r300_state.c

mesa-6.5-minigui源码

		fprintf(stderr,"VERTEX_SHADER_FRAGMENT must have length divisible by 4\n");
		exit(-1);
		}

	switch((dest>>8) & 0xf){
	case 0:
		R300_STATECHANGE(r300, vpi);
		for(i=0;i<vsf->length;i++)
			r300->hw.vpi.cmd[R300_VPI_INSTR_0+i+4*(dest & 0xff)]=(vsf->body.d[i]);
		bump_vpu_count(r300->hw.vpi.cmd, vsf->length+4*(dest & 0xff));
		break;

	case 2:
		R300_STATECHANGE(r300, vpp);
		for(i=0;i<vsf->length;i++)
			r300->hw.vpp.cmd[R300_VPP_PARAM_0+i+4*(dest & 0xff)]=(vsf->body.d[i]);
		bump_vpu_count(r300->hw.vpp.cmd, vsf->length+4*(dest & 0xff));
		break;
	case 4:
		R300_STATECHANGE(r300, vps);
		for(i=0;i<vsf->length;i++)
			r300->hw.vps.cmd[1+i+4*(dest & 0xff)]=(vsf->body.d[i]);
		bump_vpu_count(r300->hw.vps.cmd, vsf->length+4*(dest & 0xff));
		break;
	default:
		fprintf(stderr, "%s:%s don't know how to handle dest %04x\n", __FILE__, __FUNCTION__, dest);
		exit(-1);
	}
}

void r300SetupVertexProgram(r300ContextPtr rmesa);

/* just a skeleton for now.. */

/* Generate a vertex shader that simply transforms vertex and texture coordinates,
   while leaving colors intact. Nothing fancy (like lights) 
   
   If implementing lights make a copy first, so it is easy to switch between the two versions */
static void r300GenerateSimpleVertexShader(r300ContextPtr r300)
{
	int i;
	GLuint o_reg = 0;

	/* Allocate parameters */
	r300->state.vap_param.transform_offset=0x0;  /* transform matrix */
	r300->state.vertex_shader.param_offset=0x0;
	r300->state.vertex_shader.param_count=0x4;  /* 4 vector values - 4x4 matrix */
	
	r300->state.vertex_shader.program_start=0x0;
	r300->state.vertex_shader.unknown_ptr1=0x4; /* magic value ? */
	r300->state.vertex_shader.program_end=0x0;
	
	r300->state.vertex_shader.unknown_ptr2=0x0; /* magic value */
	r300->state.vertex_shader.unknown_ptr3=0x4; /* magic value */
	
	/* Initialize matrix and vector parameters.. these should really be restructured */
	/* TODO: fix vertex_shader structure */
	r300->state.vertex_shader.matrix[0].length=16;
	r300->state.vertex_shader.matrix[1].length=0;
	r300->state.vertex_shader.matrix[2].length=0;
	r300->state.vertex_shader.vector[0].length=0;
	r300->state.vertex_shader.vector[1].length=0;
	r300->state.vertex_shader.unknown1.length=0;
	r300->state.vertex_shader.unknown2.length=0;
	
#define WRITE_OP(oper,source1,source2,source3)	{\
	r300->state.vertex_shader.program.body.i[r300->state.vertex_shader.program_end].op=(oper); \
	r300->state.vertex_shader.program.body.i[r300->state.vertex_shader.program_end].src1=(source1); \
	r300->state.vertex_shader.program.body.i[r300->state.vertex_shader.program_end].src2=(source2); \
	r300->state.vertex_shader.program.body.i[r300->state.vertex_shader.program_end].src3=(source3); \
	r300->state.vertex_shader.program_end++; \
	}

	/* Multiply vertex coordinates with transform matrix */
			
	WRITE_OP(
		EASY_VSF_OP(MUL, 0, ALL, TMP),
		VSF_PARAM(3),
		VSF_ATTR_W(0),
		EASY_VSF_SOURCE(0, W, W, W, W, NONE, NONE)
		)
	
	WRITE_OP(
		EASY_VSF_OP(MUL, 1, ALL, RESULT),
		VSF_REG(1),
		VSF_ATTR_UNITY(1),
		VSF_UNITY(1)
		)
	
