40_VP_INST_DEST_TC(1); break; + case NVS_FR_TEXCOORD2: hwidx = NV40_VP_INST_DEST_TC(2); break; + case NVS_FR_TEXCOORD3: hwidx = NV40_VP_INST_DEST_TC(3); break; + case NVS_FR_TEXCOORD4: hwidx = NV40_VP_INST_DEST_TC(4); break; + case NVS_FR_TEXCOORD5: hwidx = NV40_VP_INST_DEST_TC(5); break; + case NVS_FR_TEXCOORD6: hwidx = NV40_VP_INST_DEST_TC(6); break; + case NVS_FR_TEXCOORD7: hwidx = NV40_VP_INST_DEST_TC(7); break; + default: + WARN_ONCE("unknown vtxprog output %d\n", dest->index); + hwidx = 0; + break; + } + shader->inst[3] |= (hwidx << NV40_VP_INST_DEST_SHIFT); + + if (slot) { + shader->inst[3] |= NV40_VP_INST_SCA_RESULT; + shader->inst[3] |= NV40_VP_INST_SCA_DEST_TEMP_MASK; + } else { + shader->inst[0] |= NV40_VP_INST_VEC_RESULT; + shader->inst[0] |= NV40_VP_INST_VEC_DEST_TEMP_MASK | (1<<20); + } + } else { + /* NVS_FILE_TEMP || NVS_FILE_ADDRESS */ + if (slot) + shader->inst[3] |= (dest->index << NV40_VP_INST_SCA_DEST_TEMP_SHIFT); + else + shader->inst[0] |= (dest->index << NV40_VP_INST_VEC_DEST_TEMP_SHIFT); + } + + if (slot) shader->inst[3] |= (hwmask << NV40_VP_INST_SCA_WRITEMASK_SHIFT); + else shader->inst[3] |= (hwmask << NV40_VP_INST_VEC_WRITEMASK_SHIFT); +} + +static void +NV40VPInsertSource(nvsFunc *shader, unsigned int hw, int pos) +{ + switch (pos) { + case 0: + shader->inst[1] |= ((hw & NV40_VP_SRC0_HIGH_MASK) >> + NV40_VP_SRC0_HIGH_SHIFT) + << NV40_VP_INST_SRC0H_SHIFT; + shader->inst[2] |= (hw & NV40_VP_SRC0_LOW_MASK) + << NV40_VP_INST_SRC0L_SHIFT; + break; + case 1: + shader->inst[2] |= hw + << NV40_VP_INST_SRC1_SHIFT; + break; + case 2: + shader->inst[2] |= ((hw & NV40_VP_SRC2_HIGH_MASK) >> + NV40_VP_SRC2_HIGH_SHIFT) + << NV40_VP_INST_SRC2H_SHIFT; + shader->inst[3] |= (hw & NV40_VP_SRC2_LOW_MASK) + << NV40_VP_INST_SRC2L_SHIFT; + break; + default: + assert(0); + break; + } +} + +static void +NV40VPSetSource(nvsFunc *shader, nvsRegister * src, int pos) +{ + unsigned int hw = 0; + + switch (src->file) { + case NVS_FILE_ADDRESS: + break; + case NVS_FILE_ATTRIB: + hw |= (NV40_VP_SRC_REG_TYPE_INPUT << NV40_VP_SRC_REG_TYPE_SHIFT); + + shader->inst[1] |= (src->index << NV40_VP_INST_INPUT_SRC_SHIFT); + if (src->indexed) { + shader->inst[0] |= NV40_VP_INST_INDEX_INPUT; + if (src->addr_reg) + shader->inst[0] |= NV40_VP_INST_ADDR_REG_SELECT_1; + shader->inst[0] |= (src->addr_comp << NV40_VP_INST_ADDR_SWZ_SHIFT); + } + break; + case NVS_FILE_CONST: + hw |= (NV40_VP_SRC_REG_TYPE_CONST << NV40_VP_SRC_REG_TYPE_SHIFT); + + shader->inst[1] |= (src->index << NV40_VP_INST_CONST_SRC_SHIFT); + if (src->indexed) { + shader->inst[3] |= NV40_VP_INST_INDEX_CONST; + if (src->addr_reg) + shader->inst[0] |= NV40_VP_INST_ADDR_REG_SELECT_1; + shader->inst[0] |= (src->addr_comp << NV40_VP_INST_ADDR_SWZ_SHIFT); + } + break; + case NVS_FILE_TEMP: + hw |= (NV40_VP_SRC_REG_TYPE_TEMP << NV40_VP_SRC_REG_TYPE_SHIFT); + hw |= (src->index << NV40_VP_SRC_TEMP_SRC_SHIFT); + break; + default: + fprintf(stderr, "unknown source file %d\n", src->file); + assert(0); + break; + } + + if (src->file != NVS_FILE_ADDRESS) { + if (src->negate) + hw |= NV40_VP_SRC_NEGATE; + if (src->abs) + shader->inst[0] |= (1 << (21 + pos)); + hw |= (src->swizzle[0] << NV40_VP_SRC_SWZ_X_SHIFT); + hw |= (src->swizzle[1] << NV40_VP_SRC_SWZ_Y_SHIFT); + hw |= (src->swizzle[2] << NV40_VP_SRC_SWZ_Z_SHIFT); + hw |= (src->swizzle[3] << NV40_VP_SRC_SWZ_W_SHIFT); + + NV40VPInsertSource(shader, hw, pos); + } +} + +static void +NV40VPSetUnusedSource(nvsFunc *shader, int pos) +{ + unsigned int hw; + + hw = ((NV40_VP_SRC_REG_TYPE_INPUT << NV40_VP_SRC_REG_TYPE_SHIFT) | + (NVS_SWZ_X << NV40_VP_SRC_SWZ_X_SHIFT) | + (NVS_SWZ_Y << NV40_VP_SRC_SWZ_Y_SHIFT) | + (NVS_SWZ_Z << NV40_VP_SRC_SWZ_Z_SHIFT) | + (NVS_SWZ_W << NV40_VP_SRC_SWZ_W_SHIFT)); + + NV40VPInsertSource(shader, hw, pos); +} + +static void +NV40VPSetLastInst(nvsFunc *shader, int pos) +{ + shader->inst[3] |= 1; +} + +/***************************************************************************** + * Disassembly routines + */ +static int +NV40VPHasMergedInst(nvsFunc * shader) +{ + if (shader->GetOpcodeHW(shader, 0) != NV40_VP_INST_OP_NOP && + shader->GetOpcodeHW(shader, 1) != NV40_VP_INST_OP_NOP) + return 1; + return 0; +} + +static unsigned int +NV40VPGetOpcodeHW(nvsFunc * shader, int slot) +{ + int op; + + if (slot) + op = (shader->inst[1] & NV40_VP_INST_SCA_OPCODE_MASK) + >> NV40_VP_INST_SCA_OPCODE_SHIFT; + else + op = (shader->inst[1] & NV40_VP_INST_VEC_OPCODE_MASK) + >> NV40_VP_INST_VEC_OPCODE_SHIFT; + + return op; +} + +static nvsRegFile +NV40VPGetDestFile(nvsFunc * shader, int merged) +{ + nvsOpcode op; + + op = shader->GetOpcode(shader, merged); + switch (op) { + case NVS_OP_ARL: + case NVS_OP_ARR: + case NVS_OP_ARA: + case NVS_OP_POPA: + return NVS_FILE_ADDRESS; + default: + if (shader->GetOpcodeSlot(shader, merged)) { + if (shader->inst[3] & NV40_VP_INST_SCA_RESULT) + return NVS_FILE_RESULT; + } + else { + if (shader->inst[0] & NV40_VP_INST_VEC_RESULT) + return NVS_FILE_RESULT; + } + return NVS_FILE_TEMP; + } + +} + +static unsigned int +NV40VPGetDestID(nvsFunc * shader, int merged) +{ + int id; + + switch (shader->GetDestFile(shader, merged)) { + case NVS_FILE_RESULT: + id = ((shader->inst[3] & NV40_VP_INST_DEST_MASK) + >> NV40_VP_INST_DEST_SHIFT); + switch (id) { + case NV40_VP_INST_DEST_POS : return NVS_FR_POSITION; + case NV40_VP_INST_DEST_COL0: return NVS_FR_COL0; + case NV40_VP_INST_DEST_COL1: return NVS_FR_COL1; + case NV40_VP_INST_DEST_BFC0: return NVS_FR_BFC0; + case NV40_VP_INST_DEST_BFC1: return NVS_FR_BFC1; + case NV40_VP_INST_DEST_FOGC: { + int mask = shader->GetDestMask(shader, merged); + switch (mask) { + case SMASK_X: return NVS_FR_FOGCOORD; + case SMASK_Y: return NVS_FR_CLIP0; + case SMASK_Z: return NVS_FR_CLIP1; + case SMASK_W: return NVS_FR_CLIP2; + default: + printf("more than 1 mask component set in FOGC writemask!\n"); + return NVS_FR_UNKNOWN; + } + } + case NV40_VP_INST_DEST_PSZ: + { + int mask = shader->GetDestMask(shader, merged); + switch (mask) { + case SMASK_X: return NVS_FR_POINTSZ; + case SMASK_Y: return NVS_FR_CLIP3; + case SMASK_Z: return NVS_FR_CLIP4; + case SMASK_W: return NVS_FR_CLIP5; + default: + printf("more than 1 mask component set in PSZ writemask!\n"); + return NVS_FR_UNKNOWN; + } + } + case NV40_VP_INST_DEST_TC(0): return NVS_FR_TEXCOORD0; + case NV40_VP_INST_DEST_TC(1): return NVS_FR_TEXCOORD1; + case NV40_VP_INST_DEST_TC(2): return NVS_FR_TEXCOORD2; + case NV40_VP_INST_DEST_TC(3): return NVS_FR_TEXCOORD3; + case NV40_VP_INST_DEST_TC(4): return NVS_FR_TEXCOORD4; + case NV40_VP_INST_DEST_TC(5): return NVS_FR_TEXCOORD5; + case NV40_VP_INST_DEST_TC(6): return NVS_FR_TEXCOORD6; + case NV40_VP_INST_DEST_TC(7): return NVS_FR_TEXCOORD7; + default: + return -1; + } + case NVS_FILE_ADDRESS: + /* Instructions that write address regs are encoded as if + * they would write temps. + */ + case NVS_FILE_TEMP: + if (shader->GetOpcodeSlot(shader, merged)) + id = ((shader->inst[3] & NV40_VP_INST_SCA_DEST_TEMP_MASK) + >> NV40_VP_INST_SCA_DEST_TEMP_SHIFT); + else + id = ((shader->inst[0] & NV40_VP_INST_VEC_DEST_TEMP_MASK) + >> NV40_VP_INST_VEC_DEST_TEMP_SHIFT); + return id; + default: + return -1; + } +} + +static unsigned int +NV40VPGetDestMask(nvsFunc * shader, int merged) +{ + unsigned int mask = 0; + + if (shader->GetOpcodeSlot(shader, merged)) { + if (shader->inst[3] & NV40_VP_INST_SCA_WRITEMASK_X) mask |= SMASK_X; + if (shader->inst[3] & NV40_VP_INST_SCA_WRITEMASK_Y) mask |= SMASK_Y; + if (shader->inst[3] & NV40_VP_INST_SCA_WRITEMASK_Z) mask |= SMASK_Z; + if (shader->inst[3] & NV40_VP_INST_SCA_WRITEMASK_W) mask |= SMASK_W; + } else { + if (shader->inst[3] & NV40_VP_INST_VEC_WRITEMASK_X) mask |= SMASK_X; + if (shader->inst[3] & NV40_VP_INST_VEC_WRITEMASK_Y) mask |= SMASK_Y; + if (shader->inst[3] & NV40_VP_INST_VEC_WRITEMASK_Z) mask |= SMASK_Z; + if (shader->inst[3] & NV40_VP_INST_VEC_WRITEMASK_W) mask |= SMASK_W; + } + + return mask; +} + +static unsigned int +NV40VPGetSourceHW(nvsFunc * shader, int merged, int pos) +{ + struct _op_xlat *opr; + unsigned int src; + + opr = shader->GetOPTXRec(shader, merged); + if (!opr) + return -1; + + switch (opr->srcpos[pos]) { + case 0: + src = ((shader->inst[1] & NV40_VP_INST_SRC0H_MASK) + >> NV40_VP_INST_SRC0H_SHIFT) + << NV40_VP_SRC0_HIGH_SHIFT; + src |= ((shader->inst[2] & NV40_VP_INST_SRC0L_MASK) + >> NV40_VP_INST_SRC0L_SHIFT); + break; + case 1: + src = ((shader->inst[2] & NV40_VP_INST_SRC1_MASK) + >> NV40_VP_INST_SRC1_SHIFT); + break; + case 2: + src = ((shader->inst[2] & NV40_VP_INST_SRC2H_MASK) + >> NV40_VP_INST_SRC2H_SHIFT) + << NV40_VP_SRC2_HIGH_SHIFT; + src |= ((shader->inst[3] & NV40_VP_INST_SRC2L_MASK) + >> NV40_VP_INST_SRC2L_SHIFT); + break; + default: + src = -1; + } + + return src; +} + +static nvsRegFile +NV40VPGetSourceFile(nvsFunc * shader, int merged, int pos) +{ + unsigned int src; + struct _op_xlat *opr; + int file; + + opr = shader->GetOPTXRec(shader, merged); + if (!opr || opr->srcpos[pos] == -1) + return -1; + + switch (opr->srcpos[pos]) { + case SPOS_ADDRESS: return NVS_FILE_ADDRESS; + default: + src = shader->GetSourceHW(shader, merged, pos); + file = (src & NV40_VP_SRC_REG_TYPE_MASK) >> NV40_VP_SRC_REG_TYPE_SHIFT; + + switch (file) { + case NV40_VP_SRC_REG_TYPE_TEMP : return NVS_FILE_TEMP; + case NV40_VP_SRC_REG_TYPE_INPUT: return NVS_FILE_ATTRIB; + case NV40_VP_SRC_REG_TYPE_CONST: return NVS_FILE_CONST; + default: + return NVS_FILE_UNKNOWN; + } + } +} + +static int +NV40VPGetSourceID(nvsFunc * shader, int merged, int pos) +{ + switch (shader->GetSourceFile(shader, merged, pos)) { + case NVS_FILE_ATTRIB: + switch ((shader->inst[1] & NV40_VP_INST_INPUT_SRC_MASK) + >> NV40_VP_INST_INPUT_SRC_SHIFT) { + case NV40_VP_INST_IN_POS: return NVS_FR_POSITION; + case NV40_VP_INST_IN_WEIGHT: return NVS_FR_WEIGHT; + case NV40_VP_INST_IN_NORMAL: return NVS_FR_NORMAL; + case NV40_VP_INST_IN_COL0: return NVS_FR_COL0; + case NV40_VP_INST_IN_COL1: return NVS_FR_COL1; + case NV40_VP_INST_IN_FOGC: return NVS_FR_FOGCOORD; + case NV40_VP_INST_IN_TC(0): return NVS_FR_TEXCOORD0; + case NV40_VP_INST_IN_TC(1): return NVS_FR_TEXCOORD1; + case NV40_VP_INST_IN_TC(2): return NVS_FR_TEXCOORD2; + case NV40_VP_INST_IN_TC(3): return NVS_FR_TEXCOORD3; + case NV40_VP_INST_IN_TC(4): return NVS_FR_TEXCOORD4; + case NV40_VP_INST_IN_TC(5): return NVS_FR_TEXCOORD5; + case NV40_VP_INST_IN_TC(6): return NVS_FR_TEXCOORD6; + case NV40_VP_INST_IN_TC(7): return NVS_FR_TEXCOORD7; + default: + return -1; + } + break; + case NVS_FILE_CONST: + return ((shader->inst[1] & NV40_VP_INST_CONST_SRC_MASK) + >> NV40_VP_INST_CONST_SRC_SHIFT); + case NVS_FILE_TEMP: + { + unsigned int src; + + src = shader->GetSourceHW(shader, merged, pos); + return ((src & NV40_VP_SRC_TEMP_SRC_MASK) >> + NV40_VP_SRC_TEMP_SRC_SHIFT); + } + default: + return -1; + } +} + +static int +NV40VPGetSourceNegate(nvsFunc * shader, int merged, int pos) +{ + unsigned int src; + + src = shader->GetSourceHW(shader, merged, pos); + + if (src == -1) + return -1; + return ((src & NV40_VP_SRC_NEGATE) ? 1 : 0); +} + +static void +NV40VPGetSourceSwizzle(nvsFunc * shader, int merged, int pos, nvsSwzComp *swz) +{ + unsigned int src; + int swzbits; + + src = shader->GetSourceHW(shader, merged, pos); + swzbits = (src & NV40_VP_SRC_SWZ_ALL_MASK) >> NV40_VP_SRC_SWZ_ALL_SHIFT; + NV20VPTXSwizzle(swzbits, swz); +} + +static int +NV40VPGetSourceIndexed(nvsFunc * shader, int merged, int pos) +{ + switch (shader->GetSourceFile(shader, merged, pos)) { + case NVS_FILE_ATTRIB: + return ((shader->inst[0] & NV40_VP_INST_INDEX_INPUT) ? 1 : 0); + case NVS_FILE_CONST: + return ((shader->inst[3] & NV40_VP_INST_INDEX_CONST) ? 1 : 0); + default: + return 0; + } +} + +static nvsSwzComp +NV40VPGetAddressRegSwizzle(nvsFunc * shader) +{ + nvsSwzComp swz; + + swz = NV20VP_TX_SWIZZLE[(shader->inst[0] & NV40_VP_INST_ADDR_SWZ_MASK) + >> NV40_VP_INST_ADDR_SWZ_SHIFT]; + return swz; +} + +static int +NV40VPSupportsConditional(nvsFunc * shader) +{ + /*FIXME: Is this true of all ops? */ + return 1; +} + +static int +NV40VPGetConditionUpdate(nvsFunc * shader) +{ + return ((shader->inst[0] & NV40_VP_INST_COND_UPDATE_ENABLE) ? 1 : 0); +} + +static int +NV40VPGetConditionTest(nvsFunc * shader) +{ + int op; + + /* The condition test is unconditionally enabled on some + * instructions. ie: the condition test bit does *NOT* have + * to be set. + * + * FIXME: check other relevant ops for this situation. + */ + op = shader->GetOpcodeHW(shader, 1); + switch (op) { + case NV40_VP_INST_OP_BRA: + return 1; + default: + return ((shader->inst[0] & NV40_VP_INST_COND_TEST_ENABLE) ? 1 : 0); + } +} + +static nvsCond +NV40VPGetCondition(nvsFunc * shader) +{ + int cond; + + cond = ((shader->inst[0] & NV40_VP_INST_COND_MASK) + >> NV40_VP_INST_COND_SHIFT); + + switch (cond) { + case NV40_VP_INST_COND_FL: return NVS_COND_FL; + case NV40_VP_INST_COND_LT: return NVS_COND_LT; + case NV40_VP_INST_COND_EQ: return NVS_COND_EQ; + case NV40_VP_INST_COND_LE: return NVS_COND_LE; + case NV40_VP_INST_COND_GT: return NVS_COND_GT; + case NV40_VP_INST_COND_NE: return NVS_COND_NE; + case