#include "r300_fragprog_swizzle.h"
+struct r300_emit_state {
+ struct r300_fragment_program_compiler * compiler;
+
+ unsigned current_node : 2;
+ unsigned node_first_tex : 8;
+ unsigned node_first_alu : 8;
+ uint32_t node_flags;
+};
+
#define PROG_CODE \
- struct r300_fragment_program_compiler *c = (struct r300_fragment_program_compiler*)data; \
+ struct r300_fragment_program_compiler *c = emit->compiler; \
struct r300_fragment_program_code *code = &c->code->code.r300
#define error(fmt, args...) do { \
/**
* Mark a temporary register as used.
*/
-static void use_temporary(struct r300_fragment_program_code *code, GLuint index)
+static void use_temporary(struct r300_fragment_program_code *code, unsigned int index)
{
- if (index > code->max_temp_idx)
- code->max_temp_idx = index;
+ if (index > code->pixsize)
+ code->pixsize = index;
}
+static unsigned int use_source(struct r300_fragment_program_code* code, struct radeon_pair_instruction_source src)
+{
+ if (src.File == RC_FILE_CONSTANT) {
+ return src.Index | (1 << 5);
+ } else if (src.File == RC_FILE_TEMPORARY) {
+ use_temporary(code, src.Index);
+ return src.Index;
+ }
-static GLuint translate_rgb_opcode(struct r300_fragment_program_compiler * c, GLuint opcode)
+ return 0;
+}
+
+
+static unsigned int translate_rgb_opcode(struct r300_fragment_program_compiler * c, rc_opcode opcode)
{
switch(opcode) {
- case OPCODE_CMP: return R300_ALU_OUTC_CMP;
- case OPCODE_DP3: return R300_ALU_OUTC_DP3;
- case OPCODE_DP4: return R300_ALU_OUTC_DP4;
- case OPCODE_FRC: return R300_ALU_OUTC_FRC;
+ case RC_OPCODE_CMP: return R300_ALU_OUTC_CMP;
+ case RC_OPCODE_DP3: return R300_ALU_OUTC_DP3;
+ case RC_OPCODE_DP4: return R300_ALU_OUTC_DP4;
+ case RC_OPCODE_FRC: return R300_ALU_OUTC_FRC;
default:
error("translate_rgb_opcode(%i): Unknown opcode", opcode);
/* fall through */
- case OPCODE_NOP:
+ case RC_OPCODE_NOP:
/* fall through */
- case OPCODE_MAD: return R300_ALU_OUTC_MAD;
- case OPCODE_MAX: return R300_ALU_OUTC_MAX;
- case OPCODE_MIN: return R300_ALU_OUTC_MIN;
- case OPCODE_REPL_ALPHA: return R300_ALU_OUTC_REPL_ALPHA;
+ case RC_OPCODE_MAD: return R300_ALU_OUTC_MAD;
+ case RC_OPCODE_MAX: return R300_ALU_OUTC_MAX;
+ case RC_OPCODE_MIN: return R300_ALU_OUTC_MIN;
+ case RC_OPCODE_REPL_ALPHA: return R300_ALU_OUTC_REPL_ALPHA;
}
}
-static GLuint translate_alpha_opcode(struct r300_fragment_program_compiler * c, GLuint opcode)
+static unsigned int translate_alpha_opcode(struct r300_fragment_program_compiler * c, rc_opcode opcode)