	WRITE_OP(
		EASY_VSF_OP(MAD, 0, ALL, TMP),
		VSF_PARAM(2),
		VSF_ATTR_Z(0),
		VSF_TMP(0)
		)
	
	WRITE_OP(
		EASY_VSF_OP(MAD, 0, ALL, TMP),
		VSF_PARAM(1),
		VSF_ATTR_Y(0),
		VSF_TMP(0)
		)
	
	WRITE_OP(
		EASY_VSF_OP(MAD, 0, ALL, RESULT),
		VSF_PARAM(0),
		VSF_ATTR_X(0),
		VSF_TMP(0)
		)
	o_reg += 2;
	
	if (r300->state.render_inputs & _TNL_BIT_COLOR1) {
		WRITE_OP(
			EASY_VSF_OP(MUL, o_reg++, ALL, RESULT),
			VSF_REG(r300->state.vap_reg.i_color[1]),
			VSF_ATTR_UNITY(r300->state.vap_reg.i_color[1]),
			VSF_UNITY(r300->state.vap_reg.i_color[1])
		)
	}
	
	/* Pass through texture coordinates, if any */
	for(i=0;i < r300->radeon.glCtx->Const.MaxTextureUnits;i++)
		if(r300->state.render_inputs & (_TNL_BIT_TEX0<<i)){
			// fprintf(stderr, "i_tex[%d]=%d\n", i, r300->state.vap_reg.i_tex[i]);
			WRITE_OP(
				EASY_VSF_OP(MUL, o_reg++ /* 2+i */, ALL, RESULT),
				VSF_REG(r300->state.vap_reg.i_tex[i]),
				VSF_ATTR_UNITY(r300->state.vap_reg.i_tex[i]),
				VSF_UNITY(r300->state.vap_reg.i_tex[i])
				)
			}
	
	r300->state.vertex_shader.program_end--; /* r300 wants program length to be one more - no idea why */
	r300->state.vertex_shader.program.length=(r300->state.vertex_shader.program_end+1)*4;
	
	r300->state.vertex_shader.unknown_ptr1=r300->state.vertex_shader.program_end; /* magic value ? */
	r300->state.vertex_shader.unknown_ptr2=r300->state.vertex_shader.program_end; /* magic value ? */
	r300->state.vertex_shader.unknown_ptr3=r300->state.vertex_shader.program_end; /* magic value ? */
	
}


void r300SetupVertexShader(r300ContextPtr rmesa)
{
	GLcontext* ctx = rmesa->radeon.glCtx;

	/* Reset state, in case we don't use something */
	((drm_r300_cmd_header_t*)rmesa->hw.vpp.cmd)->vpu.count = 0;
	((drm_r300_cmd_header_t*)rmesa->hw.vpi.cmd)->vpu.count = 0;
	((drm_r300_cmd_header_t*)rmesa->hw.vps.cmd)->vpu.count = 0;

	/* Not sure why this doesnt work...
	   0x400 area might have something to do with pixel shaders as it appears right after pfs programming.
	   0x406 is set to { 0.0, 0.0, 1.0, 0.0 } most of the time but should change with smooth points and in other rare cases. */
	//setup_vertex_shader_fragment(rmesa, 0x406, &unk4);
	if(hw_tcl_on && ((struct r300_vertex_program *)CURRENT_VERTEX_SHADER(ctx))->translated){
		r300SetupVertexProgram(rmesa);
		return ;
	}

/* This needs to be replaced by vertex shader generation code */


#if 0
	/* textures enabled ? */
	if(rmesa->state.texture.tc_count>0){
		rmesa->state.vertex_shader=SINGLE_TEXTURE_VERTEX_SHADER;
		} else {
		rmesa->state.vertex_shader=FLAT_COLOR_VERTEX_SHADER;
		}
#endif

	r300GenerateSimpleVertexShader(rmesa);

        rmesa->state.vertex_shader.matrix[0].length=16;
        memcpy(rmesa->state.vertex_shader.matrix[0].body.f, ctx->_ModelProjectMatrix.m, 16*4);

	setup_vertex_shader_fragment(rmesa, VSF_DEST_PROGRAM, &(rmesa->state.vertex_shader.program));

