Merge branch 'master' into instanced-arrays

Conflicts:
	src/gallium/auxiliary/tgsi/tgsi_dump.c
	src/gallium/include/pipe/p_shader_tokens.h

10 files changed, 438 insertions, 201 deletions

diff --git a/src/gallium/drivers/nv50/nv50_context.c b/src/gallium/drivers/nv50/nv50_context.c
index d21b80eab8..5997456e4c 100644
--- a/src/gallium/drivers/nv50/nv50_context.c
+++ b/src/gallium/drivers/nv50/nv50_context.c
@@ -43,6 +43,39 @@ nv50_destroy(struct pipe_context *pipe)
 {
 	struct nv50_context *nv50 = nv50_context(pipe);
 
+        if (nv50->state.fb)
+		so_ref(NULL, &nv50->state.fb);
+	if (nv50->state.blend)
+		so_ref(NULL, &nv50->state.blend);
+	if (nv50->state.blend_colour)
+		so_ref(NULL, &nv50->state.blend_colour);
+	if (nv50->state.zsa)
+		so_ref(NULL, &nv50->state.zsa);
+	if (nv50->state.rast)
+		so_ref(NULL, &nv50->state.rast);
+	if (nv50->state.stipple)
+		so_ref(NULL, &nv50->state.stipple);
+	if (nv50->state.scissor)
+		so_ref(NULL, &nv50->state.scissor);
+	if (nv50->state.viewport)
+		so_ref(NULL, &nv50->state.viewport);
+	if (nv50->state.tsc_upload)
+		so_ref(NULL, &nv50->state.tsc_upload);
+	if (nv50->state.tic_upload)
+		so_ref(NULL, &nv50->state.tic_upload);
+	if (nv50->state.vertprog)
+		so_ref(NULL, &nv50->state.vertprog);
+	if (nv50->state.fragprog)
+		so_ref(NULL, &nv50->state.fragprog);
+	if (nv50->state.programs)
+		so_ref(NULL, &nv50->state.programs);
+	if (nv50->state.vtxfmt)
+		so_ref(NULL, &nv50->state.vtxfmt);
+	if (nv50->state.vtxbuf)
+		so_ref(NULL, &nv50->state.vtxbuf);
+	if (nv50->state.vtxattr)
+		so_ref(NULL, &nv50->state.vtxattr);
+
 	draw_destroy(nv50->draw);
 	FREE(nv50);
 }
diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c
index 679c28ce4b..2d0b1818ef 100644
--- a/src/gallium/drivers/nv50/nv50_program.c
+++ b/src/gallium/drivers/nv50/nv50_program.c
@@ -154,26 +154,17 @@ struct nv50_pc {
 	int if_lvl, loop_lvl;
 	unsigned loop_pos[NV50_MAX_LOOP_NESTING];
 
+	unsigned *insn_pos; /* actual program offset of each TGSI insn */
+	boolean in_subroutine;
+
 	/* current instruction and total number of insns */
 	unsigned insn_cur;
 	unsigned insn_nr;
 
 	boolean allow32;
-};
-
-static INLINE struct nv50_reg *
-reg_instance(struct nv50_pc *pc, struct nv50_reg *reg)
-{
-	struct nv50_reg *ri;
 
-	assert(pc->reg_instance_nr < 16);
-	ri = &pc->reg_instances[pc->reg_instance_nr++];
-	if (reg) {
-		*ri = *reg;
-		reg->mod = 0;
-	}
-	return ri;
-}
+	uint8_t edgeflag_out;
+};
 
 static INLINE void
 ctor_reg(struct nv50_reg *reg, unsigned type, int index, int hw)
@@ -253,6 +244,21 @@ alloc_reg(struct nv50_pc *pc, struct nv50_reg *reg)
 	assert(0);
 }
 
+static INLINE struct nv50_reg *
+reg_instance(struct nv50_pc *pc, struct nv50_reg *reg)
+{
+	struct nv50_reg *ri;
+
+	assert(pc->reg_instance_nr < 16);
+	ri = &pc->reg_instances[pc->reg_instance_nr++];
+	if (reg) {
+		alloc_reg(pc, reg);
+		*ri = *reg;
+		reg->mod = 0;
+	}
+	return ri;
+}
+
 /* XXX: For shaders that aren't executed linearly (e.g. shaders that
  * contain loops), we need to assign all hw regs to TGSI TEMPs early,
  * lest we risk temp_temps overwriting regs alloc'd "later".
@@ -279,22 +285,6 @@ alloc_temp(struct nv50_pc *pc, struct nv50_reg *dst)
 	return NULL;
 }
 
-/* Assign the hw of the discarded temporary register src
- * to the tgsi register dst and free src.
- */
-static void
-assimilate_temp(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src)
-{
-	assert(src->index == -1 && src->hw != -1);
-
-	if (dst->hw != -1)
-		pc->r_temp[dst->hw] = NULL;
-	pc->r_temp[src->hw] = dst;
-	dst->hw = src->hw;
-
-	FREE(src);
-}
-
 /* release the hardware resource held by r */
 static void
 release_hw(struct nv50_pc *pc, struct nv50_reg *r)
@@ -451,10 +441,19 @@ is_immd(struct nv50_program_exec *e)
 	return FALSE;
 }
 
+static boolean
+is_join(struct nv50_program_exec *e)
+{
+	if (is_long(e) && (e->inst[1] & 3) == 2)
+		return TRUE;
+	return FALSE;
+}
+
 static INLINE void
 set_pred(struct nv50_pc *pc, unsigned pred, unsigned idx,
 	 struct nv50_program_exec *e)
 {
+	assert(!is_immd(e));
 	set_long(pc, e);
 	e->inst[1] &= ~((0x1f << 7) | (0x3 << 12));
 	e->inst[1] |= (pred << 7) | (idx << 12);
@@ -497,15 +496,6 @@ set_dst(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_program_exec *e)
 static INLINE void
 set_immd(struct nv50_pc *pc, struct nv50_reg *imm, struct nv50_program_exec *e)
 {
-	union {
-		float f;
-		uint32_t ui;
-	} u;
-	u.ui = pc->immd_buf[imm->hw];
-
-	u.f = (imm->mod & NV50_MOD_ABS) ? fabsf(u.f) : u.f;
-	u.f = (imm->mod & NV50_MOD_NEG) ? -u.f : u.f;
-
 	set_long(pc, e);
 	/* XXX: can't be predicated - bits overlap; cases where both
 	 * are required should be avoided by using pc->allow32 */
@@ -513,8 +503,8 @@ set_immd(struct nv50_pc *pc, struct nv50_reg *imm, struct nv50_program_exec *e)
 	set_pred_wr(pc, 0, 0, e);
 
 	e->inst[1] |= 0x00000002 | 0x00000001;
-	e->inst[0] |= (u.ui & 0x3f) << 16;
-	e->inst[1] |= (u.ui >> 6) << 2;
+	e->inst[0] |= (pc->immd_buf[imm->hw] & 0x3f) << 16;
+	e->inst[1] |= (pc->immd_buf[imm->hw] >> 6) << 2;
 }
 
 static INLINE void
@@ -663,6 +653,7 @@ set_data(struct nv50_pc *pc, struct nv50_reg *src, unsigned m, unsigned s,
 	e->inst[1] |= (((src->type == P_IMMD) ? 0 : 1) << 22);
 }
 