{
switch(opcode) {
- case OPCODE_CMP: return R300_ALU_OUTA_CMP;
- case OPCODE_DP3: return R300_ALU_OUTA_DP4;
- case OPCODE_DP4: return R300_ALU_OUTA_DP4;
- case OPCODE_EX2: return R300_ALU_OUTA_EX2;
- case OPCODE_FRC: return R300_ALU_OUTA_FRC;
- case OPCODE_LG2: return R300_ALU_OUTA_LG2;
+ case RC_OPCODE_CMP: return R300_ALU_OUTA_CMP;
+ case RC_OPCODE_DP3: return R300_ALU_OUTA_DP4;
+ case RC_OPCODE_DP4: return R300_ALU_OUTA_DP4;
+ case RC_OPCODE_EX2: return R300_ALU_OUTA_EX2;
+ case RC_OPCODE_FRC: return R300_ALU_OUTA_FRC;
+ case RC_OPCODE_LG2: return R300_ALU_OUTA_LG2;
default:
error("translate_rgb_opcode(%i): Unknown opcode", opcode);
/* fall through */
- case OPCODE_NOP:
+ case RC_OPCODE_NOP:
/* fall through */
- case OPCODE_MAD: return R300_ALU_OUTA_MAD;
- case OPCODE_MAX: return R300_ALU_OUTA_MAX;
- case OPCODE_MIN: return R300_ALU_OUTA_MIN;
- case OPCODE_RCP: return R300_ALU_OUTA_RCP;
- case OPCODE_RSQ: return R300_ALU_OUTA_RSQ;
+ case RC_OPCODE_MAD: return R300_ALU_OUTA_MAD;
+ case RC_OPCODE_MAX: return R300_ALU_OUTA_MAX;
+ case RC_OPCODE_MIN: return R300_ALU_OUTA_MIN;
+ case RC_OPCODE_RCP: return R300_ALU_OUTA_RCP;
+ case RC_OPCODE_RSQ: return R300_ALU_OUTA_RSQ;
}
}
/**
* Emit one paired ALU instruction.
*/
-static GLboolean emit_alu(void* data, struct radeon_pair_instruction* inst)
+static int emit_alu(struct r300_emit_state * emit, struct rc_pair_instruction* inst)
{
PROG_CODE;
if (code->alu.length >= R300_PFS_MAX_ALU_INST) {
error("Too many ALU instructions");
- return GL_FALSE;
+ return 0;
}
int ip = code->alu.length++;
int j;
- code->node[code->cur_node].alu_end++;
- code->alu.inst[ip].inst0 = translate_rgb_opcode(c, inst->RGB.Opcode);
- code->alu.inst[ip].inst2 = translate_alpha_opcode(c, inst->Alpha.Opcode);
+ code->alu.inst[ip].rgb_inst = translate_rgb_opcode(c, inst->RGB.Opcode);
+ code->alu.inst[ip].alpha_inst = translate_alpha_opcode(c, inst->Alpha.Opcode);
for(j = 0; j < 3; ++j) {
- GLuint src = inst->RGB.Src[j].Index | (inst->RGB.Src[j].Constant << 5);
- if (!inst->RGB.Src[j].Constant)
- use_temporary(code, inst->RGB.Src[j].Index);
- code->alu.inst[ip].inst1 |= src << (6*j);
+ unsigned int src = use_source(code, inst->RGB.Src[j]);
+ code->alu.inst[ip].rgb_addr |= src << (6*j);
- src = inst->Alpha.Src[j].Index | (inst->Alpha.Src[j].Constant << 5);
- if (!inst->Alpha.Src[j].Constant)
- use_temporary(code, inst->Alpha.Src[j].Index);
- code->alu.inst[ip].inst3 |= src << (6*j);