	setup_vertex_shader_fragment(rmesa, VSF_DEST_MATRIX0, &(rmesa->state.vertex_shader.matrix[0]));
#if 0
	setup_vertex_shader_fragment(rmesa, VSF_DEST_MATRIX1, &(rmesa->state.vertex_shader.matrix[0]));
	setup_vertex_shader_fragment(rmesa, VSF_DEST_MATRIX2, &(rmesa->state.vertex_shader.matrix[0]));

	setup_vertex_shader_fragment(rmesa, VSF_DEST_VECTOR0, &(rmesa->state.vertex_shader.vector[0]));
	setup_vertex_shader_fragment(rmesa, VSF_DEST_VECTOR1, &(rmesa->state.vertex_shader.vector[1]));
#endif

#if 0
	setup_vertex_shader_fragment(rmesa, VSF_DEST_UNKNOWN1, &(rmesa->state.vertex_shader.unknown1));
	setup_vertex_shader_fragment(rmesa, VSF_DEST_UNKNOWN2, &(rmesa->state.vertex_shader.unknown2));
#endif

	R300_STATECHANGE(rmesa, pvs);
	rmesa->hw.pvs.cmd[R300_PVS_CNTL_1]=(rmesa->state.vertex_shader.program_start << R300_PVS_CNTL_1_PROGRAM_START_SHIFT)
		| (rmesa->state.vertex_shader.unknown_ptr1 << R300_PVS_CNTL_1_POS_END_SHIFT)
		| (rmesa->state.vertex_shader.program_end << R300_PVS_CNTL_1_PROGRAM_END_SHIFT);
	rmesa->hw.pvs.cmd[R300_PVS_CNTL_2]=(rmesa->state.vertex_shader.param_offset << R300_PVS_CNTL_2_PARAM_OFFSET_SHIFT)
		| (rmesa->state.vertex_shader.param_count << R300_PVS_CNTL_2_PARAM_COUNT_SHIFT);
	rmesa->hw.pvs.cmd[R300_PVS_CNTL_3]=(rmesa->state.vertex_shader.unknown_ptr2 << R300_PVS_CNTL_3_PROGRAM_UNKNOWN_SHIFT)
	| (rmesa->state.vertex_shader.unknown_ptr3 << 0);

	/* This is done for vertex shader fragments, but also needs to be done for vap_pvs,
	so I leave it as a reminder */
#if 0
	reg_start(R300_VAP_PVS_WAITIDLE,0);
		e32(0x00000000);
#endif
}

void r300SetupVertexProgram(r300ContextPtr rmesa)
{
	GLcontext* ctx = rmesa->radeon.glCtx;
	int inst_count;
	int param_count;
	struct r300_vertex_program *prog=(struct r300_vertex_program *)CURRENT_VERTEX_SHADER(ctx);
			

	((drm_r300_cmd_header_t*)rmesa->hw.vpp.cmd)->vpu.count = 0;
	R300_STATECHANGE(rmesa, vpp);
	param_count = r300VertexProgUpdateParams(ctx, prog, (float *)&rmesa->hw.vpp.cmd[R300_VPP_PARAM_0]);
	bump_vpu_count(rmesa->hw.vpp.cmd, param_count);
	param_count /= 4;
	
	/* Reset state, in case we don't use something */
	((drm_r300_cmd_header_t*)rmesa->hw.vpi.cmd)->vpu.count = 0;
	((drm_r300_cmd_header_t*)rmesa->hw.vps.cmd)->vpu.count = 0;

	setup_vertex_shader_fragment(rmesa, VSF_DEST_PROGRAM, &(prog->program));

#if 0
	setup_vertex_shader_fragment(rmesa, VSF_DEST_UNKNOWN1, &(rmesa->state.vertex_shader.unknown1));
	setup_vertex_shader_fragment(rmesa, VSF_DEST_UNKNOWN2, &(rmesa->state.vertex_shader.unknown2));
#endif

	inst_count=prog->program.length/4 - 1;

