#include "pipe/p_context.h"
#include "pipe/p_defines.h"
#include "pipe/p_state.h"
#include "util/u_inlines.h"
#include "util/u_debug.h"

#include "pipe/p_shader_tokens.h"
#include "tgsi/tgsi_parse.h"
#include "tgsi/tgsi_util.h"
#include "tgsi/tgsi_dump.h"

#include "nvfx_context.h"
#include "nvfx_shader.h"
#include "nvfx_resource.h"

#define MAX_CONSTS 128
#define MAX_IMM 32
struct nvfx_fpc {
	struct nvfx_fragment_program *fp;

	unsigned r_temps;
	unsigned r_temps_discard;
	struct nvfx_reg r_result[PIPE_MAX_SHADER_OUTPUTS];
	struct nvfx_reg *r_temp;

	int num_regs;

	unsigned inst_offset;
	unsigned have_const;

	struct {
		int pipe;
		float vals[4];
	} consts[MAX_CONSTS];
	int nr_consts;

	struct nvfx_reg imm[MAX_IMM];
	unsigned nr_imm;

	unsigned char generic_to_slot[256]; /* semantic idx for each input semantic */
};

static INLINE struct nvfx_reg
temp(struct nvfx_fpc *fpc)
{
	int idx = ffs(~fpc->r_temps) - 1;

	if (idx < 0) {
		NOUVEAU_ERR("out of temps!!\n");
		assert(0);
		return nvfx_reg(NVFXSR_TEMP, 0);
	}

	fpc->r_temps |= (1 << idx);
	fpc->r_temps_discard |= (1 << idx);
	return nvfx_reg(NVFXSR_TEMP, idx);
}

static INLINE void
release_temps(struct nvfx_fpc *fpc)
{
	fpc->r_temps &= ~fpc->r_temps_discard;
	fpc->r_temps_discard = 0;
}

static INLINE struct nvfx_reg
constant(struct nvfx_fpc *fpc, int pipe, float vals[4])
{
	int idx;

	if (fpc->nr_consts == MAX_CONSTS)
		assert(0);
	idx = fpc->nr_consts++;

	fpc->consts[idx].pipe = pipe;
	if (pipe == -1)
		memcpy(fpc->consts[idx].vals, vals, 4 * sizeof(float));
	return nvfx_reg(NVFXSR_CONST, idx);
}

static void
grow_insns(struct nvfx_fpc *fpc, int size)
{
	struct nvfx_fragment_program *fp = fpc->fp;

	fp->insn_len += size;
	fp->insn = realloc(fp->insn, sizeof(uint32_t) * fp->insn_len);
}

static void
emit_src(struct nvfx_fpc *fpc, int pos, struct nvfx_src src)
{
	struct nvfx_fragment_program *fp = fpc->fp;
	uint32_t *hw = &fp->insn[fpc->inst_offset];
	uint32_t sr = 0;

	switch (src.reg.type) {
	case NVFXSR_INPUT:
		sr |= (NVFX_FP_REG_TYPE_INPUT << NVFX_FP_REG_TYPE_SHIFT);
		hw[0] |= (src.reg.index << NVFX_FP_OP_INPUT_SRC_SHIFT);
		break;
	case NVFXSR_OUTPUT:
		sr |= NVFX_FP_REG_SRC_HALF;
		/* fall-through */
	case NVFXSR_TEMP:
		sr |= (NVFX_FP_REG_TYPE_TEMP << NVFX_FP_REG_TYPE_SHIFT);
		sr |= (src.reg.index << NVFX_FP_REG_SRC_SHIFT);
		break;
	case NVFXSR_RELOCATED:
		sr |= (NVFX_FP_REG_TYPE_INPUT << NVFX_FP_REG_TYPE_SHIFT);
		//printf("adding relocation at %x for %x\n", fpc->inst_offset, src.index);
		util_dynarray_append(&fpc->fp->slot_relocations[src.reg.index], unsigned, fpc->inst_offset);
		break;
	case NVFXSR_CONST:
		if (!fpc->have_const) {
			grow_insns(fpc, 4);
			fpc->have_const = 1;
		}

		hw = &fp->insn[fpc->inst_offset];
		if (fpc->consts[src.reg.index].pipe >= 0) {
			struct nvfx_fragment_program_data *fpd;

			fp->consts = realloc(fp->consts, ++fp->nr_consts *
					     sizeof(*fpd));
			fpd = &fp->consts[fp->nr_consts - 1];
			fpd->offset = fpc->inst_offset + 4;
			fpd->index = fpc->consts[src.reg.index].pipe;
			memset(&fp->insn[fpd->offset], 0, sizeof(uint32_t) * 4);
		} else {
			memcpy(&fp->insn[fpc->inst_offset + 4],
				fpc->consts[src.reg.index].vals,
				sizeof(uint32_t) * 4);
		}

		sr |= (NVFX_FP_REG_TYPE_CONST << NVFX_FP_REG_TYPE_SHIFT);
		break;
	case NVFXSR_NONE:
		sr |= (NVFX_FP_REG_TYPE_INPUT << NVFX_FP_REG_TYPE_SHIFT);
		break;
	default:
		assert(0);
	}

	if (src.negate)
		sr |= NVFX_FP_REG_NEGATE;

	if (src.abs)
		hw[1] |= (1 << (29 + pos));

	sr |= ((src.swz[0] << NVFX_FP_REG_SWZ_X_SHIFT) |
	       (src.swz[1] << NVFX_FP_REG_SWZ_Y_SHIFT) |
	       (src.swz[2] << NVFX_FP_REG_SWZ_Z_SHIFT) |
	       (src.swz[3] << NVFX_FP_REG_SWZ_W_SHIFT));

	hw[pos + 1] |= sr;
}

static void
emit_dst(struct nvfx_fpc *fpc, struct nvfx_reg dst)
{
	struct nvfx_fragment_program *fp = fpc->fp;
	uint32_t *hw = &fp->insn[fpc->inst_offset];