NV40_VP_INST_COND_GE: return NVS_COND_GE; + case NV40_VP_INST_COND_TR: return NVS_COND_TR; + default: + return NVS_COND_UNKNOWN; + } +} + +static void +NV40VPGetCondRegSwizzle(nvsFunc * shader, nvsSwzComp *swz) +{ + int swzbits; + + swzbits = (shader->inst[0] & NV40_VP_INST_COND_SWZ_ALL_MASK) + >> NV40_VP_INST_COND_SWZ_ALL_SHIFT; + NV20VPTXSwizzle(swzbits, swz); +} + +static int +NV40VPGetCondRegID(nvsFunc * shader) +{ + return ((shader->inst[0] & NV40_VP_INST_COND_REG_SELECT_1) ? 1 : 0); +} + +static int +NV40VPGetBranch(nvsFunc * shader) +{ + int addr; + + addr = ((shader->inst[2] & NV40_VP_INST_IADDRH_MASK) + >> NV40_VP_INST_IADDRH_SHIFT) << 3; + addr |= ((shader->inst[3] & NV40_VP_INST_IADDRL_MASK) + >> NV40_VP_INST_IADDRL_SHIFT); + return addr; +} + +void +NV40VPInitShaderFuncs(nvsFunc * shader) +{ + /* Inherit NV30 VP code, we share some of it */ + NV30VPInitShaderFuncs(shader); + + /* Limits */ + shader->MaxInst = 4096; + shader->MaxAttrib = 16; + shader->MaxTemp = 32; + shader->MaxAddress = 2; + shader->MaxConst = 256; + shader->caps = SCAP_SRC_ABS; + + /* Add extra opcodes for NV40+ */ +// MOD_OPCODE(NVVP_TX_VOP, NV40_VP_INST_OP_TXWHAT, NVS_OP_TEX , 0, 4, -1); + MOD_OPCODE(NVVP_TX_SOP, NV40_VP_INST_OP_PUSHA, NVS_OP_PUSHA, 3, -1, -1); + MOD_OPCODE(NVVP_TX_SOP, NV40_VP_INST_OP_POPA , NVS_OP_POPA , -1, -1, -1); + + shader->SupportsOpcode = NV40VPSupportsOpcode; + shader->SetOpcode = NV40VPSetOpcode; + shader->SetCCUpdate = NV40VPSetCCUpdate; + shader->SetCondition = NV40VPSetCondition; + shader->SetResult = NV40VPSetResult; + shader->SetSource = NV40VPSetSource; + shader->SetUnusedSource = NV40VPSetUnusedSource; + shader->SetLastInst = NV40VPSetLastInst; + + shader->HasMergedInst = NV40VPHasMergedInst; + shader->GetOpcodeHW = NV40VPGetOpcodeHW; + + shader->GetDestFile = NV40VPGetDestFile; + shader->GetDestID = NV40VPGetDestID; + shader->GetDestMask = NV40VPGetDestMask; + + shader->GetSourceHW = NV40VPGetSourceHW; + shader->GetSourceFile = NV40VPGetSourceFile; + shader->GetSourceID = NV40VPGetSourceID; + shader->GetSourceNegate = NV40VPGetSourceNegate; + shader->GetSourceSwizzle = NV40VPGetSourceSwizzle; + shader->GetSourceIndexed = NV40VPGetSourceIndexed; + + shader->GetRelAddressSwizzle = NV40VPGetAddressRegSwizzle; + + shader->SupportsConditional = NV40VPSupportsConditional; + shader->GetConditionUpdate = NV40VPGetConditionUpdate; + shader->GetConditionTest = NV40VPGetConditionTest; + shader->GetCondition = NV40VPGetCondition; + shader->GetCondRegSwizzle = NV40VPGetCondRegSwizzle; + shader->GetCondRegID = NV40VPGetCondRegID; + + shader->GetBranch = NV40VPGetBranch; +}