+/* Never apply nv50_reg::mod in emit_mov, or carefully check the code !!! */
 static void
 emit_mov(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src)
 {
@@ -715,6 +706,34 @@ emit_mov_immdval(struct nv50_pc *pc, struct nv50_reg *dst, float f)
 	FREE(imm);
 }
 
+/* Assign the hw of the discarded temporary register src
+ * to the tgsi register dst and free src.
+ */
+static void
+assimilate_temp(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src)
+{
+	assert(src->index == -1 && src->hw != -1);
+
+	if (pc->if_lvl || pc->loop_lvl ||
+	    (dst->type != P_TEMP) ||
+	    (src->hw < pc->result_nr * 4 &&
+	     pc->p->type == PIPE_SHADER_FRAGMENT) ||
+	    pc->p->info.opcode_count[TGSI_OPCODE_CAL] ||
+	    pc->p->info.opcode_count[TGSI_OPCODE_BRA]) {
+
+		emit_mov(pc, dst, src);
+		free_temp(pc, src);
+		return;
+	}
+
+	if (dst->hw != -1)
+		pc->r_temp[dst->hw] = NULL;
+	pc->r_temp[src->hw] = dst;
+	dst->hw = src->hw;
+
+	FREE(src);
+}
+
 static void
 emit_nop(struct nv50_pc *pc)
 {
@@ -886,7 +905,7 @@ emit_mul(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src0,
 	set_dst(pc, dst, e);
 	set_src_0(pc, src0, e);
 	if (src1->type == P_IMMD && !is_long(e)) {
-		if (src0->mod & NV50_MOD_NEG)
+		if (src0->mod ^ src1->mod)
 			e->inst[0] |= 0x00008000;
 		set_immd(pc, src1, e);
 	} else {
@@ -997,6 +1016,8 @@ emit_bitop2(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src0,
 	    op != TGSI_OPCODE_XOR)
 		assert(!"invalid bit op");
 
+	assert(!(src0->mod | src1->mod));
+
 	if (src1->type == P_IMMD && src0->type == P_TEMP && pc->allow32) {
 		set_immd(pc, src1, e);
 		if (op == TGSI_OPCODE_OR)
@@ -1048,6 +1069,14 @@ emit_msb(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src0,
 	src2->mod ^= NV50_MOD_NEG;
 }
 
+#define NV50_FLOP_RCP 0
+#define NV50_FLOP_RSQ 2
+#define NV50_FLOP_LG2 3
+#define NV50_FLOP_SIN 4
+#define NV50_FLOP_COS 5
+#define NV50_FLOP_EX2 6
+
+/* rcp, rsqrt, lg2 support neg and abs */
 static void
 emit_flop(struct nv50_pc *pc, unsigned sub,
 	  struct nv50_reg *dst, struct nv50_reg *src)
@@ -1055,17 +1084,20 @@ emit_flop(struct nv50_pc *pc, unsigned sub,
 	struct nv50_program_exec *e = exec(pc);
 
 	e->inst[0] |= 0x90000000;
-	if (sub) {
+	if (sub || src->mod) {
 		set_long(pc, e);
 		e->inst[1] |= (sub << 29);
 	}
 
 	set_dst(pc, dst, e);
+	set_src_0_restricted(pc, src, e);
 
-	if (sub == 0 || sub == 2)
-		set_src_0_restricted(pc, src, e);
-	else
-		set_src_0(pc, src, e);
+	assert(!src->mod || sub < 4);
+
+	if (src->mod & NV50_MOD_NEG)
+		e->inst[1] |= 0x04000000;
+	if (src->mod & NV50_MOD_ABS)
+		e->inst[1] |= 0x00100000;
 
 	emit(pc, e);
 }
@@ -1082,6 +1114,11 @@ emit_preex2(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src)
 	set_long(pc, e);
 	e->inst[1] |= (6 << 29) | 0x00004000;
 
+	if (src->mod & NV50_MOD_NEG)
+		e->inst[1] |= 0x04000000;
+	if (src->mod & NV50_MOD_ABS)
+		e->inst[1] |= 0x00100000;
+
 	emit(pc, e);
 }
 
@@ -1097,6 +1134,11 @@ emit_precossin(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src)
 	set_long(pc, e);
 	e->inst[1] |= (6 << 29);
 
+	if (src->mod & NV50_MOD_NEG)
+		e->inst[1] |= 0x04000000;
+	if (src->mod & NV50_MOD_ABS)
+		e->inst[1] |= 0x00100000;
+
 	emit(pc, e);
 }
 
@@ -1231,10 +1273,10 @@ emit_pow(struct nv50_pc *pc, struct nv50_reg *dst,
 {
 	struct nv50_reg *temp = alloc_temp(pc, NULL);
 
-	emit_flop(pc, 3, temp, v);
+	emit_flop(pc, NV50_FLOP_LG2, temp, v);
 	emit_mul(pc, temp, temp, e);
 	emit_preex2(pc, temp, temp);
-	emit_flop(pc, 6, dst, temp);
+	emit_flop(pc, NV50_FLOP_EX2, dst, temp);
 
 	free_temp(pc, temp);
 }
@@ -1336,66 +1378,53 @@ emit_kil(struct nv50_pc *pc, struct nv50_reg *src)
 }
 
 static struct nv50_program_exec *
-emit_breakaddr(struct nv50_pc *pc)
+emit_control_flow(struct nv50_pc *pc, unsigned op, int pred, unsigned cc)
 {
 	struct nv50_program_exec *e = exec(pc);
 
-	e->inst[0] = 0x40000002;
+	e->inst[0] = (op << 28) | 2;
 	set_long(pc, e);
+	if (pred >= 0)
+		set_pred(pc, cc, pred, e);
 
 	emit(pc, e);
 	return e;
 }
 
-static void
-emit_break(struct nv50_pc *pc, int pred, unsigned cc)
+static INLINE struct nv50_program_exec *
+emit_breakaddr(struct nv50_pc *pc)
 {
-	struct nv50_program_exec *e = exec(pc);
-
-	e->inst[0] = 0x50000002;
-	set_long(pc, e);
-	if (pred >= 0)
-		set_pred(pc, cc, pred, e);
+	return emit_control_flow(pc, 0x4, -1, 0);
+}
 
-	emit(pc, e);
+static INLINE void
+emit_break(struct nv50_pc *pc, int pred, unsigned cc)
+{
+	emit_control_flow(pc, 0x5, pred, cc);
 }
 
-static struct nv50_program_exec *
+static INLINE struct nv50_program_exec *
 emit_joinat(struct nv50_pc *pc)
 {
-	struct nv50_program_exec *e = exec(pc);
-
-	e->inst[0] = 0xa0000002;
-	set_long(pc, e);
-
-	emit(pc, e);
-	return e;
+	return emit_control_flow(pc, 0xa, -1, 0);
 }
 
-static struct nv50_program_exec *
+static INLINE struct nv50_program_exec *
 emit_branch(struct nv50_pc *pc, int pred, unsigned cc)
 {
-	struct nv50_program_exec *e = exec(pc);
+	return emit_control_flow(pc, 0x1, pred, cc);
+}
 
-	e->inst[0] = 0x10000002;
-	set_long(pc, e);
-	if (pred >= 0)
-		set_pred(pc, cc, pred, e);
-	emit(pc, e);
-	return pc->p->exec_tail;
+static INLINE struct nv50_program_exec *
+emit_call(struct nv50_pc *pc, int pred, unsigned cc)
+{
+	return emit_control_flow(pc, 0x2, pred, cc);
 }
 