	switch (dst.type) {
	case NVFXSR_TEMP:
		if (fpc->num_regs < (dst.index + 1))
			fpc->num_regs = dst.index + 1;
		break;
	case NVFXSR_OUTPUT:
		if (dst.index == 1) {
			fp->fp_control |= 0xe;
		} else {
			hw[0] |= NVFX_FP_OP_OUT_REG_HALF;
		}
		break;
	case NVFXSR_NONE:
		hw[0] |= (1 << 30);
		break;
	default:
		assert(0);
	}

	hw[0] |= (dst.index << NVFX_FP_OP_OUT_REG_SHIFT);
}

static void
nvfx_fp_emit(struct nvfx_fpc *fpc, struct nvfx_insn insn)
{
	struct nvfx_fragment_program *fp = fpc->fp;
	uint32_t *hw;

	fpc->inst_offset = fp->insn_len;
	fpc->have_const = 0;
	grow_insns(fpc, 4);
	hw = &fp->insn[fpc->inst_offset];
	memset(hw, 0, sizeof(uint32_t) * 4);

	if (insn.op == NVFX_FP_OP_OPCODE_KIL)
		fp->fp_control |= NV34TCL_FP_CONTROL_USES_KIL;
	hw[0] |= (insn.op << NVFX_FP_OP_OPCODE_SHIFT);
	hw[0] |= (insn.mask << NVFX_FP_OP_OUTMASK_SHIFT);
	hw[2] |= (insn.scale << NVFX_FP_OP_DST_SCALE_SHIFT);

	if (insn.sat)
		hw[0] |= NVFX_FP_OP_OUT_SAT;

	if (insn.cc_update)
		hw[0] |= NVFX_FP_OP_COND_WRITE_ENABLE;
	hw[1] |= (insn.cc_test << NVFX_FP_OP_COND_SHIFT);
	hw[1] |= ((insn.cc_swz[0] << NVFX_FP_OP_COND_SWZ_X_SHIFT) |
		  (insn.cc_swz[1] << NVFX_FP_OP_COND_SWZ_Y_SHIFT) |
		  (insn.cc_swz[2] << NVFX_FP_OP_COND_SWZ_Z_SHIFT) |
		  (insn.cc_swz[3] << NVFX_FP_OP_COND_SWZ_W_SHIFT));

	if(insn.unit >= 0)
	{
		hw[0] |= (insn.unit << NVFX_FP_OP_TEX_UNIT_SHIFT);
		fp->samplers |= (1 << insn.unit);
	}

	emit_dst(fpc, insn.dst);
	emit_src(fpc, 0, insn.src[0]);
	emit_src(fpc, 1, insn.src[1]);
	emit_src(fpc, 2, insn.src[2]);
}

#define arith(s,o,d,m,s0,s1,s2) \
       nvfx_insn((s), NVFX_FP_OP_OPCODE_##o, -1, \
                       (d), (m), (s0), (s1), (s2))

#define tex(s,o,u,d,m,s0,s1,s2) \
	nvfx_insn((s), NVFX_FP_OP_OPCODE_##o, (u), \
                   (d), (m), (s0), none, none)

static INLINE struct nvfx_src
tgsi_src(struct nvfx_fpc *fpc, const struct tgsi_full_src_register *fsrc)
{
	struct nvfx_src src;

	switch (fsrc->Register.File) {
	case TGSI_FILE_INPUT:
		if(fpc->fp->info.input_semantic_name[fsrc->Register.Index] == TGSI_SEMANTIC_POSITION) {
			assert(fpc->fp->info.input_semantic_index[fsrc->Register.Index] == 0);
			src.reg = nvfx_reg(NVFXSR_INPUT, NVFX_FP_OP_INPUT_SRC_POSITION);
		} else if(fpc->fp->info.input_semantic_name[fsrc->Register.Index] == TGSI_SEMANTIC_COLOR) {
			if(fpc->fp->info.input_semantic_index[fsrc->Register.Index] == 0)
				src.reg = nvfx_reg(NVFXSR_INPUT, NVFX_FP_OP_INPUT_SRC_COL0);
			else if(fpc->fp->info.input_semantic_index[fsrc->Register.Index] == 1)
				src.reg = nvfx_reg(NVFXSR_INPUT, NVFX_FP_OP_INPUT_SRC_COL1);
			else
				assert(0);
		} else if(fpc->fp->info.input_semantic_name[fsrc->Register.Index] == TGSI_SEMANTIC_FOG) {
			assert(fpc->fp->info.input_semantic_index[fsrc->Register.Index] == 0);
			src.reg = nvfx_reg(NVFXSR_INPUT, NVFX_FP_OP_INPUT_SRC_FOGC);
		} else if(fpc->fp->info.input_semantic_name[fsrc->Register.Index] == TGSI_SEMANTIC_FACE) {
			/* TODO: check this has the correct values */
			/* XXX: what do we do for nv30 here (assuming it lacks facing)?!  */
			assert(fpc->fp->info.input_semantic_index[fsrc->Register.Index] == 0);
			src.reg = nvfx_reg(NVFXSR_INPUT, NV40_FP_OP_INPUT_SRC_FACING);
		} else {
			assert(fpc->fp->info.input_semantic_name[fsrc->Register.Index] == TGSI_SEMANTIC_GENERIC);
			src.reg = nvfx_reg(NVFXSR_RELOCATED, fpc->generic_to_slot[fpc->fp->info.input_semantic_index[fsrc->Register.Index]]);
		}
		break;
	case TGSI_FILE_CONSTANT:
		src.reg = constant(fpc, fsrc->Register.Index, NULL);
		break;
	case TGSI_FILE_IMMEDIATE:
		assert(fsrc->Register.Index < fpc->nr_imm);
		src.reg = fpc->imm[fsrc->Register.Index];
		break;
	case TGSI_FILE_TEMPORARY:
		src.reg = fpc->r_temp[fsrc->Register.Index];
		break;
	/* NV40 fragprog result regs are just temps, so this is simple */
	case TGSI_FILE_OUTPUT:
		src.reg = fpc->r_result[fsrc->Register.Index];
		break;
	default:
		NOUVEAU_ERR("bad src file\n");
		break;
	}

	src.abs = fsrc->Register.Absolute;
	src.negate = fsrc->Register.Negate;
	src.swz[0] = fsrc->Register.SwizzleX;
	src.swz[1] = fsrc->Register.SwizzleY;
	src.swz[2] = fsrc->Register.SwizzleZ;
	src.swz[3] = fsrc->Register.SwizzleW;
	return src;
}