	R300_STATECHANGE(rmesa, pvs);
	rmesa->hw.pvs.cmd[R300_PVS_CNTL_1]=(0 << R300_PVS_CNTL_1_PROGRAM_START_SHIFT)
		| (inst_count/*pos_end*/ << R300_PVS_CNTL_1_POS_END_SHIFT)
		| (inst_count << R300_PVS_CNTL_1_PROGRAM_END_SHIFT);
	rmesa->hw.pvs.cmd[R300_PVS_CNTL_2]=(0 << R300_PVS_CNTL_2_PARAM_OFFSET_SHIFT)
		| (param_count << R300_PVS_CNTL_2_PARAM_COUNT_SHIFT);
	rmesa->hw.pvs.cmd[R300_PVS_CNTL_3]=(0/*rmesa->state.vertex_shader.unknown_ptr2*/ << R300_PVS_CNTL_3_PROGRAM_UNKNOWN_SHIFT)
	| (inst_count /*rmesa->state.vertex_shader.unknown_ptr3*/ << 0);

	/* This is done for vertex shader fragments, but also needs to be done for vap_pvs,
	so I leave it as a reminder */
#if 0
	reg_start(R300_VAP_PVS_WAITIDLE,0);
		e32(0x00000000);
#endif
}

extern void _tnl_UpdateFixedFunctionProgram( GLcontext *ctx );

extern int future_hw_tcl_on;
void r300UpdateShaders(r300ContextPtr rmesa)
{
	GLcontext *ctx;
	struct r300_vertex_program *vp;
	
	ctx = rmesa->radeon.glCtx;
	
	/* Disable tnl programs when doing software vertex programs.
	   I can only hope this actually disables it at the right time. */
	ctx->_MaintainTnlProgram = hw_tcl_on;
	
	if (rmesa->NewGLState && hw_tcl_on) {
		rmesa->NewGLState = 0;
		
		_tnl_UpdateFixedFunctionProgram(ctx);
	
		vp = (struct r300_vertex_program *)CURRENT_VERTEX_SHADER(ctx);
		if (vp->translated == GL_FALSE)
			r300_translate_vertex_shader(vp);
		if (vp->translated == GL_FALSE) {
			fprintf(stderr, "Failing back to sw-tcl\n");
			hw_tcl_on = future_hw_tcl_on = 0;
			r300ResetHwState(rmesa);

			return ;
		}
	}
	
}

void r300UpdateShaderStates(r300ContextPtr rmesa)
{
	GLcontext *ctx;
	ctx = rmesa->radeon.glCtx;
	
#ifdef CB_DPATH
	r300UpdateTextureState(ctx);
#endif

	r300SetupPixelShader(rmesa);
	r300_setup_textures(ctx);
	
	r300SetupVertexShader(rmesa);
	r300_setup_rs_unit(ctx);
}

/* This is probably wrong for some values, I need to test this
 * some more.  Range checking would be a good idea also..
 * 
 * But it works for most things.  I'll fix it later if someone
 * else with a better clue doesn't
 */
static unsigned int r300PackFloat24(float f)
{
	float mantissa;
	int exponent;
	unsigned int float24 = 0;

	if (f == 0.0) return 0;

	mantissa = frexpf(f, &exponent);

	/* Handle -ve */
	if (mantissa < 0) {
		float24 |= (1<<23);
		mantissa = mantissa * -1.0;
	}
	/* Handle exponent, bias of 63 */
	exponent += 62;
	float24 |= (exponent << 16);
	/* Kill 7 LSB of mantissa */
	float24 |= (r300PackFloat32(mantissa) & 0x7FFFFF)  >> 7;

	return float24;
}

void r300SetupPixelShader(r300ContextPtr rmesa)
{
	GLcontext *ctx = rmesa->radeon.glCtx;
	struct r300_fragment_program *rp =
		(struct r300_fragment_program *)
		(char *)ctx->FragmentProgram._Current;
	int i,k;

	if (!rp)	/* should only happenen once, just after context is created */
		return;
	
	r300_translate_fragment_shader(rp);
	if (!rp->translated) {
		fprintf(stderr, "%s: No valid fragment shader, exiting\n", __func__);
		exit(-1);
	}
	
#define OUTPUT_FIELD(st, reg, field)  \
		R300_STATECHANGE(rmesa, st); \
		for(i=0;i<=rp->alu_end;i++) \
			rmesa->hw.st.cmd[R300_FPI_INSTR_0+i]=rp->alu.inst[i].field;\
		rmesa->hw.st.cmd[R300_FPI_CMD_0]=cmdpacket0(reg, rp->alu_end+1);