-static void
+static INLINE void
 emit_ret(struct nv50_pc *pc, int pred, unsigned cc)
 {
-	struct nv50_program_exec *e = exec(pc);
-
-	e->inst[0] = 0x30000002;
-	set_long(pc, e);
-	if (pred >= 0)
-		set_pred(pc, cc, pred, e);
-
-	emit(pc, e);
+	emit_control_flow(pc, 0x3, pred, cc);
 }
 
 #define QOP_ADD 0
@@ -1458,7 +1487,7 @@ load_cube_tex_coords(struct nv50_pc *pc, struct nv50_reg *t[4],
 	if (arg == 4) /* there is no textureProj(samplerCubeShadow) */
 		emit_mov(pc, t[3], src[3]);
 
-	emit_flop(pc, 0, t[2], t[2]);
+	emit_flop(pc, NV50_FLOP_RCP, t[2], t[2]);
 
 	emit_mul(pc, t[0], src[0], t[2]);
 	emit_mul(pc, t[1], src[1], t[2]);
@@ -1476,7 +1505,7 @@ load_proj_tex_coords(struct nv50_pc *pc, struct nv50_reg *t[4],
 
 		t[3]->rhw = src[3]->rhw;
 		emit_interp(pc, t[3], NULL, (mode & INTERP_CENTROID));
-		emit_flop(pc, 0, t[3], t[3]);
+		emit_flop(pc, NV50_FLOP_RCP, t[3], t[3]);
 
 		for (c = 0; c < dim; ++c) {
 			t[c]->rhw = src[c]->rhw;
@@ -1490,7 +1519,7 @@ load_proj_tex_coords(struct nv50_pc *pc, struct nv50_reg *t[4],
 		/* XXX: for some reason the blob sometimes uses MAD
 		 * (mad f32 $rX $rY $rZ neg $r63)
 		 */
-		emit_flop(pc, 0, t[3], src[3]);
+		emit_flop(pc, NV50_FLOP_RCP, t[3], src[3]);
 		for (c = 0; c < dim; ++c)
 			emit_mul(pc, t[c], src[c], t[3]);
 		if (arg != dim) /* depth reference value */
@@ -1537,7 +1566,13 @@ emit_texlod_sequence(struct nv50_pc *pc, struct nv50_reg *tlod,
 		     struct nv50_reg *src, struct nv50_program_exec *tex)
 {
 	struct nv50_program_exec *join_at;
-	unsigned i, target = pc->p->exec_size + 7 * 2;
+	unsigned i, target = pc->p->exec_size + 9 * 2;
+
+	if (pc->p->type != PIPE_SHADER_FRAGMENT) {
+		emit(pc, tex);
+		return;
+	}
+	pc->allow32 = FALSE;
 
 	/* Subtract lod of each pixel from lod of top left pixel, jump
 	 * texlod insn if result is 0, then repeat for 2 other pixels.
@@ -1663,6 +1698,7 @@ emit_tex(struct nv50_pc *pc, struct nv50_reg **dst, unsigned mask,
 		emit(pc, e);
 	} else
 	if (bias_lod < 0) {
+		assert(pc->p->type == PIPE_SHADER_FRAGMENT);
 		e->inst[0] |= arg << 22;
 		e->inst[1] |= 0x20000000; /* texbias */
 		emit_mov(pc, t[arg], src[3]);
@@ -1782,20 +1818,24 @@ static boolean
 negate_supported(const struct tgsi_full_instruction *insn, int i)
 {
 	switch (insn->Instruction.Opcode) {
+	case TGSI_OPCODE_ADD:
+	case TGSI_OPCODE_COS:
 	case TGSI_OPCODE_DDX:
 	case TGSI_OPCODE_DDY:
 	case TGSI_OPCODE_DP3:
 	case TGSI_OPCODE_DP4:
-	case TGSI_OPCODE_MUL:
+	case TGSI_OPCODE_EX2:
 	case TGSI_OPCODE_KIL:
-	case TGSI_OPCODE_ADD:
-	case TGSI_OPCODE_SUB:
+	case TGSI_OPCODE_LG2:
 	case TGSI_OPCODE_MAD:
-		return TRUE;
+	case TGSI_OPCODE_MUL:
 	case TGSI_OPCODE_POW:
-		if (i == 1)
-			return TRUE;
-		return FALSE;
+	case TGSI_OPCODE_RCP:
+	case TGSI_OPCODE_RSQ: /* ignored, RSQ = rsqrt(abs(src.x)) */
+	case TGSI_OPCODE_SCS:
+	case TGSI_OPCODE_SIN:
+	case TGSI_OPCODE_SUB:
+		return TRUE;
 	default:
 		return FALSE;
 	}
@@ -1820,7 +1860,9 @@ nv50_tgsi_src_mask(const struct tgsi_full_instruction *insn, int c)
 	case TGSI_OPCODE_DST:
 		return mask & (c ? 0xa : 0x6);
 	case TGSI_OPCODE_EX2:
+	case TGSI_OPCODE_EXP:
 	case TGSI_OPCODE_LG2:
+	case TGSI_OPCODE_LOG:
 	case TGSI_OPCODE_POW:
 	case TGSI_OPCODE_RCP:
 	case TGSI_OPCODE_RSQ:
@@ -2042,6 +2084,8 @@ nv50_tgsi_dst_revdep(unsigned op, int s, int c)
 			assert(0);
 			return 0x0;
 		}
+	case TGSI_OPCODE_EXP:
+	case TGSI_OPCODE_LOG:
 	case TGSI_OPCODE_LIT:
 	case TGSI_OPCODE_SCS:
 	case TGSI_OPCODE_TEX:
@@ -2082,6 +2126,8 @@ nv50_kill_branch(struct nv50_pc *pc)
 
 	if (pc->if_insn[lvl]->next != pc->p->exec_tail)
 		return FALSE;
+	if (is_immd(pc->p->exec_tail))
+		return FALSE;
 