static INLINE struct nvfx_reg
tgsi_dst(struct nvfx_fpc *fpc, const struct tgsi_full_dst_register *fdst) {
	switch (fdst->Register.File) {
	case TGSI_FILE_OUTPUT:
		return fpc->r_result[fdst->Register.Index];
	case TGSI_FILE_TEMPORARY:
		return fpc->r_temp[fdst->Register.Index];
	case TGSI_FILE_NULL:
		return nvfx_reg(NVFXSR_NONE, 0);
	default:
		NOUVEAU_ERR("bad dst file %d\n", fdst->Register.File);
		return nvfx_reg(NVFXSR_NONE, 0);
	}
}

static INLINE int
tgsi_mask(uint tgsi)
{
	int mask = 0;

	if (tgsi & TGSI_WRITEMASK_X) mask |= NVFX_FP_MASK_X;
	if (tgsi & TGSI_WRITEMASK_Y) mask |= NVFX_FP_MASK_Y;
	if (tgsi & TGSI_WRITEMASK_Z) mask |= NVFX_FP_MASK_Z;
	if (tgsi & TGSI_WRITEMASK_W) mask |= NVFX_FP_MASK_W;
	return mask;
}

static boolean
nvfx_fragprog_parse_instruction(struct nvfx_context* nvfx, struct nvfx_fpc *fpc,
				const struct tgsi_full_instruction *finst)
{
	const struct nvfx_src none = nvfx_src(nvfx_reg(NVFXSR_NONE, 0));
	struct nvfx_insn insn;
	struct nvfx_src src[3], tmp;
	struct nvfx_reg dst;
	int mask, sat, unit = 0;
	int ai = -1, ci = -1, ii = -1;
	int i;

	if (finst->Instruction.Opcode == TGSI_OPCODE_END)
		return TRUE;

	for (i = 0; i < finst->Instruction.NumSrcRegs; i++) {
		const struct tgsi_full_src_register *fsrc;

		fsrc = &finst->Src[i];
		if (fsrc->Register.File == TGSI_FILE_TEMPORARY) {
			src[i] = tgsi_src(fpc, fsrc);
		}
	}

	for (i = 0; i < finst->Instruction.NumSrcRegs; i++) {
		const struct tgsi_full_src_register *fsrc;

		fsrc = &finst->Src[i];

		switch (fsrc->Register.File) {
		case TGSI_FILE_INPUT:
			if (ai == -1 || ai == fsrc->Register.Index) {
				ai = fsrc->Register.Index;
				src[i] = tgsi_src(fpc, fsrc);
			} else {
				src[i] = nvfx_src(temp(fpc));
				nvfx_fp_emit(fpc, arith(0, MOV, src[i].reg, NVFX_FP_MASK_ALL, tgsi_src(fpc, fsrc), none, none));
			}
			break;
		case TGSI_FILE_CONSTANT:
			if ((ci == -1 && ii == -1) ||
			    ci == fsrc->Register.Index) {
				ci = fsrc->Register.Index;
				src[i] = tgsi_src(fpc, fsrc);
			} else {
				src[i] = nvfx_src(temp(fpc));
				nvfx_fp_emit(fpc, arith(0, MOV, src[i].reg, NVFX_FP_MASK_ALL, tgsi_src(fpc, fsrc), none, none));
			}
			break;
		case TGSI_FILE_IMMEDIATE:
			if ((ci == -1 && ii == -1) ||
			    ii == fsrc->Register.Index) {
				ii = fsrc->Register.Index;
				src[i] = tgsi_src(fpc, fsrc);
			} else {
				src[i] = nvfx_src(temp(fpc));
				nvfx_fp_emit(fpc, arith(0, MOV, src[i].reg, NVFX_FP_MASK_ALL, tgsi_src(fpc, fsrc), none, none));
			}
			break;
		case TGSI_FILE_TEMPORARY:
			/* handled above */
			break;
		case TGSI_FILE_SAMPLER:
			unit = fsrc->Register.Index;
			break;
		case TGSI_FILE_OUTPUT:
			break;
		default:
			NOUVEAU_ERR("bad src file\n");
			return FALSE;
		}
	}

	dst  = tgsi_dst(fpc, &finst->Dst[0]);
	mask = tgsi_mask(finst->Dst[0].Register.WriteMask);
	sat  = (finst->Instruction.Saturate == TGSI_SAT_ZERO_ONE);

	switch (finst->Instruction.Opcode) {
	case TGSI_OPCODE_ABS:
		nvfx_fp_emit(fpc, arith(sat, MOV, dst, mask, abs(src[0]), none, none));
		break;
	case TGSI_OPCODE_ADD:
		nvfx_fp_emit(fpc, arith(sat, ADD, dst, mask, src[0], src[1], none));
		break;
	case TGSI_OPCODE_CMP:
		insn = arith(0, MOV, none.reg, 0xf, src[0], none, none);
		insn.cc_update = 1;
		nvfx_fp_emit(fpc, insn);

		insn = arith(sat, MOV, dst, mask, src[2], none, none);
		insn.cc_test = NVFX_COND_GE;
		nvfx_fp_emit(fpc, insn);

