#include "tnl/tnl.h"
#include "compiler/radeon_compiler.h"
-#include "compiler/radeon_nqssadce.h"
+#include "radeon_mesa_to_rc.h"
#include "r300_context.h"
+#include "r300_fragprog_common.h"
#include "r300_state.h"
-
-static int r300VertexProgUpdateParams(GLcontext * ctx, struct gl_vertex_program *vp, float *dst)
+/**
+ * Write parameter array for the given vertex program into dst.
+ * Return the total number of components written.
+ */
+static int r300VertexProgUpdateParams(GLcontext * ctx, struct r300_vertex_program *vp, float *dst)
{
- int pi;
- float *dst_o = dst;
- struct gl_program_parameter_list *paramList;
+ int i;
- if (vp->IsNVProgram) {
+ if (vp->Base->IsNVProgram) {
_mesa_load_tracked_matrices(ctx);
+ } else {
+ if (vp->Base->Base.Parameters) {
+ _mesa_load_state_parameters(ctx, vp->Base->Base.Parameters);
+ }
+ }
- for (pi = 0; pi < MAX_NV_VERTEX_PROGRAM_PARAMS; pi++) {
- *dst++ = ctx->VertexProgram.Parameters[pi][0];
- *dst++ = ctx->VertexProgram.Parameters[pi][1];
- *dst++ = ctx->VertexProgram.Parameters[pi][2];
- *dst++ = ctx->VertexProgram.Parameters[pi][3];
+ for(i = 0; i < vp->code.constants.Count; ++i) {
+ const float * src = 0;
+ const struct rc_constant * constant = &vp->code.constants.Constants[i];
+
+ switch(constant->Type) {
+ case RC_CONSTANT_EXTERNAL:
+ if (vp->Base->IsNVProgram) {
+ src = ctx->VertexProgram.Parameters[constant->u.External];
+ } else {
+ src = vp->Base->Base.Parameters->ParameterValues[constant->u.External];
+ }
+ break;
+
+ case RC_CONSTANT_IMMEDIATE:
+ src = constant->u.Immediate;
+ break;
}
- return dst - dst_o;
+
+ dst[4*i] = src[0];
+ dst[4*i + 1] = src[1];
+ dst[4*i + 2] = src[2];
+ dst[4*i + 3] = src[3];
}
- if (!vp->Base.Parameters)
- return 0;
+ return 4 * vp->code.constants.Count;
+}
+
+static GLbitfield compute_required_outputs(struct gl_vertex_program * vp, GLbitfield fpreads)
+{
+ GLbitfield outputs = 0;
+ int i;
- _mesa_load_state_parameters(ctx, vp->Base.Parameters);
+#define ADD_OUTPUT(fp_attr, vp_result) \
+ do { \
+ if (fpreads & (1 << (fp_attr))) \
+ outputs |= (1 << (vp_result)); \
+ } while (0)
- if (vp->Base.Parameters->NumParameters * 4 >
- VSF_MAX_FRAGMENT_LENGTH) {
- fprintf(stderr, "%s:Params exhausted\n", __FUNCTION__);
- _mesa_exit(-1);
+ ADD_OUTPUT(FRAG_ATTRIB_COL0, VERT_RESULT_COL0);
+ ADD_OUTPUT(FRAG_ATTRIB_COL1, VERT_RESULT_COL1);
+
+ for (i = 0; i <= 7; ++i) {
+ ADD_OUTPUT(FRAG_ATTRIB_TEX0 + i, VERT_RESULT_TEX0 + i);
}
- paramList = vp->Base.Parameters;
- for (pi = 0; pi < paramList->NumParameters; pi++) {
- switch (paramList->Parameters[pi].Type) {
- case PROGRAM_STATE_VAR:
- case PROGRAM_NAMED_PARAM:
- //fprintf(stderr, "%s", vp->Parameters->Parameters[pi].Name);
- case PROGRAM_CONSTANT:
- *dst++ = paramList->ParameterValues[pi][0];
- *dst++ = paramList->ParameterValues[pi][1];
- *dst++ = paramList->ParameterValues[pi][2];
- *dst++ = paramList->ParameterValues[pi][3];
- break;
- default:
- _mesa_problem(NULL, "Bad param type in %s",
- __FUNCTION__);
+#undef ADD_OUTPUT
+
+ if ((fpreads & (1 << FRAG_ATTRIB_COL0)) &&
+ (vp->Base.OutputsWritten & (1 << VERT_RESULT_BFC0)))
+ outputs |= 1 << VERT_RESULT_BFC0;
+ if ((fpreads & (1 << FRAG_ATTRIB_COL1)) &&
+ (vp->Base.OutputsWritten & (1 << VERT_RESULT_BFC1)))
+ outputs |= 1 << VERT_RESULT_BFC1;
+
+ outputs |= 1 << VERT_RESULT_HPOS;
+ if (vp->Base.OutputsWritten & (1 << VERT_RESULT_PSIZ))
+ outputs |= 1 << VERT_RESULT_PSIZ;
+
+ return outputs;
+}
+
+
+static void t_inputs_outputs(struct r300_vertex_program_compiler * c)
+{
+ int i;
+ int cur_reg;
+ GLuint OutputsWritten, InputsRead;
+
+ OutputsWritten = c->Base.Program.OutputsWritten;
+ InputsRead = c->Base.Program.InputsRead;
+
+ cur_reg = -1;
+ for (i = 0; i < VERT_ATTRIB_MAX; i++) {
+ if (InputsRead & (1 << i))
+ c->code->inputs[i] = ++cur_reg;
+ else
+ c->code->inputs[i] = -1;
+ }
+
+ cur_reg = 0;
+ for (i = 0; i < VERT_RESULT_MAX; i++)
+ c->code->outputs[i] = -1;
+
+ assert(OutputsWritten & (1 << VERT_RESULT_HPOS));
+
+ if (OutputsWritten & (1 << VERT_RESULT_HPOS)) {
+ c->code->outputs[VERT_RESULT_HPOS] = cur_reg++;
+ }
+
+ if (OutputsWritten & (1 << VERT_RESULT_PSIZ)) {
+ c->code->outputs[VERT_RESULT_PSIZ] = cur_reg++;
+ }
+
+ /* If we're writing back facing colors we need to send
+ * four colors to make front/back face colors selection work.