 	/* if ccode == 'true', the BRA is from an ELSE and the predicate
 	 * reg may no longer be valid, since we currently always use $p0
@@ -2215,10 +2261,22 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 		pc->loop_pos[pc->loop_lvl++] = pc->p->exec_size;
 		terminate_mbb(pc);
 		break;
+	case TGSI_OPCODE_BGNSUB:
+		assert(!pc->in_subroutine);
+		pc->in_subroutine = TRUE;
+		/* probably not necessary, but align to 8 byte boundary */
+		if (!is_long(pc->p->exec_tail))
+			convert_to_long(pc, pc->p->exec_tail);
+		break;
 	case TGSI_OPCODE_BRK:
 		assert(pc->loop_lvl > 0);
 		emit_break(pc, -1, 0);
 		break;
+	case TGSI_OPCODE_CAL:
+		assert(inst->Label.Label < pc->insn_nr);
+		emit_call(pc, -1, 0)->param.index = inst->Label.Label;
+		/* replaced by actual offset in nv50_program_fixup_insns */
+		break;
 	case TGSI_OPCODE_CEIL:
 		for (c = 0; c < 4; c++) {
 			if (!(mask & (1 << c)))
@@ -2239,17 +2297,22 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 			set_pred(pc, 0x6, 1, pc->p->exec_tail); /* @NSF */
 		}
 		break;
+	case TGSI_OPCODE_CONT:
+		assert(pc->loop_lvl > 0);
+		emit_branch(pc, -1, 0)->param.index =
+			pc->loop_pos[pc->loop_lvl - 1];
+		break;
 	case TGSI_OPCODE_COS:
 		if (mask & 8) {
 			emit_precossin(pc, temp, src[0][3]);
-			emit_flop(pc, 5, dst[3], temp);
+			emit_flop(pc, NV50_FLOP_COS, dst[3], temp);
 			if (!(mask &= 7))
 				break;
 			if (temp == dst[3])
 				temp = brdc = temp_temp(pc);
 		}
 		emit_precossin(pc, temp, src[0][0]);
-		emit_flop(pc, 5, brdc, temp);
+		emit_flop(pc, NV50_FLOP_COS, brdc, temp);
 		break;
 	case TGSI_OPCODE_DDX:
 		for (c = 0; c < 4; c++) {
@@ -2321,9 +2384,40 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 		pc->loop_brka[pc->loop_lvl]->param.index = pc->p->exec_size;
 		terminate_mbb(pc);
 		break;
+	case TGSI_OPCODE_ENDSUB:
+		assert(pc->in_subroutine);
+		pc->in_subroutine = FALSE;
+		break;
 	case TGSI_OPCODE_EX2:
 		emit_preex2(pc, temp, src[0][0]);
-		emit_flop(pc, 6, brdc, temp);
+		emit_flop(pc, NV50_FLOP_EX2, brdc, temp);
+		break;
+	case TGSI_OPCODE_EXP:
+	{
+		struct nv50_reg *t[2];
+
+		assert(!temp);
+		t[0] = temp_temp(pc);
+		t[1] = temp_temp(pc);
+
+		if (mask & 0x6)
+			emit_mov(pc, t[0], src[0][0]);
+		if (mask & 0x3)
+			emit_flr(pc, t[1], src[0][0]);
+
+		if (mask & (1 << 1))
+			emit_sub(pc, dst[1], t[0], t[1]);
+		if (mask & (1 << 0)) {
+			emit_preex2(pc, t[1], t[1]);
+			emit_flop(pc, NV50_FLOP_EX2, dst[0], t[1]);
+		}
+		if (mask & (1 << 2)) {
+			emit_preex2(pc, t[0], t[0]);
+			emit_flop(pc, NV50_FLOP_EX2, dst[2], t[0]);
+		}
+		if (mask & (1 << 3))
+			emit_mov_immdval(pc, dst[3], 1.0f);
+	}
 		break;
 	case TGSI_OPCODE_FLR:
 		for (c = 0; c < 4; c++) {
@@ -2363,7 +2457,35 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 		emit_lit(pc, &dst[0], mask, &src[0][0]);
 		break;
 	case TGSI_OPCODE_LG2:
-		emit_flop(pc, 3, brdc, src[0][0]);
+		emit_flop(pc, NV50_FLOP_LG2, brdc, src[0][0]);
+		break;
+	case TGSI_OPCODE_LOG:
+	{
+		struct nv50_reg *t[2];
+
+		t[0] = temp_temp(pc);
+		if (mask & (1 << 1))
+			t[1] = temp_temp(pc);
+		else
+			t[1] = t[0];
+
+		emit_abs(pc, t[0], src[0][0]);
+		emit_flop(pc, NV50_FLOP_LG2, t[1], t[0]);
+		if (mask & (1 << 2))
+			emit_mov(pc, dst[2], t[1]);
+		emit_flr(pc, t[1], t[1]);
+		if (mask & (1 << 0))
+			emit_mov(pc, dst[0], t[1]);
+		if (mask & (1 << 1)) {
+			t[1]->mod = NV50_MOD_NEG;
+			emit_preex2(pc, t[1], t[1]);
+			t[1]->mod = 0;
+			emit_flop(pc, NV50_FLOP_EX2, t[1], t[1]);
+			emit_mul(pc, dst[1], t[0], t[1]);
+		}
+		if (mask & (1 << 3))
+			emit_mov_immdval(pc, dst[3], 1.0f);
+	}
 		break;
 	case TGSI_OPCODE_LRP:
 		temp = temp_temp(pc);
@@ -2413,24 +2535,25 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 		emit_pow(pc, brdc, src[0][0], src[1][0]);
 		break;
 	case TGSI_OPCODE_RCP:
-		emit_flop(pc, 0, brdc, src[0][0]);
+		emit_flop(pc, NV50_FLOP_RCP, brdc, src[0][0]);
 		break;
 	case TGSI_OPCODE_RET:
-		if (pc->p->type == PIPE_SHADER_FRAGMENT)
+		if (pc->p->type == PIPE_SHADER_FRAGMENT && !pc->in_subroutine)
 			nv50_fp_move_results(pc);
 		emit_ret(pc, -1, 0);
 		break;
 	case TGSI_OPCODE_RSQ:
-		emit_flop(pc, 2, brdc, src[0][0]);
+		src[0][0]->mod |= NV50_MOD_ABS;
+		emit_flop(pc, NV50_FLOP_RSQ, brdc, src[0][0]);
 		break;
 	case TGSI_OPCODE_SCS:
 		temp = temp_temp(pc);
 		if (mask & 3)
 			emit_precossin(pc, temp, src[0][0]);
 		if (mask & (1 << 0))
-			emit_flop(pc, 5, dst[0], temp);
+			emit_flop(pc, NV50_FLOP_COS, dst[0], temp);
 		if (mask & (1 << 1))
-			emit_flop(pc, 4, dst[1], temp);
+			emit_flop(pc, NV50_FLOP_SIN, dst[1], temp);
 		if (mask & (1 << 2))
 			emit_mov_immdval(pc, dst[2], 0.0);
 		if (mask & (1 << 3))
@@ -2439,14 +2562,14 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 	case TGSI_OPCODE_SIN:
 		if (mask & 8) {
 			emit_precossin(pc, temp, src[0][3]);
-			emit_flop(pc, 4, dst[3], temp);
+			emit_flop(pc, NV50_FLOP_SIN, dst[3], temp);
 			if (!(mask &= 7))
 				break;
 			if (temp == dst[3])
 				temp = brdc = temp_temp(pc);
 		}
 		emit_precossin(pc, temp, src[0][0]);
-		emit_flop(pc, 4, brdc, temp);
+		emit_flop(pc, NV50_FLOP_SIN, brdc, temp);
 		break;
 	case TGSI_OPCODE_SLT:
 	case TGSI_OPCODE_SGE:
@@ -2510,6 +2633,17 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 			emit_mov_immdval(pc, dst[3], 1.0);
 		break;
 	case TGSI_OPCODE_END:
+		if (pc->p->type == PIPE_SHADER_FRAGMENT)
+			nv50_fp_move_results(pc);
+
+		/* last insn must be long so it can have the exit bit set */
+		if (!is_long(pc->p->exec_tail))
+			convert_to_long(pc, pc->p->exec_tail);
+		else
+		if (is_immd(pc->p->exec_tail) || is_join(pc->p->exec_tail))
+			emit_nop(pc);
+
+		pc->p->exec_tail->inst[1] |= 1; /* set exit bit */
 		break;
 	default:
 		NOUVEAU_ERR("invalid opcode %d\n", inst->Instruction.Opcode);
@@ -2554,10 +2688,16 @@ prep_inspect_insn(struct nv50_pc *pc, const struct tgsi_full_instruction *insn)
 	mask = dst->WriteMask;
 
         if (dst->File == TGSI_FILE_TEMPORARY)
-                reg = pc->temp;
+		reg = pc->temp;
         else
-        if (dst->File == TGSI_FILE_OUTPUT)
-                reg = pc->result;
+	if (dst->File == TGSI_FILE_OUTPUT) {
+		reg = pc->result;
+
+		if (insn->Instruction.Opcode == TGSI_OPCODE_MOV &&
+		    dst->Index == pc->edgeflag_out &&
+		    insn->Src[0].Register.File == TGSI_FILE_INPUT)
+			pc->p->cfg.edgeflag_in = insn->Src[0].Register.Index;
+	}
 
 	if (reg) {
 		for (c = 0; c < 4; c++) {
@@ -2724,7 +2864,7 @@ nv50_tgsi_insn(struct nv50_pc *pc, const union tgsi_full_token *tok)
 	}
 	pc->r_brdc = NULL;
 