		insn = arith(sat, MOV, dst, mask, src[1], none, none);
		insn.cc_test = NVFX_COND_LT;
		nvfx_fp_emit(fpc, insn);
		break;
	case TGSI_OPCODE_COS:
		nvfx_fp_emit(fpc, arith(sat, COS, dst, mask, src[0], none, none));
		break;
	case TGSI_OPCODE_DDX:
		if (mask & (NVFX_FP_MASK_Z | NVFX_FP_MASK_W)) {
			tmp = nvfx_src(temp(fpc));
			nvfx_fp_emit(fpc, arith(sat, DDX, tmp.reg, NVFX_FP_MASK_X | NVFX_FP_MASK_Y, swz(src[0], Z, W, Z, W), none, none));
			nvfx_fp_emit(fpc, arith(0, MOV, tmp.reg, NVFX_FP_MASK_Z | NVFX_FP_MASK_W, swz(tmp, X, Y, X, Y), none, none));
			nvfx_fp_emit(fpc, arith(sat, DDX, tmp.reg, NVFX_FP_MASK_X | NVFX_FP_MASK_Y, src[0], none, none));
			nvfx_fp_emit(fpc, arith(0, MOV, dst, mask, tmp, none, none));
		} else {
			nvfx_fp_emit(fpc, arith(sat, DDX, dst, mask, src[0], none, none));
		}
		break;
	case TGSI_OPCODE_DDY:
		if (mask & (NVFX_FP_MASK_Z | NVFX_FP_MASK_W)) {
			tmp = nvfx_src(temp(fpc));
			nvfx_fp_emit(fpc, arith(sat, DDY, tmp.reg, NVFX_FP_MASK_X | NVFX_FP_MASK_Y, swz(src[0], Z, W, Z, W), none, none));
			nvfx_fp_emit(fpc, arith(0, MOV, tmp.reg, NVFX_FP_MASK_Z | NVFX_FP_MASK_W, swz(tmp, X, Y, X, Y), none, none));
			nvfx_fp_emit(fpc, arith(sat, DDY, tmp.reg, NVFX_FP_MASK_X | NVFX_FP_MASK_Y, src[0], none, none));
			nvfx_fp_emit(fpc, arith(0, MOV, dst, mask, tmp, none, none));
		} else {
			nvfx_fp_emit(fpc, arith(sat, DDY, dst, mask, src[0], none, none));
		}
		break;
	case TGSI_OPCODE_DP3:
		nvfx_fp_emit(fpc, arith(sat, DP3, dst, mask, src[0], src[1], none));
		break;
	case TGSI_OPCODE_DP4:
		nvfx_fp_emit(fpc, arith(sat, DP4, dst, mask, src[0], src[1], none));
		break;
	case TGSI_OPCODE_DPH:
		tmp = nvfx_src(temp(fpc));
		nvfx_fp_emit(fpc, arith(0, DP3, tmp.reg, NVFX_FP_MASK_X, src[0], src[1], none));
		nvfx_fp_emit(fpc, arith(sat, ADD, dst, mask, swz(tmp, X, X, X, X), swz(src[1], W, W, W, W), none));
		break;
	case TGSI_OPCODE_DST:
		nvfx_fp_emit(fpc, arith(sat, DST, dst, mask, src[0], src[1], none));
		break;
	case TGSI_OPCODE_EX2:
		nvfx_fp_emit(fpc, arith(sat, EX2, dst, mask, src[0], none, none));
		break;
	case TGSI_OPCODE_FLR:
		nvfx_fp_emit(fpc, arith(sat, FLR, dst, mask, src[0], none, none));
		break;
	case TGSI_OPCODE_FRC:
		nvfx_fp_emit(fpc, arith(sat, FRC, dst, mask, src[0], none, none));
		break;
	case TGSI_OPCODE_KILP:
		nvfx_fp_emit(fpc, arith(0, KIL, none.reg, 0, none, none, none));
		break;
	case TGSI_OPCODE_KIL:
		insn = arith(0, MOV, none.reg, NVFX_FP_MASK_ALL, src[0], none, none);
		insn.cc_update = 1;
		nvfx_fp_emit(fpc, insn);