+ * If the vertex program doesn't write all 4 colors, lets
+ * pretend it does by skipping output index reg so the colors
+ * get written into appropriate output vectors.
+ */
+ if (OutputsWritten & (1 << VERT_RESULT_COL0)) {
+ c->code->outputs[VERT_RESULT_COL0] = cur_reg++;
+ } else if (OutputsWritten & (1 << VERT_RESULT_BFC0) ||
+ OutputsWritten & (1 << VERT_RESULT_BFC1)) {
+ cur_reg++;
+ }
+
+ if (OutputsWritten & (1 << VERT_RESULT_COL1)) {
+ c->code->outputs[VERT_RESULT_COL1] = cur_reg++;
+ } else if (OutputsWritten & (1 << VERT_RESULT_BFC0) ||
+ OutputsWritten & (1 << VERT_RESULT_BFC1)) {
+ cur_reg++;
+ }
+
+ if (OutputsWritten & (1 << VERT_RESULT_BFC0)) {
+ c->code->outputs[VERT_RESULT_BFC0] = cur_reg++;
+ } else if (OutputsWritten & (1 << VERT_RESULT_BFC1)) {
+ cur_reg++;
+ }
+
+ if (OutputsWritten & (1 << VERT_RESULT_BFC1)) {
+ c->code->outputs[VERT_RESULT_BFC1] = cur_reg++;
+ } else if (OutputsWritten & (1 << VERT_RESULT_BFC0)) {
+ cur_reg++;
+ }
+
+ for (i = VERT_RESULT_TEX0; i <= VERT_RESULT_TEX7; i++) {
+ if (OutputsWritten & (1 << i)) {
+ c->code->outputs[i] = cur_reg++;
}
+ }
+ if (OutputsWritten & (1 << VERT_RESULT_FOGC)) {
+ c->code->outputs[VERT_RESULT_FOGC] = cur_reg++;
}
+}
- return dst - dst_o;
+/**
+ * The NV_vertex_program spec mandates that all registers be
+ * initialized to zero. We do this here unconditionally.
+ *
+ * \note We rely on dead-code elimination in the compiler.
+ */
+static void initialize_NV_registers(struct radeon_compiler * compiler)
+{
+ unsigned int reg;
+ struct rc_instruction * inst;
+
+ for(reg = 0; reg < 12; ++reg) {
+ inst = rc_insert_new_instruction(compiler, &compiler->Program.Instructions);
+ inst->U.I.Opcode = RC_OPCODE_MOV;
+ inst->U.I.DstReg.File = RC_FILE_TEMPORARY;
+ inst->U.I.DstReg.Index = reg;
+ inst->U.I.SrcReg[0].File = RC_FILE_NONE;
+ inst->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_0000;
+ }
+
+ inst = rc_insert_new_instruction(compiler, &compiler->Program.Instructions);
+ inst->U.I.Opcode = RC_OPCODE_ARL;
+ inst->U.I.DstReg.File = RC_FILE_ADDRESS;
+ inst->U.I.DstReg.Index = 0;
+ inst->U.I.DstReg.WriteMask = WRITEMASK_X;
+ inst->U.I.SrcReg[0].File = RC_FILE_NONE;
+ inst->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_0000;
}
static struct r300_vertex_program *build_program(GLcontext *ctx,
- struct r300_vertex_program_external_state *wanted_key,
+ struct r300_vertex_program_key *wanted_key,
const struct gl_vertex_program *mesa_vp)
{
struct r300_vertex_program *vp;
_mesa_memcpy(&vp->key, wanted_key, sizeof(vp->key));
rc_init(&compiler.Base);
- compiler.Base.Debug = (RADEON_DEBUG & DEBUG_VERTS) ? GL_TRUE : GL_FALSE;
+ compiler.Base.Debug = (RADEON_DEBUG & RADEON_VERTS) ? GL_TRUE : GL_FALSE;
compiler.code = &vp->code;
- compiler.state = vp->key;
- compiler.program = vp->Base;
+ compiler.RequiredOutputs = compute_required_outputs(vp->Base, vp->key.FpReads);
+ compiler.SetHwInputOutput = &t_inputs_outputs;
if (compiler.Base.Debug) {
fprintf(stderr, "Initial vertex program:\n");
- _mesa_print_program(compiler.program);
- fflush(stdout);