+ src = use_source(code, inst->Alpha.Src[j]);
+ code->alu.inst[ip].alpha_addr |= src << (6*j);
- GLuint arg = r300FPTranslateRGBSwizzle(inst->RGB.Arg[j].Source, inst->RGB.Arg[j].Swizzle);
+ unsigned int arg = r300FPTranslateRGBSwizzle(inst->RGB.Arg[j].Source, inst->RGB.Arg[j].Swizzle);
arg |= inst->RGB.Arg[j].Abs << 6;
arg |= inst->RGB.Arg[j].Negate << 5;
- code->alu.inst[ip].inst0 |= arg << (7*j);
+ code->alu.inst[ip].rgb_inst |= arg << (7*j);
arg = r300FPTranslateAlphaSwizzle(inst->Alpha.Arg[j].Source, inst->Alpha.Arg[j].Swizzle);
arg |= inst->Alpha.Arg[j].Abs << 6;
arg |= inst->Alpha.Arg[j].Negate << 5;
- code->alu.inst[ip].inst2 |= arg << (7*j);
+ code->alu.inst[ip].alpha_inst |= arg << (7*j);
}
if (inst->RGB.Saturate)
- code->alu.inst[ip].inst0 |= R300_ALU_OUTC_CLAMP;
+ code->alu.inst[ip].rgb_inst |= R300_ALU_OUTC_CLAMP;
if (inst->Alpha.Saturate)
- code->alu.inst[ip].inst2 |= R300_ALU_OUTA_CLAMP;
+ code->alu.inst[ip].alpha_inst |= R300_ALU_OUTA_CLAMP;
if (inst->RGB.WriteMask) {
use_temporary(code, inst->RGB.DestIndex);
- code->alu.inst[ip].inst1 |=
+ code->alu.inst[ip].rgb_addr |=
(inst->RGB.DestIndex << R300_ALU_DSTC_SHIFT) |
(inst->RGB.WriteMask << R300_ALU_DSTC_REG_MASK_SHIFT);
}
if (inst->RGB.OutputWriteMask) {
- code->alu.inst[ip].inst1 |= (inst->RGB.OutputWriteMask << R300_ALU_DSTC_OUTPUT_MASK_SHIFT);
- code->node[code->cur_node].flags |= R300_RGBA_OUT;
+ code->alu.inst[ip].rgb_addr |=
+ (inst->RGB.OutputWriteMask << R300_ALU_DSTC_OUTPUT_MASK_SHIFT) |
+ R300_RGB_TARGET(inst->RGB.Target);
+ emit->node_flags |= R300_RGBA_OUT;
}
if (inst->Alpha.WriteMask) {
use_temporary(code, inst->Alpha.DestIndex);
- code->alu.inst[ip].inst3 |=
+ code->alu.inst[ip].alpha_addr |=
(inst->Alpha.DestIndex << R300_ALU_DSTA_SHIFT) |
R300_ALU_DSTA_REG;
}
if (inst->Alpha.OutputWriteMask) {
- code->alu.inst[ip].inst3 |= R300_ALU_DSTA_OUTPUT;
- code->node[code->cur_node].flags |= R300_RGBA_OUT;
+ code->alu.inst[ip].alpha_addr |= R300_ALU_DSTA_OUTPUT |
+ R300_ALPHA_TARGET(inst->Alpha.Target);
+ emit->node_flags |= R300_RGBA_OUT;
}
if (inst->Alpha.DepthWriteMask) {
- code->alu.inst[ip].inst3 |= R300_ALU_DSTA_DEPTH;
- code->node[code->cur_node].flags |= R300_W_OUT;
- c->code->writes_depth = GL_TRUE;
+ code->alu.inst[ip].alpha_addr |= R300_ALU_DSTA_DEPTH;
+ emit->node_flags |= R300_W_OUT;
+ c->code->writes_depth = 1;
}
- return GL_TRUE;
+ return 1;
}
/**
* Finish the current node without advancing to the next one.