		insn = arith(0, KIL, none.reg, 0, none, none, none);
		insn.cc_test = NVFX_COND_LT;
		nvfx_fp_emit(fpc, insn);
		break;
	case TGSI_OPCODE_LG2:
		nvfx_fp_emit(fpc, arith(sat, LG2, dst, mask, src[0], none, none));
		break;
//	case TGSI_OPCODE_LIT:
	case TGSI_OPCODE_LRP:
		if(!nvfx->is_nv4x)
			nvfx_fp_emit(fpc, arith(sat, LRP_NV30, dst, mask, src[0], src[1], src[2]));
		else {
			tmp = nvfx_src(temp(fpc));
			nvfx_fp_emit(fpc, arith(0, MAD, tmp.reg, mask, neg(src[0]), src[2], src[2]));
			nvfx_fp_emit(fpc, arith(sat, MAD, dst, mask, src[0], src[1], tmp));
		}
		break;
	case TGSI_OPCODE_MAD:
		nvfx_fp_emit(fpc, arith(sat, MAD, dst, mask, src[0], src[1], src[2]));
		break;
	case TGSI_OPCODE_MAX:
		nvfx_fp_emit(fpc, arith(sat, MAX, dst, mask, src[0], src[1], none));
		break;
	case TGSI_OPCODE_MIN:
		nvfx_fp_emit(fpc, arith(sat, MIN, dst, mask, src[0], src[1], none));
		break;
	case TGSI_OPCODE_MOV:
		nvfx_fp_emit(fpc, arith(sat, MOV, dst, mask, src[0], none, none));
		break;
	case TGSI_OPCODE_MUL:
		nvfx_fp_emit(fpc, arith(sat, MUL, dst, mask, src[0], src[1], none));
		break;
	case TGSI_OPCODE_POW:
		if(!nvfx->is_nv4x)
			nvfx_fp_emit(fpc, arith(sat, POW_NV30, dst, mask, src[0], src[1], none));
		else {
			tmp = nvfx_src(temp(fpc));
			nvfx_fp_emit(fpc, arith(0, LG2, tmp.reg, NVFX_FP_MASK_X, swz(src[0], X, X, X, X), none, none));
			nvfx_fp_emit(fpc, arith(0, MUL, tmp.reg, NVFX_FP_MASK_X, swz(tmp, X, X, X, X), swz(src[1], X, X, X, X), none));
			nvfx_fp_emit(fpc, arith(sat, EX2, dst, mask, swz(tmp, X, X, X, X), none, none));
		}
		break;
	case TGSI_OPCODE_RCP:
		nvfx_fp_emit(fpc, arith(sat, RCP, dst, mask, src[0], none, none));
		break;
	case TGSI_OPCODE_RET:
		assert(0);
		break;
	case TGSI_OPCODE_RFL:
		if(!nvfx->is_nv4x)
			nvfx_fp_emit(fpc, arith(0, RFL_NV30, dst, mask, src[0], src[1], none));
		else {
			tmp = nvfx_src(temp(fpc));
			nvfx_fp_emit(fpc, arith(0, DP3, tmp.reg, NVFX_FP_MASK_X, src[0], src[0], none));
			nvfx_fp_emit(fpc, arith(0, DP3, tmp.reg, NVFX_FP_MASK_Y, src[0], src[1], none));
			insn = arith(0, DIV, tmp.reg, NVFX_FP_MASK_Z, swz(tmp, Y, Y, Y, Y), swz(tmp, X, X, X, X), none);
			insn.scale = NVFX_FP_OP_DST_SCALE_2X;
			nvfx_fp_emit(fpc, insn);
			nvfx_fp_emit(fpc, arith(sat, MAD, dst, mask, swz(tmp, Z, Z, Z, Z), src[0], neg(src[1])));
		}
		break;
	case TGSI_OPCODE_RSQ:
		if(!nvfx->is_nv4x)
			nvfx_fp_emit(fpc, arith(sat, RSQ_NV30, dst, mask, abs(swz(src[0], X, X, X, X)), none, none));
		else {
			tmp = nvfx_src(temp(fpc));
			insn = arith(0, LG2, tmp.reg, NVFX_FP_MASK_X, abs(swz(src[0], X, X, X, X)), none, none);
			insn.scale = NVFX_FP_OP_DST_SCALE_INV_2X;
			nvfx_fp_emit(fpc, insn);
			nvfx_fp_emit(fpc, arith(sat, EX2, dst, mask, neg(swz(tmp, X, X, X, X)), none, none));
		}
		break;
	case TGSI_OPCODE_SCS:
		/* avoid overwriting the source */
		if(src[0].swz[NVFX_SWZ_X] != NVFX_SWZ_X)
		{
			if (mask & NVFX_FP_MASK_X)
				nvfx_fp_emit(fpc, arith(sat, COS, dst, NVFX_FP_MASK_X, swz(src[0], X, X, X, X), none, none));
			if (mask & NVFX_FP_MASK_Y)
				nvfx_fp_emit(fpc, arith(sat, SIN, dst, NVFX_FP_MASK_Y, swz(src[0], X, X, X, X), none, none));
		}
		else
		{
			if (mask & NVFX_FP_MASK_Y)
				nvfx_fp_emit(fpc, arith(sat, SIN, dst, NVFX_FP_MASK_Y, swz(src[0], X, X, X, X), none, none));
			if (mask & NVFX_FP_MASK_X)
				nvfx_fp_emit(fpc, arith(sat, COS, dst, NVFX_FP_MASK_X, swz(src[0], X, X, X, X), none, none));
		}
		break;
	case TGSI_OPCODE_SEQ:
		nvfx_fp_emit(fpc, arith(sat, SEQ, dst, mask, src[0], src[1], none));
		break;
	case TGSI_OPCODE_SFL:
		nvfx_fp_emit(fpc, arith(sat, SFL, dst, mask, src[0], src[1], none));
		break;
	case TGSI_OPCODE_SGE:
		nvfx_fp_emit(fpc, arith(sat, SGE, dst, mask, src[0], src[1], none));
		break;
	case TGSI_OPCODE_SGT:
		nvfx_fp_emit(fpc, arith(sat, SGT, dst, mask, src[0], src[1], none));
		break;
	case TGSI_OPCODE_SIN:
		nvfx_fp_emit(fpc, arith(sat, SIN, dst, mask, src[0], none, none));
		break;
	case TGSI_OPCODE_SLE:
		nvfx_fp_emit(fpc, arith(sat, SLE, dst, mask, src[0], src[1], none));
		break;
	case TGSI_OPCODE_SLT:
		nvfx_fp_emit(fpc, arith(sat, SLT, dst, mask, src[0], src[1], none));
		break;
	case TGSI_OPCODE_SNE:
		nvfx_fp_emit(fpc, arith(sat, SNE, dst, mask, src[0], src[1], none));
		break;
	case TGSI_OPCODE_STR:
		nvfx_fp_emit(fpc, arith(sat, STR, dst, mask, src[0], src[1], none));
		break;
	case TGSI_OPCODE_SUB:
		nvfx_fp_emit(fpc, arith(sat, ADD, dst, mask, src[0], neg(src[1]), none));
		break;
	case TGSI_OPCODE_TEX:
		nvfx_fp_emit(fpc, tex(sat, TEX, unit, dst, mask, src[0], none, none));
		break;
	case TGSI_OPCODE_TXB:
		nvfx_fp_emit(fpc, tex(sat, TXB, unit, dst, mask, src[0], none, none));
		break;
	case TGSI_OPCODE_TXP:
		nvfx_fp_emit(fpc, tex(sat, TXP, unit, dst, mask, src[0], none, none));
		break;
	case TGSI_OPCODE_XPD:
		tmp = nvfx_src(temp(fpc));
		nvfx_fp_emit(fpc, arith(0, MUL, tmp.reg, mask, swz(src[0], Z, X, Y, Y), swz(src[1], Y, Z, X, X), none));
		nvfx_fp_emit(fpc, arith(sat, MAD, dst, (mask & ~NVFX_FP_MASK_W), swz(src[0], Y, Z, X, X), swz(src[1], Z, X, Y, Y), neg(tmp)));
		break;
	default:
		NOUVEAU_ERR("invalid opcode %d\n", finst->Instruction.Opcode);
		return FALSE;
	}

	release_temps(fpc);
	return TRUE;
}

static boolean
nvfx_fragprog_parse_decl_output(struct nvfx_context* nvfx, struct nvfx_fpc *fpc,
				const struct tgsi_full_declaration *fdec)
{
	unsigned idx = fdec->Range.First;
	unsigned hw;

	switch (fdec->Semantic.Name) {
	case TGSI_SEMANTIC_POSITION:
		hw = 1;
		break;
	case TGSI_SEMANTIC_COLOR:
		hw = ~0;
		switch (fdec->Semantic.Index) {
		case 0: hw = 0; break;
		case 1: hw = 2; break;
		case 2: hw = 3; break;
		case 3: hw = 4; break;
		}
		if(hw > ((nvfx->is_nv4x) ? 4 : 2)) {
			NOUVEAU_ERR("bad rcol index\n");
			return FALSE;
		}
		break;
	default:
		NOUVEAU_ERR("bad output semantic\n");
		return FALSE;
	}