	OUTPUT_FIELD(fpi[0], R300_PFS_INSTR0_0, inst0);
	OUTPUT_FIELD(fpi[1], R300_PFS_INSTR1_0, inst1);
	OUTPUT_FIELD(fpi[2], R300_PFS_INSTR2_0, inst2);
	OUTPUT_FIELD(fpi[3], R300_PFS_INSTR3_0, inst3);
#undef OUTPUT_FIELD

	R300_STATECHANGE(rmesa, fp);
	/* I just want to say, the way these nodes are stored.. weird.. */
	for (i=0,k=(4-(rp->cur_node+1));i<4;i++,k++) {
		if (i<(rp->cur_node+1)) {
			rmesa->hw.fp.cmd[R300_FP_NODE0+k]=
				(rp->node[i].alu_offset << R300_PFS_NODE_ALU_OFFSET_SHIFT)
				| (rp->node[i].alu_end  << R300_PFS_NODE_ALU_END_SHIFT)
				| (rp->node[i].tex_offset << R300_PFS_NODE_TEX_OFFSET_SHIFT)
				| (rp->node[i].tex_end  << R300_PFS_NODE_TEX_END_SHIFT)
				| rp->node[i].flags; /*  ( (k==3) ? R300_PFS_NODE_LAST_NODE : 0); */
		} else {
			rmesa->hw.fp.cmd[R300_FP_NODE0+(3-i)] = 0;
		}
	}

		/*  PFS_CNTL_0 */
	rmesa->hw.fp.cmd[R300_FP_CNTL0]=
		rp->cur_node
		| (rp->first_node_has_tex<<3);
		/* PFS_CNTL_1 */
	rmesa->hw.fp.cmd[R300_FP_CNTL1]=rp->max_temp_idx;
		/* PFS_CNTL_2 */
	rmesa->hw.fp.cmd[R300_FP_CNTL2]=
		(rp->alu_offset << R300_PFS_CNTL_ALU_OFFSET_SHIFT)
		| (rp->alu_end << R300_PFS_CNTL_ALU_END_SHIFT)
		| (rp->tex_offset << R300_PFS_CNTL_TEX_OFFSET_SHIFT)
		| (rp->tex_end << R300_PFS_CNTL_TEX_END_SHIFT);

	R300_STATECHANGE(rmesa, fpp);
	for(i=0;i<rp->const_nr;i++){
		rmesa->hw.fpp.cmd[R300_FPP_PARAM_0+4*i+0]=r300PackFloat24(rp->constant[i][0]);
		rmesa->hw.fpp.cmd[R300_FPP_PARAM_0+4*i+1]=r300PackFloat24(rp->constant[i][1]);
		rmesa->hw.fpp.cmd[R300_FPP_PARAM_0+4*i+2]=r300PackFloat24(rp->constant[i][2]);
		rmesa->hw.fpp.cmd[R300_FPP_PARAM_0+4*i+3]=r300PackFloat24(rp->constant[i][3]);
	}
	rmesa->hw.fpp.cmd[R300_FPP_CMD_0]=cmdpacket0(R300_PFS_PARAM_0_X, rp->const_nr*4);
}

/**
 * Called by Mesa after an internal state update.
 */
static void r300InvalidateState(GLcontext * ctx, GLuint new_state)
{
	r300ContextPtr r300 = R300_CONTEXT(ctx);
	
	_swrast_InvalidateState(ctx, new_state);
	_swsetup_InvalidateState(ctx, new_state);
	_ac_InvalidateState(ctx, new_state);
	_tnl_InvalidateState(ctx, new_state);
	_ae_invalidate_state(ctx, new_state);

	if (new_state & (_NEW_BUFFERS | _NEW_COLOR | _NEW_PIXEL)) {
		r300UpdateDrawBuffer(ctx);
	}
#ifndef CB_DPATH
	/* Go inefficiency! */
	r300ResetHwState(r300);
#endif
#ifdef HW_VBOS
	if(new_state & _NEW_ARRAY)
		r300->state.VB.lock_uptodate = GL_FALSE;
#endif
	r300->NewGLState |= new_state;
}

/**
 * Completely recalculates hardware state based on the Mesa state.
 */
void r300ResetHwState(r300ContextPtr r300)