*/
-static GLboolean finish_node(struct r300_fragment_program_compiler *c)
+static int finish_node(struct r300_emit_state * emit)
{
- struct r300_fragment_program_code *code = &c->code->code.r300;
- struct r300_fragment_program_node *node = &code->node[code->cur_node];
+ struct r300_fragment_program_compiler * c = emit->compiler;
+ struct r300_fragment_program_code *code = &emit->compiler->code->code.r300;
- if (node->alu_end < 0) {
+ if (code->alu.length == emit->node_first_alu) {
/* Generate a single NOP for this node */
- struct radeon_pair_instruction inst;
- _mesa_bzero(&inst, sizeof(inst));
- if (!emit_alu(c, &inst))
- return GL_FALSE;
+ struct rc_pair_instruction inst;
+ memset(&inst, 0, sizeof(inst));
+ if (!emit_alu(emit, &inst))
+ return 0;
}
- if (node->tex_end < 0) {
- if (code->cur_node == 0) {
- node->tex_end = 0;
- } else {
- error("Node %i has no TEX instructions", code->cur_node);
- return GL_FALSE;
+ unsigned alu_offset = emit->node_first_alu;
+ unsigned alu_end = code->alu.length - alu_offset - 1;
+ unsigned tex_offset = emit->node_first_tex;
+ unsigned tex_end = code->tex.length - tex_offset - 1;
+
+ if (code->tex.length == emit->node_first_tex) {
+ if (emit->current_node > 0) {
+ error("Node %i has no TEX instructions", emit->current_node);
+ return 0;
}
+
+ tex_end = 0;
} else {
- if (code->cur_node == 0)
- code->first_node_has_tex = 1;
+ if (emit->current_node == 0)
+ code->config |= R300_PFS_CNTL_FIRST_NODE_HAS_TEX;
}
- return GL_TRUE;
+ /* Write the config register.
+ * Note: The order in which the words for each node are written
+ * is not correct here and needs to be fixed up once we're entirely
+ * done
+ *
+ * Also note that the register specification from AMD is slightly
+ * incorrect in its description of this register. */
+ code->code_addr[emit->current_node] =
+ (alu_offset << R300_ALU_START_SHIFT) |
+ (alu_end << R300_ALU_SIZE_SHIFT) |
+ (tex_offset << R300_TEX_START_SHIFT) |
+ (tex_end << R300_TEX_SIZE_SHIFT) |
+ emit->node_flags;
+
+ return 1;
}
* Begin a block of texture instructions.
* Create the necessary indirection.
*/
-static GLboolean begin_tex(void* data)
+static int begin_tex(struct r300_emit_state * emit)
{
PROG_CODE;
- if (code->cur_node == 0) {
- if (code->node[0].alu_end < 0 &&
- code->node[0].tex_end < 0)
- return GL_TRUE;
+ if (code->alu.length == emit->node_first_alu &&
+ code->tex.length == emit->node_first_tex) {
+ return 1;
}
- if (code->cur_node == 3) {
+ if (emit->current_node == 3) {
error("Too many texture indirections");
- return GL_FALSE;
+ return 0;
}
- if (!finish_node(c))
- return GL_FALSE;
+ if (!finish_node(emit))
+ return 0;
- struct r300_fragment_program_node *node = &code->node[++code->cur_node];
- node->alu_offset = code->alu.length;
- node->alu_end = -1;
- node->tex_offset = code->tex.length;
- node->tex_end = -1;
- return GL_TRUE;
+ emit->current_node++;
+ emit->node_first_tex = code->tex.length;
+ emit->node_first_alu = code->alu.length;
+ emit->node_flags = 0;
+ return 1;
}
-static GLboolean emit_tex(void* data, struct radeon_pair_texture_instruction* inst)
+static int emit_tex(struct r300_emit_state * emit, struct rc_instruction * inst)
{
PROG_CODE;
if (code->tex.length >= R300_PFS_MAX_TEX_INST) {
error("Too many TEX instructions");
- return GL_FALSE;
+ return 0;
}
- GLuint unit = inst->TexSrcUnit;