	fpc->r_result[idx] = nvfx_reg(NVFXSR_OUTPUT, hw);
	fpc->r_temps |= (1 << hw);
	return TRUE;
}

static boolean
nvfx_fragprog_prepare(struct nvfx_context* nvfx, struct nvfx_fpc *fpc)
{
	struct tgsi_parse_context p;
	int high_temp = -1, i;
	struct util_semantic_set set;

	fpc->fp->num_slots = util_semantic_set_from_program_file(&set, fpc->fp->pipe.tokens, TGSI_FILE_INPUT);
	if(fpc->fp->num_slots > 8)
		return FALSE;
	util_semantic_layout_from_set(fpc->fp->slot_to_generic, &set, 0, 8);
	util_semantic_table_from_layout(fpc->generic_to_slot, fpc->fp->slot_to_generic, 0, 8);

	memset(fpc->fp->slot_to_fp_input, 0xff, sizeof(fpc->fp->slot_to_fp_input));

	tgsi_parse_init(&p, fpc->fp->pipe.tokens);
	while (!tgsi_parse_end_of_tokens(&p)) {
		const union tgsi_full_token *tok = &p.FullToken;

		tgsi_parse_token(&p);
		switch(tok->Token.Type) {
		case TGSI_TOKEN_TYPE_DECLARATION:
		{
			const struct tgsi_full_declaration *fdec;
			fdec = &p.FullToken.FullDeclaration;
			switch (fdec->Declaration.File) {
			case TGSI_FILE_OUTPUT:
				if (!nvfx_fragprog_parse_decl_output(nvfx, fpc, fdec))
					goto out_err;
				break;
			case TGSI_FILE_TEMPORARY:
				if (fdec->Range.Last > high_temp) {
					high_temp =
						fdec->Range.Last;
				}
				break;
			default:
				break;
			}
		}
			break;
		case TGSI_TOKEN_TYPE_IMMEDIATE:
		{
			struct tgsi_full_immediate *imm;
			float vals[4];

			imm = &p.FullToken.FullImmediate;
			assert(imm->Immediate.DataType == TGSI_IMM_FLOAT32);
			assert(fpc->nr_imm < MAX_IMM);

			vals[0] = imm->u[0].Float;
			vals[1] = imm->u[1].Float;
			vals[2] = imm->u[2].Float;
			vals[3] = imm->u[3].Float;
			fpc->imm[fpc->nr_imm++] = constant(fpc, -1, vals);
		}
			break;
		default:
			break;
		}
	}
	tgsi_parse_free(&p);

	if (++high_temp) {
		fpc->r_temp = CALLOC(high_temp, sizeof(struct nvfx_reg));
		for (i = 0; i < high_temp; i++)
			fpc->r_temp[i] = temp(fpc);
		fpc->r_temps_discard = 0;
	}

	return TRUE;

out_err:
	if (fpc->r_temp)
		FREE(fpc->r_temp);
	tgsi_parse_free(&p);
	return FALSE;
}

DEBUG_GET_ONCE_BOOL_OPTION(nvfx_dump_fp, "NVFX_DUMP_FP", FALSE)

static void
nvfx_fragprog_translate(struct nvfx_context *nvfx,
			struct nvfx_fragment_program *fp)
{
	struct tgsi_parse_context parse;
	struct nvfx_fpc *fpc = NULL;

	fpc = CALLOC(1, sizeof(struct nvfx_fpc));
	if (!fpc)
		return;
	fpc->fp = fp;
	fpc->num_regs = 2;

	if (!nvfx_fragprog_prepare(nvfx, fpc)) {
		FREE(fpc);
		return;
	}

	tgsi_parse_init(&parse, fp->pipe.tokens);

	while (!tgsi_parse_end_of_tokens(&parse)) {
		tgsi_parse_token(&parse);

		switch (parse.FullToken.Token.Type) {
		case TGSI_TOKEN_TYPE_INSTRUCTION:
		{
			const struct tgsi_full_instruction *finst;

			finst = &parse.FullToken.FullInstruction;
			if (!nvfx_fragprog_parse_instruction(nvfx, fpc, finst))
				goto out_err;
		}
			break;
		default:
			break;
		}
	}

	if(!nvfx->is_nv4x)
		fp->fp_control |= (fpc->num_regs-1)/2;
	else
		fp->fp_control |= fpc->num_regs << NV40TCL_FP_CONTROL_TEMP_COUNT_SHIFT;

	/* Terminate final instruction */
	if(fp->insn)
		fp->insn[fpc->inst_offset] |= 0x00000001;

	/* Append NOP + END instruction, may or may not be necessary. */
	fpc->inst_offset = fp->insn_len;
	grow_insns(fpc, 4);
	fp->insn[fpc->inst_offset + 0] = 0x00000001;
	fp->insn[fpc->inst_offset + 1] = 0x00000000;
	fp->insn[fpc->inst_offset + 2] = 0x00000000;
	fp->insn[fpc->inst_offset + 3] = 0x00000000;

	if(debug_get_option_nvfx_dump_fp())
	{
		debug_printf("\n");
		tgsi_dump(fp->pipe.tokens, 0);

		debug_printf("\n%s fragment program:\n", nvfx->is_nv4x ? "nv4x" : "nv3x");
		for (unsigned i = 0; i < fp->insn_len; i += 4)
			debug_printf("%3u: %08x %08x %08x %08x\n", i >> 2, fp->insn[i], fp->insn[i + 1], fp->insn[i + 2], fp->insn[i + 3]);
		debug_printf("\n");
	}

	fp->translated = TRUE;
out_err:
	tgsi_parse_free(&parse);
	if (fpc->r_temp)
		FREE(fpc->r_temp);
	FREE(fpc);
}

static inline void
nvfx_fp_memcpy(void* dst, const void* src, size_t len)
{
#ifndef WORDS_BIGENDIAN
	memcpy(dst, src, len);
#else
	size_t i;
	for(i = 0; i < len; i += 4) {
		uint32_t v = (uint32_t*)((char*)src + i);
		*(uint32_t*)((char*)dst + i) = (v >> 16) | (v << 16);
	}
#endif
}

void
nvfx_fragprog_validate(struct nvfx_context *nvfx)
{
	struct nouveau_channel* chan = nvfx->screen->base.channel;
	struct nvfx_fragment_program *fp = nvfx->fragprog;
	int update = 0;

	if (!fp->translated)
	{
		const int min_size = 4096;

		nvfx_fragprog_translate(nvfx, fp);
		if (!fp->translated) {
			static unsigned dummy[8] = {1, 0, 0, 0, 1, 0, 0, 0};
			static int warned = 0;
			if(!warned)
			{
				fprintf(stderr, "nvfx: failed to translate fragment program!\n");
				warned = 1;
			}