-	if (!deqs)
+	if (!deqs || (!rdep[0] && !rdep[1] && !rdep[2] && !rdep[3]))
 		return nv50_program_tx_insn(pc, &insn);
 
 	deqs = nv50_revdep_reorder(m, rdep);
@@ -2775,7 +2915,7 @@ load_interpolant(struct nv50_pc *pc, struct nv50_reg *reg)
 		iv->rhw = popcnt4(pc->p->cfg.regs[1] >> 24) - 1;
 
 		emit_interp(pc, iv, NULL, mode & INTERP_CENTROID);
-		emit_flop(pc, 0, iv, iv);
+		emit_flop(pc, NV50_FLOP_RCP, iv, iv);
 
 		/* XXX: when loading interpolants dynamically, move these
 		 * to the program head, or make sure it can't be skipped.
@@ -2856,6 +2996,9 @@ nv50_program_tx_prep(struct nv50_pc *pc)
 					if (p->cfg.io_nr > first)
 						p->cfg.io_nr = first;
 					break;
+				case TGSI_SEMANTIC_EDGEFLAG:
+					pc->edgeflag_out = first;
+					break;
 					/*
 				case TGSI_SEMANTIC_CLIP_DISTANCE:
 					p->cfg.clpd = MIN2(p->cfg.clpd, first);
@@ -3104,6 +3247,8 @@ ctor_nv50_pc(struct nv50_pc *pc, struct nv50_program *p)
 	p->cfg.two_side[0].hw = 0x40;
 	p->cfg.two_side[1].hw = 0x40;
 
+	p->cfg.edgeflag_in = pc->edgeflag_out = 0xff;
+
 	switch (p->type) {
 	case PIPE_SHADER_VERTEX:
 		p->cfg.psiz = 0x40;
@@ -3192,16 +3337,6 @@ nv50_program_fixup_insns(struct nv50_pc *pc)
 		if (e->param.index >= 0 && !e->param.mask)
 			bra_list[n++] = e;
 
-	/* last instruction must be long so it can have the exit bit set */
-	if (!is_long(pc->p->exec_tail))
-		convert_to_long(pc, pc->p->exec_tail);
-	/* set exit bit */
-	pc->p->exec_tail->inst[1] |= 1;
-
-	/* !immd on exit insn simultaneously means !join */
-	assert(!is_immd(pc->p->exec_head));
-	assert(!is_immd(pc->p->exec_tail));
-
 	/* Make sure we don't have any single 32 bit instructions. */
 	for (e = pc->p->exec_head, pos = 0; e; e = e->next) {
 		pos += is_long(e) ? 2 : 1;
@@ -3210,12 +3345,24 @@ nv50_program_fixup_insns(struct nv50_pc *pc)
 			for (i = 0; i < n; ++i)
 				if (bra_list[i]->param.index >= pos)
 					bra_list[i]->param.index += 1;
+			for (i = 0; i < pc->insn_nr; ++i)
+				if (pc->insn_pos[i] >= pos)
+					pc->insn_pos[i] += 1;
 			convert_to_long(pc, e);
 			++pos;
 		}
 	}
 
 	FREE(bra_list);
+
+	if (!pc->p->info.opcode_count[TGSI_OPCODE_CAL])
+		return;
+
+	/* fill in CALL offsets */
+	for (e = pc->p->exec_head; e; e = e->next) {
+		if ((e->inst[0] & 2) && (e->inst[0] >> 28) == 0x2)
+			e->param.index = pc->insn_pos[e->param.index];
+	}
 }
 
 static boolean
@@ -3237,19 +3384,20 @@ nv50_program_tx(struct nv50_program *p)
 	if (ret == FALSE)
 		goto out_cleanup;
 
+	pc->insn_pos = MALLOC(pc->insn_nr * sizeof(unsigned));
+
 	tgsi_parse_init(&parse, pc->p->pipe.tokens);
 	while (!tgsi_parse_end_of_tokens(&parse)) {
 		const union tgsi_full_token *tok = &parse.FullToken;
 
-		/* don't allow half insn/immd on first and last instruction */
+		/* previously allow32 was FALSE for first & last instruction */
 		pc->allow32 = TRUE;
-		if (pc->insn_cur == 0 || pc->insn_cur + 2 == pc->insn_nr)
-			pc->allow32 = FALSE;
 
 		tgsi_parse_token(&parse);
 
 		switch (tok->Token.Type) {
 		case TGSI_TOKEN_TYPE_INSTRUCTION:
+			pc->insn_pos[pc->insn_cur] = pc->p->exec_size;
 			++pc->insn_cur;
 			ret = nv50_tgsi_insn(pc, tok);
 			if (ret == FALSE)
@@ -3260,9 +3408,6 @@ nv50_program_tx(struct nv50_program *p)
 		}
 	}
 
-	if (pc->p->type == PIPE_SHADER_FRAGMENT)
-		nv50_fp_move_results(pc);
-
 	nv50_program_fixup_insns(pc);
 
 	p->param_nr = pc->param_nr * 4;
@@ -3480,7 +3625,7 @@ nv50_fragprog_validate(struct nv50_context *nv50)
 	so_data  (so, p->cfg.high_temp);
 	so_method(so, tesla, NV50TCL_FP_RESULT_COUNT, 1);
 	so_data  (so, p->cfg.high_result);
-	so_method(so, tesla, NV50TCL_FP_CTRL_UNK19A8, 1);
+	so_method(so, tesla, NV50TCL_FP_CONTROL, 1);
 	so_data  (so, p->cfg.regs[2]);
 	so_method(so, tesla, NV50TCL_FP_CTRL_UNK196C, 1);
 	so_data  (so, p->cfg.regs[3]);
@@ -3652,7 +3797,7 @@ nv50_linkage_validate(struct nv50_context *nv50)
 	so_method(so, tesla, NV50TCL_FP_INTERPOLANT_CTRL, 1);
 	so_data  (so, reg[4]);
 
-	so_method(so, tesla, 0x1540, 4);
+	so_method(so, tesla, NV50TCL_NOPERSPECTIVE_BITMAP(0), 4);
 	so_datap (so, lin, 4);
 
 	if (nv50->rasterizer->pipe.point_sprite) {
diff --git a/src/gallium/drivers/nv50/nv50_program.h b/src/gallium/drivers/nv50/nv50_program.h
index 4a90c372ce..461fec1d89 100644
--- a/src/gallium/drivers/nv50/nv50_program.h
+++ b/src/gallium/drivers/nv50/nv50_program.h
@@ -58,6 +58,7 @@ struct nv50_program {
 		/* VP only */
 		uint8_t clpd, clpd_nr;
 		uint8_t psiz;
+		uint8_t edgeflag_in;
 	} cfg;
 };
 
diff --git a/src/gallium/drivers/nv50/nv50_query.c b/src/gallium/drivers/nv50/nv50_query.c
index 268c9823f7..5d9e18218a 100644
--- a/src/gallium/drivers/nv50/nv50_query.c
+++ b/src/gallium/drivers/nv50/nv50_query.c
@@ -77,9 +77,9 @@ nv50_query_begin(struct pipe_context *pipe, struct pipe_query *pq)
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
 	struct nv50_query *q = nv50_query(pq);
 