+ _mesa_print_program(&vp->Base->Base);
+ fflush(stderr);
}
if (mesa_vp->IsPositionInvariant) {
- _mesa_insert_mvp_code(ctx, (struct gl_vertex_program *)compiler.program);
+ _mesa_insert_mvp_code(ctx, vp->Base);
+ }
+
+ radeon_mesa_to_rc_program(&compiler.Base, &vp->Base->Base);
+
+ if (mesa_vp->IsNVProgram)
+ initialize_NV_registers(&compiler.Base);
+
+ rc_move_output(&compiler.Base, VERT_RESULT_PSIZ, VERT_RESULT_PSIZ, WRITEMASK_X);
+
+ if (vp->key.WPosAttr != FRAG_ATTRIB_MAX) {
+ rc_copy_output(&compiler.Base,
+ VERT_RESULT_HPOS,
+ vp->key.WPosAttr - FRAG_ATTRIB_TEX0 + VERT_RESULT_TEX0);
+ }
+
+ if (vp->key.FogAttr != FRAG_ATTRIB_MAX) {
+ rc_move_output(&compiler.Base,
+ VERT_RESULT_FOGC,
+ vp->key.FogAttr - FRAG_ATTRIB_TEX0 + VERT_RESULT_TEX0, WRITEMASK_X);
}
r3xx_compile_vertex_program(&compiler);
+
+ if (vp->code.constants.Count > ctx->Const.VertexProgram.MaxParameters) {
+ rc_error(&compiler.Base, "Program exceeds constant buffer size limit\n");
+ }
+
vp->error = compiler.Base.Error;
+ vp->Base->Base.InputsRead = vp->code.InputsRead;
+ vp->Base->Base.OutputsWritten = vp->code.OutputsWritten;
+
rc_destroy(&compiler.Base);
return vp;
struct r300_vertex_program * r300SelectAndTranslateVertexShader(GLcontext *ctx)
{
r300ContextPtr r300 = R300_CONTEXT(ctx);
- struct r300_vertex_program_external_state wanted_key = { 0 };
+ struct r300_vertex_program_key wanted_key = { 0 };
struct r300_vertex_program_cont *vpc;
struct r300_vertex_program *vp;
vpc = (struct r300_vertex_program_cont *)ctx->VertexProgram._Current;
+
+ if (!r300->selected_fp) {
+ /* This can happen when GetProgramiv is called to check
+ * whether the program runs natively.
+ *
+ * To be honest, this is not a very good solution,
+ * but solving the problem of reporting good values
+ * for those queries is tough anyway considering that
+ * we recompile vertex programs based on the precise
+ * fragment program that is in use.
+ */
+ r300SelectAndTranslateFragmentShader(ctx);
+ }
+
wanted_key.FpReads = r300->selected_fp->InputsRead;
- wanted_key.FogAttr = r300->selected_fp->code.fog_attr;
- wanted_key.WPosAttr = r300->selected_fp->code.wpos_attr;
+ wanted_key.FogAttr = r300->selected_fp->fog_attr;
+ wanted_key.WPosAttr = r300->selected_fp->wpos_attr;
for (vp = vpc->progs; vp; vp = vp->next) {
if (_mesa_memcmp(&vp->key, &wanted_key, sizeof(wanted_key))
#define bump_vpu_count(ptr, new_count) do { \
drm_r300_cmd_header_t* _p=((drm_r300_cmd_header_t*)(ptr)); \
int _nc=(new_count)/4; \
- assert(_nc < 256); \
if(_nc>_p->vpu.count)_p->vpu.count=_nc; \
} while(0)
assert((code->length > 0) && (code->length % 4 == 0));
+ R300_STATECHANGE( r300, vap_flush );
+
switch ((dest >> 8) & 0xf) {
case 0:
R300_STATECHANGE(r300, vpi);
((drm_r300_cmd_header_t *) rmesa->hw.vpi.cmd)->vpu.count = 0;
((drm_r300_cmd_header_t *) rmesa->hw.vps.cmd)->vpu.count = 0;
+ R300_STATECHANGE(rmesa, vap_flush);
R300_STATECHANGE(rmesa, vpp);
- param_count = r300VertexProgUpdateParams(ctx, prog->Base, (float *)&rmesa->hw.vpp.cmd[R300_VPP_PARAM_0]);
+ 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;