			/* use dummy program: we cannot fail here */
			fp->translated = TRUE;
			fp->insn = malloc(sizeof(dummy));
			memcpy(fp->insn, dummy, sizeof(dummy));
			fp->insn_len = sizeof(dummy) / sizeof(dummy[0]);
		}
		update = TRUE;

		fp->prog_size = (fp->insn_len * 4 + 63) & ~63;

		if(fp->prog_size >= min_size)
			fp->progs_per_bo = 1;
		else
			fp->progs_per_bo = min_size / fp->prog_size;
		fp->bo_prog_idx = fp->progs_per_bo - 1;
	}

	/* we must update constants even on "just" fragprog changes, because
	   we don't check whether the current constant buffer matches the latest
	   one bound to this fragment program */
	if (nvfx->dirty & (NVFX_NEW_FRAGCONST | NVFX_NEW_FRAGPROG))
		update = TRUE;

	struct nvfx_vertex_program* vp = nvfx->render_mode == HW ? nvfx->vertprog : nvfx->swtnl.vertprog;
	if (fp->last_vp_id != vp->id) {
		char* vp_sem_table = vp->generic_to_fp_input;
		unsigned char* fp_semantics = fp->slot_to_generic;
		unsigned diff = 0;
		fp->last_vp_id = nvfx->vertprog->id;
		unsigned char* cur_slots = fp->slot_to_fp_input;
		for(unsigned i = 0; i < fp->num_slots; ++i) {
			unsigned char slot_mask = vp_sem_table[fp_semantics[i]];
			diff |= (slot_mask >> 4) & (slot_mask ^ cur_slots[i]);
		}

		if(diff)
		{
			for(unsigned i = 0; i < fp->num_slots; ++i) {
				/* if 0xff, then this will write to the dummy value at fp->last_layout_mask[0] */
				fp->slot_to_fp_input[i] = vp_sem_table[fp_semantics[i]] & 0xf;
				//printf("fp: GENERIC[%i] from fpreg %i\n", fp_semantics[i], fp->slot_to_fp_input[i]);
			}

			fp->progs_left_with_obsolete_slot_assignments = fp->progs;
			update = TRUE;
		}
	}

	// last_sprite_coord_enable
	unsigned sprite_coord_enable = nvfx->rasterizer->pipe.point_quad_rasterization * nvfx->rasterizer->pipe.sprite_coord_enable;
	if(fp->last_sprite_coord_enable != sprite_coord_enable)
	{
		unsigned texcoord_mask = vp->texcoord_ouput_mask;
		fp->last_sprite_coord_enable = sprite_coord_enable;
		fp->point_sprite_control = 0;
		for(unsigned i = 0; i < fp->num_slots; ++i) {
			if((1 << fp->slot_to_generic[i]) & sprite_coord_enable)
			{
				unsigned fpin = fp->slot_to_fp_input[i];
				//printf("sprite: slot %i generic %i had texcoord %i\n", i, fp->slot_to_generic[i], fpin - NVFX_FP_OP_INPUT_SRC_TC0);
				if(fpin >= 0x0f)
				{
					unsigned tc = __builtin_ctz(~texcoord_mask);
					texcoord_mask |= (1 << tc);
					fp->slot_to_fp_input[i] = fpin = NVFX_FP_OP_INPUT_SRC_TC(tc);

					fp->progs_left_with_obsolete_slot_assignments = fp->progs;
					update = TRUE;
				}
				//printf("sprite: slot %i texcoord %i\n", i, fpin - NVFX_FP_OP_INPUT_SRC_TC0);
				fp->point_sprite_control |= (1 << (fpin - NVFX_FP_OP_INPUT_SRC_TC0 + 8));
			}
			else
			{
				unsigned fpin = fp->slot_to_fp_input[i];
				if(!(vp->texcoord_ouput_mask & (1 << (fpin - NVFX_FP_OP_INPUT_SRC_TC0))))
				{
					fp->slot_to_fp_input[i] = 0x0f;

					fp->progs_left_with_obsolete_slot_assignments = fp->progs;
					update = TRUE;
				}
			}
		}
	}

	if(update) {
		++fp->bo_prog_idx;
		if(fp->bo_prog_idx >= fp->progs_per_bo)
		{
			if(fp->fpbo && !nouveau_bo_busy(fp->fpbo->next->bo, NOUVEAU_BO_WR))
			{
				fp->fpbo = fp->fpbo->next;
			}
			else
			{
				struct nvfx_fragment_program_bo* fpbo = os_malloc_aligned(sizeof(struct nvfx_fragment_program) + (fp->prog_size + 8) * fp->progs_per_bo, 16);
				fpbo->slots = (unsigned char*)&fpbo->insn[(fp->prog_size) * fp->progs_per_bo];
				memset(fpbo->slots, 0, 8 * fp->progs_per_bo);
				if(fp->fpbo)
				{
					fpbo->next = fp->fpbo->next;
					fp->fpbo->next = fpbo;
				}
				else
					fpbo->next = fpbo;
				fp->fpbo = fpbo;
				fpbo->bo = 0;
				fp->progs += fp->progs_per_bo;
				fp->progs_left_with_obsolete_slot_assignments += fp->progs_per_bo;
				nouveau_bo_new(nvfx->screen->base.device, NOUVEAU_BO_VRAM | NOUVEAU_BO_MAP, 64, fp->prog_size * fp->progs_per_bo, &fpbo->bo);
				nouveau_bo_map(fpbo->bo, NOUVEAU_BO_NOSYNC);