- GLuint dest = inst->DestIndex;
- GLuint opcode;
+ unsigned int unit = inst->U.I.TexSrcUnit;
+ unsigned int dest = inst->U.I.DstReg.Index;
+ unsigned int opcode;
- switch(inst->Opcode) {
- case RADEON_OPCODE_KIL: opcode = R300_TEX_OP_KIL; break;
- case RADEON_OPCODE_TEX: opcode = R300_TEX_OP_LD; break;
- case RADEON_OPCODE_TXB: opcode = R300_TEX_OP_TXB; break;
- case RADEON_OPCODE_TXP: opcode = R300_TEX_OP_TXP; break;
+ switch(inst->U.I.Opcode) {
+ case RC_OPCODE_KIL: opcode = R300_TEX_OP_KIL; break;
+ case RC_OPCODE_TEX: opcode = R300_TEX_OP_LD; break;
+ case RC_OPCODE_TXB: opcode = R300_TEX_OP_TXB; break;
+ case RC_OPCODE_TXP: opcode = R300_TEX_OP_TXP; break;
default:
- error("Unknown texture opcode %i", inst->Opcode);
- return GL_FALSE;
+ error("Unknown texture opcode %i", inst->U.I.Opcode);
+ return 0;
}
- if (inst->Opcode == RADEON_OPCODE_KIL) {
+ if (inst->U.I.Opcode == RC_OPCODE_KIL) {
unit = 0;
dest = 0;
} else {
use_temporary(code, dest);
}
- use_temporary(code, inst->SrcIndex);
+ use_temporary(code, inst->U.I.SrcReg[0].Index);
- code->node[code->cur_node].tex_end++;
code->tex.inst[code->tex.length++] =
- (inst->SrcIndex << R300_SRC_ADDR_SHIFT) |
+ (inst->U.I.SrcReg[0].Index << R300_SRC_ADDR_SHIFT) |
(dest << R300_DST_ADDR_SHIFT) |
(unit << R300_TEX_ID_SHIFT) |
(opcode << R300_TEX_INST_SHIFT);
- return GL_TRUE;
+ return 1;
}
-static const struct radeon_pair_handler pair_handler = {
- .EmitPaired = &emit_alu,
- .EmitTex = &emit_tex,
- .BeginTexBlock = &begin_tex,
- .MaxHwTemps = R300_PFS_NUM_TEMP_REGS
-};
-
/**
* Final compilation step: Turn the intermediate radeon_program into
* machine-readable instructions.
*/
void r300BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compiler)
{
+ struct r300_emit_state emit;
struct r300_fragment_program_code *code = &compiler->code->code.r300;
- _mesa_bzero(code, sizeof(struct r300_fragment_program_code));
- code->node[0].alu_end = -1;
- code->node[0].tex_end = -1;
+ memset(&emit, 0, sizeof(emit));
+ emit.compiler = compiler;
+
+ memset(code, 0, sizeof(struct r300_fragment_program_code));
+
+ for(struct rc_instruction * inst = compiler->Base.Program.Instructions.Next;
+ inst != &compiler->Base.Program.Instructions && !compiler->Base.Error;
+ inst = inst->Next) {
+ if (inst->Type == RC_INSTRUCTION_NORMAL) {
+ if (inst->U.I.Opcode == RC_OPCODE_BEGIN_TEX) {
+ begin_tex(&emit);
+ continue;
+ }
+
+ emit_tex(&emit, inst);
+ } else {
+ emit_alu(&emit, &inst->U.P);
+ }
+ }
+
+ if (code->pixsize >= compiler->max_temp_regs)
+ rc_error(&compiler->Base, "Too many hardware temporaries used.\n");
- radeonPairProgram(&compiler->Base, &pair_handler, compiler);
if (compiler->Base.Error)
return;
- finish_node(compiler);
+ /* Finish the program */
+ finish_node(&emit);
+
+ code->config |= emit.current_node; /* FIRST_NODE_HAS_TEX set by finish_node */
+ code->code_offset =
+ (0 << R300_PFS_CNTL_ALU_OFFSET_SHIFT) |
+ ((code->alu.length-1) << R300_PFS_CNTL_ALU_END_SHIFT) |
+ (0 << R300_PFS_CNTL_TEX_OFFSET_SHIFT) |
+ ((code->tex.length ? code->tex.length-1 : 0) << R300_PFS_CNTL_TEX_END_SHIFT);
+
+ if (emit.current_node < 3) {
+ int shift = 3 - emit.current_node;
+ int i;
+ for(i = emit.current_node; i >= 0; --i)
+ code->code_addr[shift + i] = code->code_addr[i];
+ for(i = 0; i < shift; ++i)
+ code->code_addr[i] = 0;
+ }
}
-