-	BEGIN_RING(chan, tesla, 0x1530, 1);
+	BEGIN_RING(chan, tesla, NV50TCL_SAMPLECNT_RESET, 1);
 	OUT_RING  (chan, 1);
-	BEGIN_RING(chan, tesla, 0x1514, 1);
+	BEGIN_RING(chan, tesla, NV50TCL_SAMPLECNT_ENABLE, 1);
 	OUT_RING  (chan, 1);
 
 	q->ready = FALSE;
diff --git a/src/gallium/drivers/nv50/nv50_screen.c b/src/gallium/drivers/nv50/nv50_screen.c
index d443ca3ad0..7e039ea82e 100644
--- a/src/gallium/drivers/nv50/nv50_screen.c
+++ b/src/gallium/drivers/nv50/nv50_screen.c
@@ -128,7 +128,7 @@ nv50_screen_get_param(struct pipe_screen *pscreen, int param)
 	case PIPE_CAP_TEXTURE_MIRROR_REPEAT:
 		return 1;
 	case PIPE_CAP_TGSI_CONT_SUPPORTED:
-		return 0;
+		return 1;
 	case PIPE_CAP_BLEND_EQUATION_SEPARATE:
 		return 1;
 	case NOUVEAU_CAP_HW_VTXBUF:
@@ -165,6 +165,21 @@ static void
 nv50_screen_destroy(struct pipe_screen *pscreen)
 {
 	struct nv50_screen *screen = nv50_screen(pscreen);
+	unsigned i;
+
+	for (i = 0; i < 2; i++) {
+		if (screen->constbuf_parm[i])
+			nouveau_bo_ref(NULL, &screen->constbuf_parm[i]);
+	}
+
+	if (screen->constbuf_misc[0])
+		nouveau_bo_ref(NULL, &screen->constbuf_misc[0]);
+	if (screen->tic)
+		nouveau_bo_ref(NULL, &screen->tic);
+	if (screen->tsc)
+		nouveau_bo_ref(NULL, &screen->tsc);
+	if (screen->static_init)
+		so_ref(NULL, &screen->static_init);
 
 	nouveau_notifier_free(&screen->sync);
 	nouveau_grobj_free(&screen->tesla);
@@ -231,8 +246,7 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 		break;
 	case 0x80:
 	case 0x90:
-		/* this stupid name should be corrected. */
-		tesla_class = NV54TCL;
+		tesla_class = NV84TCL;
 		break;
 	case 0xa0:
 		switch (chipset) {
@@ -242,7 +256,7 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 			tesla_class = NVA0TCL;
 			break;
 		default:
-			tesla_class = 0x8597;
+			tesla_class = NVA8TCL;
 			break;
 		}
 		break;
@@ -287,7 +301,7 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 	so_data  (so, chan->vram->handle);
 	so_method(so, screen->eng2d, NV50_2D_OPERATION, 1);
 	so_data  (so, NV50_2D_OPERATION_SRCCOPY);
-	so_method(so, screen->eng2d, 0x0290, 1);
+	so_method(so, screen->eng2d, NV50_2D_CLIP_ENABLE, 1);
 	so_data  (so, 0);
 	so_method(so, screen->eng2d, 0x0888, 1);
 	so_data  (so, 1);
@@ -297,34 +311,33 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 	/* Static tesla init */
 	so = so_new(256, 20);
 
-	so_method(so, screen->tesla, 0x1558, 1);
-	so_data  (so, 1);
+	so_method(so, screen->tesla, NV50TCL_COND_MODE, 1);
+	so_data  (so, NV50TCL_COND_MODE_ALWAYS);
 	so_method(so, screen->tesla, NV50TCL_DMA_NOTIFY, 1);
 	so_data  (so, screen->sync->handle);
-	so_method(so, screen->tesla, NV50TCL_DMA_UNK0(0),
-				     NV50TCL_DMA_UNK0__SIZE);
-	for (i = 0; i < NV50TCL_DMA_UNK0__SIZE; i++)
+	so_method(so, screen->tesla, NV50TCL_DMA_ZETA, 11);
+	for (i = 0; i < 11; i++)
 		so_data(so, chan->vram->handle);
-	so_method(so, screen->tesla, NV50TCL_DMA_UNK1(0),
-				     NV50TCL_DMA_UNK1__SIZE);
-	for (i = 0; i < NV50TCL_DMA_UNK1__SIZE; i++)
+	so_method(so, screen->tesla, NV50TCL_DMA_COLOR(0),
+				     NV50TCL_DMA_COLOR__SIZE);
+	for (i = 0; i < NV50TCL_DMA_COLOR__SIZE; i++)
 		so_data(so, chan->vram->handle);
-	so_method(so, screen->tesla, 0x121c, 1);
+	so_method(so, screen->tesla, NV50TCL_RT_CONTROL, 1);
 	so_data  (so, 1);
 
 	/* activate all 32 lanes (threads) in a warp */
-	so_method(so, screen->tesla, 0x19a0, 1);
+	so_method(so, screen->tesla, NV50TCL_WARP_HALVES, 1);
 	so_data  (so, 0x2);
 	so_method(so, screen->tesla, 0x1400, 1);
 	so_data  (so, 0xf);
 
 	/* max TIC (bits 4:8) & TSC (ignored) bindings, per program type */
-	so_method(so, screen->tesla, 0x13b4, 1);
+	so_method(so, screen->tesla, NV50TCL_TEX_LIMITS(0), 1);
 	so_data  (so, 0x54);
-	so_method(so, screen->tesla, 0x13bc, 1);
+	so_method(so, screen->tesla, NV50TCL_TEX_LIMITS(2), 1);
 	so_data  (so, 0x54);
 	/* origin is top left (set to 1 for bottom left) */
-	so_method(so, screen->tesla, 0x13ac, 1);
+	so_method(so, screen->tesla, NV50TCL_Y_ORIGIN_BOTTOM, 1);
 	so_data  (so, 0);
 	so_method(so, screen->tesla, NV50TCL_VP_REG_ALLOC_RESULT, 1);
 	so_data  (so, 8);
@@ -360,7 +373,7 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 	//  B = buffer ID (maybe more than 1 byte)
 	//  N = CB index used in shader instruction
 	//  P = program type (0 = VP, 2 = GP, 3 = FP)
-	so_method(so, screen->tesla, 0x1694, 1);
+	so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1);
 	so_data  (so, 0x000BBNP1);
 	*/
 
@@ -424,23 +437,26 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 
 	/* Vertex array limits - max them out */
 	for (i = 0; i < 16; i++) {
-		so_method(so, screen->tesla, NV50TCL_UNK1080_OFFSET_HIGH(i), 2);
+		so_method(so, screen->tesla, NV50TCL_VERTEX_ARRAY_LIMIT_HIGH(i), 2);
 		so_data  (so, 0x000000ff);
 		so_data  (so, 0xffffffff);
 	}
 
-	so_method(so, screen->tesla, NV50TCL_DEPTH_RANGE_NEAR, 2);
+	so_method(so, screen->tesla, NV50TCL_DEPTH_RANGE_NEAR(0), 2);
 	so_data  (so, fui(0.0));
 	so_data  (so, fui(1.0));
 