				uint8_t* map = fpbo->bo->map;
				uint8_t* buf = (uint8_t*)fpbo->insn;
				for(unsigned i = 0; i < fp->progs_per_bo; ++i)
				{
					memcpy(buf, fp->insn, fp->insn_len * 4);
					nvfx_fp_memcpy(map, fp->insn, fp->insn_len * 4);
					map += fp->prog_size;
					buf += fp->prog_size;
				}
			}
			fp->bo_prog_idx = 0;
		}

		int offset = fp->bo_prog_idx * fp->prog_size;
		uint32_t* fpmap = (uint32_t*)((char*)fp->fpbo->bo->map + offset);

		if(nvfx->constbuf[PIPE_SHADER_FRAGMENT]) {
			struct pipe_resource* constbuf = nvfx->constbuf[PIPE_SHADER_FRAGMENT];
			uint32_t* map = (uint32_t*)nvfx_buffer(constbuf)->data;
			uint32_t* fpmap = (uint32_t*)((char*)fp->fpbo->bo->map + offset);
			uint32_t* buf = (uint32_t*)((char*)fp->fpbo->insn + offset);
			int i;
			for (i = 0; i < fp->nr_consts; ++i) {
				unsigned off = fp->consts[i].offset;
				unsigned idx = fp->consts[i].index * 4;

				/* TODO: is checking a good idea? */
				if(memcmp(&buf[off], &map[idx], 4 * sizeof(uint32_t))) {
					memcpy(&buf[off], &map[idx], 4 * sizeof(uint32_t));
					nvfx_fp_memcpy(&fpmap[off], &map[idx], 4 * sizeof(uint32_t));
				}
			}
		}

		if(fp->progs_left_with_obsolete_slot_assignments) {
			unsigned char* fpbo_slots = &fp->fpbo->slots[fp->bo_prog_idx * 8];
			for(unsigned i = 0; i < fp->num_slots; ++i) {
				unsigned value = fp->slot_to_fp_input[i];;
				if(value != fpbo_slots[i]) {
					unsigned* p = (unsigned*)fp->slot_relocations[i].data;
					unsigned* pend = (unsigned*)((char*)fp->slot_relocations[i].data + fp->slot_relocations[i].size);
					for(; p != pend; ++p) {
						unsigned off = *p;
						unsigned dw = fp->insn[off];
						dw = (dw & ~NVFX_FP_OP_INPUT_SRC_MASK) | (value << NVFX_FP_OP_INPUT_SRC_SHIFT);
						nvfx_fp_memcpy(&fpmap[*p], &dw, sizeof(dw));
					}
					fpbo_slots[i] = value;
				}
			}
			--fp->progs_left_with_obsolete_slot_assignments;
		}
	}

	if(update || (nvfx->dirty & NVFX_NEW_FRAGPROG)) {
		int offset = fp->bo_prog_idx * fp->prog_size;
		MARK_RING(chan, 8, 1);
		OUT_RING(chan, RING_3D(NV34TCL_FP_ACTIVE_PROGRAM, 1));
		OUT_RELOC(chan, fp->fpbo->bo, offset, NOUVEAU_BO_VRAM |
			      NOUVEAU_BO_GART | NOUVEAU_BO_RD | NOUVEAU_BO_LOW |
			      NOUVEAU_BO_OR, NV34TCL_FP_ACTIVE_PROGRAM_DMA0,
			      NV34TCL_FP_ACTIVE_PROGRAM_DMA1);
		OUT_RING(chan, RING_3D(NV34TCL_FP_CONTROL, 1));
		OUT_RING(chan, fp->fp_control);
		if(!nvfx->is_nv4x) {
			OUT_RING(chan, RING_3D(NV34TCL_FP_REG_CONTROL, 1));
			OUT_RING(chan, (1<<16)|0x4);
			OUT_RING(chan, RING_3D(NV34TCL_TX_UNITS_ENABLE, 1));
			OUT_RING(chan, fp->samplers);
		}
	}

	if(nvfx->dirty & (NVFX_NEW_FRAGPROG | NVFX_NEW_SPRITE))
	{
		WAIT_RING(chan, 2);
		OUT_RING(chan, RING_3D(NV34TCL_POINT_SPRITE, 1));
		OUT_RING(chan, fp->point_sprite_control | nvfx->rasterizer->pipe.point_quad_rasterization);
	}
}

void
nvfx_fragprog_relocate(struct nvfx_context *nvfx)
{
	struct nouveau_channel* chan = nvfx->screen->base.channel;
	struct nvfx_fragment_program *fp = nvfx->fragprog;
	struct nouveau_bo* bo = fp->fpbo->bo;
	int offset = fp->bo_prog_idx * fp->prog_size;
	unsigned fp_flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_RD; // TODO: GART?
	fp_flags |= NOUVEAU_BO_DUMMY;
	MARK_RING(chan, 2, 2);
	OUT_RELOC(chan, bo, RING_3D(NV34TCL_FP_ACTIVE_PROGRAM, 1), fp_flags, 0, 0);
	OUT_RELOC(chan, bo, offset, fp_flags | NOUVEAU_BO_LOW |
		      NOUVEAU_BO_OR, NV34TCL_FP_ACTIVE_PROGRAM_DMA0,
		      NV34TCL_FP_ACTIVE_PROGRAM_DMA1);
}

void
nvfx_fragprog_destroy(struct nvfx_context *nvfx,
		      struct nvfx_fragment_program *fp)
{
	unsigned i;
	struct nvfx_fragment_program_bo* fpbo = fp->fpbo;
	if(fpbo)
	{
		do
		{
			struct nvfx_fragment_program_bo* next = fpbo->next;
			nouveau_bo_unmap(fpbo->bo);
			nouveau_bo_ref(0, &fpbo->bo);
			free(fpbo);
			fpbo = next;
		}
		while(fpbo != fp->fpbo);
	}

	for(i = 0; i < 8; ++i)
		util_dynarray_fini(&fp->slot_relocations[i]);

	if (fp->insn_len)
		FREE(fp->insn);
}