 	/* no dynamic combination of TIC & TSC entries => only BIND_TIC used */
-	so_method(so, screen->tesla, 0x1234, 1);
+	so_method(so, screen->tesla, NV50TCL_LINKED_TSC, 1);
 	so_data  (so, 1);
 
 	/* activate first scissor rectangle */
-	so_method(so, screen->tesla, NV50TCL_SCISSOR_ENABLE, 1);
+	so_method(so, screen->tesla, NV50TCL_SCISSOR_ENABLE(0), 1);
 	so_data  (so, 1);
 
+	so_method(so, screen->tesla, NV50TCL_EDGEFLAG_ENABLE, 1);
+	so_data  (so, 1); /* default edgeflag to TRUE */
+
 	so_emit(chan, so);
 	so_ref (so, &screen->static_init);
 	so_ref (NULL, &so);
diff --git a/src/gallium/drivers/nv50/nv50_state.c b/src/gallium/drivers/nv50/nv50_state.c
index 88aef52d08..30b2b0f91b 100644
--- a/src/gallium/drivers/nv50/nv50_state.c
+++ b/src/gallium/drivers/nv50/nv50_state.c
@@ -295,7 +295,7 @@ nv50_rasterizer_state_create(struct pipe_context *pipe,
 	so_method(so, tesla, NV50TCL_SHADE_MODEL, 1);
 	so_data  (so, cso->flatshade ? NV50TCL_SHADE_MODEL_FLAT :
 				       NV50TCL_SHADE_MODEL_SMOOTH);
-	so_method(so, tesla, 0x1684, 1);
+	so_method(so, tesla, NV50TCL_PROVOKING_VERTEX_LAST, 1);
 	so_data  (so, cso->flatshade_first ? 0 : 1);
 
 	so_method(so, tesla, NV50TCL_VERTEX_TWO_SIDE_ENABLE, 1);
@@ -392,7 +392,7 @@ nv50_rasterizer_state_create(struct pipe_context *pipe,
 		so_method(so, tesla, NV50TCL_POLYGON_OFFSET_FACTOR, 1);
 		so_data  (so, fui(cso->offset_scale));
 		so_method(so, tesla, NV50TCL_POLYGON_OFFSET_UNITS, 1);
-		so_data  (so, fui(cso->offset_units));
+		so_data  (so, fui(cso->offset_units * 2.0f));
 	}
 
 	rso->pipe = *cso;
@@ -439,9 +439,8 @@ nv50_depth_stencil_alpha_state_create(struct pipe_context *pipe,
 		so_data  (so, 0);
 	}
 
-	/* XXX: keep hex values until header is updated (names reversed) */
 	if (cso->stencil[0].enabled) {
-		so_method(so, tesla, 0x1380, 8);
+		so_method(so, tesla, NV50TCL_STENCIL_FRONT_ENABLE, 8);
 		so_data  (so, 1);
 		so_data  (so, nvgl_stencil_op(cso->stencil[0].fail_op));
 		so_data  (so, nvgl_stencil_op(cso->stencil[0].zfail_op));
@@ -451,23 +450,23 @@ nv50_depth_stencil_alpha_state_create(struct pipe_context *pipe,
 		so_data  (so, cso->stencil[0].writemask);
 		so_data  (so, cso->stencil[0].valuemask);
 	} else {
-		so_method(so, tesla, 0x1380, 1);
+		so_method(so, tesla, NV50TCL_STENCIL_FRONT_ENABLE, 1);
 		so_data  (so, 0);
 	}
 
 	if (cso->stencil[1].enabled) {
-		so_method(so, tesla, 0x1594, 5);
+		so_method(so, tesla, NV50TCL_STENCIL_BACK_ENABLE, 5);
 		so_data  (so, 1);
 		so_data  (so, nvgl_stencil_op(cso->stencil[1].fail_op));
 		so_data  (so, nvgl_stencil_op(cso->stencil[1].zfail_op));
 		so_data  (so, nvgl_stencil_op(cso->stencil[1].zpass_op));
 		so_data  (so, nvgl_comparison_op(cso->stencil[1].func));
-		so_method(so, tesla, 0x0f54, 3);
+		so_method(so, tesla, NV50TCL_STENCIL_BACK_FUNC_REF, 3);
 		so_data  (so, cso->stencil[1].ref_value);
 		so_data  (so, cso->stencil[1].writemask);
 		so_data  (so, cso->stencil[1].valuemask);
 	} else {
-		so_method(so, tesla, 0x1594, 1);
+		so_method(so, tesla, NV50TCL_STENCIL_BACK_ENABLE, 1);
 		so_data  (so, 0);
 	}
 
diff --git a/src/gallium/drivers/nv50/nv50_state_validate.c b/src/gallium/drivers/nv50/nv50_state_validate.c
index 871e8097b6..c8bdf9dc27 100644
--- a/src/gallium/drivers/nv50/nv50_state_validate.c
+++ b/src/gallium/drivers/nv50/nv50_state_validate.c
@@ -41,7 +41,7 @@ nv50_state_validate_fb(struct nv50_context *nv50)
 	 * FP result 0 always goes to RT[0], bits 4 - 6 are ignored.
 	 * Ambiguous assignment results in no rendering (no DATA_ERROR).
 	 */
-	so_method(so, tesla, 0x121c, 1);
+	so_method(so, tesla, NV50TCL_RT_CONTROL, 1);
 	so_data  (so, fb->nr_cbufs |
 		  (0 <<  4) | (1 <<  7) | (2 << 10) | (3 << 13) |
 		  (4 << 16) | (5 << 19) | (6 << 22) | (7 << 25));
@@ -87,7 +87,7 @@ nv50_state_validate_fb(struct nv50_context *nv50)
 				level[fb->cbufs[i]->level].tile_mode << 4);
 		so_data(so, 0x00000000);
 
-		so_method(so, tesla, 0x1224, 1);
+		so_method(so, tesla, NV50TCL_RT_ARRAY_MODE, 1);
 		so_data  (so, 1);
 	}
 
@@ -124,22 +124,22 @@ nv50_state_validate_fb(struct nv50_context *nv50)
 				level[fb->zsbuf->level].tile_mode << 4);
 		so_data(so, 0x00000000);
 
-		so_method(so, tesla, 0x1538, 1);
+		so_method(so, tesla, NV50TCL_ZETA_ENABLE, 1);
 		so_data  (so, 1);
 		so_method(so, tesla, NV50TCL_ZETA_HORIZ, 3);
 		so_data  (so, fb->zsbuf->width);
 		so_data  (so, fb->zsbuf->height);
 		so_data  (so, 0x00010001);
 	} else {
-		so_method(so, tesla, 0x1538, 1);
+		so_method(so, tesla, NV50TCL_ZETA_ENABLE, 1);
 		so_data  (so, 0);
 	}
 
-	so_method(so, tesla, NV50TCL_VIEWPORT_HORIZ, 2);
+	so_method(so, tesla, NV50TCL_VIEWPORT_HORIZ(0), 2);
 	so_data  (so, w << 16);
 	so_data  (so, h << 16);
 	/* set window lower left corner */
-	so_method(so, tesla, NV50TCL_WINDOW_LEFT, 2);
+	so_method(so, tesla, NV50TCL_WINDOW_OFFSET_X, 2);
 	so_data  (so, 0);
 	so_data  (so, 0);
 	/* set screen scissor rectangle */
@@ -325,7 +325,7 @@ nv50_state_validate(struct nv50_context *nv50)
 		nv50->state.scissor_enabled = rast->scissor;
 
 		so = so_new(3, 0);
-		so_method(so, tesla, NV50TCL_SCISSOR_HORIZ, 2);
+		so_method(so, tesla, NV50TCL_SCISSOR_HORIZ(0), 2);
 		if (nv50->state.scissor_enabled) {
 			so_data(so, (s->maxx << 16) | s->minx);
 			so_data(so, (s->maxy << 16) | s->miny);
@@ -355,11 +355,11 @@ scissor_uptodate:
 
 		so = so_new(14, 0);
 		if (!bypass) {
-			so_method(so, tesla, NV50TCL_VIEWPORT_TRANSLATE(0), 3);
+			so_method(so, tesla, NV50TCL_VIEWPORT_TRANSLATE_X(0), 3);
 			so_data  (so, fui(nv50->viewport.translate[0]));
 			so_data  (so, fui(nv50->viewport.translate[1]));
 			so_data  (so, fui(nv50->viewport.translate[2]));
-			so_method(so, tesla, NV50TCL_VIEWPORT_SCALE(0), 3);
+			so_method(so, tesla, NV50TCL_VIEWPORT_SCALE_X(0), 3);
 			so_data  (so, fui(nv50->viewport.scale[0]));
 			so_data  (so, fui(nv50->viewport.scale[1]));
 			so_data  (so, fui(nv50->viewport.scale[2]));
@@ -440,7 +440,7 @@ void nv50_so_init_sifc(struct nv50_context *nv50,
 	so_data  (so, 1);
 	so_reloc (so, bo, offset, reloc | NOUVEAU_BO_HIGH, 0, 0);
 	so_reloc (so, bo, offset, reloc | NOUVEAU_BO_LOW, 0, 0);
-	so_method(so, eng2d, NV50_2D_SIFC_UNK0800, 2);
+	so_method(so, eng2d, NV50_2D_SIFC_BITMAP_ENABLE, 2);
 	so_data  (so, 0);
 	so_data  (so, NV50_2D_SIFC_FORMAT_R8_UNORM);
 	so_method(so, eng2d, NV50_2D_SIFC_WIDTH, 10);
diff --git a/src/gallium/drivers/nv50/nv50_surface.c b/src/gallium/drivers/nv50/nv50_surface.c
index 79655fc08d..6378132979 100644
--- a/src/gallium/drivers/nv50/nv50_surface.c
+++ b/src/gallium/drivers/nv50/nv50_surface.c
@@ -176,11 +176,11 @@ nv50_surface_fill(struct pipe_context *pipe, struct pipe_surface *dest,
 	if (ret)
 		return;
 
-	BEGIN_RING(chan, eng2d, 0x0580, 3);
-	OUT_RING  (chan, 4);
+	BEGIN_RING(chan, eng2d, NV50_2D_DRAW_SHAPE, 3);
+	OUT_RING  (chan, NV50_2D_DRAW_SHAPE_RECTANGLES);
 	OUT_RING  (chan, format);
 	OUT_RING  (chan, value);
-	BEGIN_RING(chan, eng2d, NV50_2D_RECT_X1, 4);
+	BEGIN_RING(chan, eng2d, NV50_2D_DRAW_POINT32_X(0), 4);
 	OUT_RING  (chan, destx);
 	OUT_RING  (chan, desty);
 	OUT_RING  (chan, width);
diff --git a/src/gallium/drivers/nv50/nv50_transfer.c b/src/gallium/drivers/nv50/nv50_transfer.c
index 4d9afa6fed..a2f1db2914 100644
--- a/src/gallium/drivers/nv50/nv50_transfer.c
+++ b/src/gallium/drivers/nv50/nv50_transfer.c
@@ -47,7 +47,7 @@ nv50_transfer_rect_m2mf(struct pipe_screen *pscreen,
 			NV50_MEMORY_TO_MEMORY_FORMAT_LINEAR_IN, 1);
 		OUT_RING  (chan, 1);
 		BEGIN_RING(chan, m2mf,
-			NV50_MEMORY_TO_MEMORY_FORMAT_PITCH_IN, 1);
+			NV04_MEMORY_TO_MEMORY_FORMAT_PITCH_IN, 1);
 		OUT_RING  (chan, src_pitch);
 		src_offset += (sy * src_pitch) + (sx * cpp);
 	} else {
@@ -66,7 +66,7 @@ nv50_transfer_rect_m2mf(struct pipe_screen *pscreen,
 			NV50_MEMORY_TO_MEMORY_FORMAT_LINEAR_OUT, 1);
 		OUT_RING  (chan, 1);
 		BEGIN_RING(chan, m2mf,
-			NV50_MEMORY_TO_MEMORY_FORMAT_PITCH_OUT, 1);
+			NV04_MEMORY_TO_MEMORY_FORMAT_PITCH_OUT, 1);
 		OUT_RING  (chan, dst_pitch);
 		dst_offset += (dy * dst_pitch) + (dx * cpp);
 	} else {
@@ -89,7 +89,7 @@ nv50_transfer_rect_m2mf(struct pipe_screen *pscreen,
 		OUT_RELOCh(chan, src_bo, src_offset, src_reloc);
 		OUT_RELOCh(chan, dst_bo, dst_offset, dst_reloc);
 		BEGIN_RING(chan, m2mf,
-			NV50_MEMORY_TO_MEMORY_FORMAT_OFFSET_IN, 2);
+			NV04_MEMORY_TO_MEMORY_FORMAT_OFFSET_IN, 2);
 		OUT_RELOCl(chan, src_bo, src_offset, src_reloc);
 		OUT_RELOCl(chan, dst_bo, dst_offset, dst_reloc);
 		if (src_bo->tile_flags) {
@@ -107,7 +107,7 @@ nv50_transfer_rect_m2mf(struct pipe_screen *pscreen,
 			dst_offset += (line_count * dst_pitch);
 		}
 		BEGIN_RING(chan, m2mf,
-			NV50_MEMORY_TO_MEMORY_FORMAT_LINE_LENGTH_IN, 4);
+			NV04_MEMORY_TO_MEMORY_FORMAT_LINE_LENGTH_IN, 4);
 		OUT_RING  (chan, width * cpp);
 		OUT_RING  (chan, line_count);
 		OUT_RING  (chan, 0x00000101);
@@ -291,7 +291,7 @@ nv50_upload_sifc(struct nv50_context *nv50,
 
 	/* NV50_2D_OPERATION_SRCCOPY assumed already set */
 
-	BEGIN_RING(chan, eng2d, NV50_2D_SIFC_UNK0800, 2);
+	BEGIN_RING(chan, eng2d, NV50_2D_SIFC_BITMAP_ENABLE, 2);
 	OUT_RING  (chan, 0);
 	OUT_RING  (chan, src_format);
 	BEGIN_RING(chan, eng2d, NV50_2D_SIFC_WIDTH, 10);
@@ -334,6 +334,6 @@ nv50_upload_sifc(struct nv50_context *nv50,
 		src += src_pitch;
 	}
 
-	BEGIN_RING(chan, tesla, 0x1440, 1);
+	BEGIN_RING(chan, tesla, NV50TCL_CODE_CB_FLUSH, 1);
 	OUT_RING  (chan, 0);
 }
diff --git a/src/gallium/drivers/nv50/nv50_vbo.c b/src/gallium/drivers/nv50/nv50_vbo.c
index f7fa0659e8..602adfc50d 100644
--- a/src/gallium/drivers/nv50/nv50_vbo.c
+++ b/src/gallium/drivers/nv50/nv50_vbo.c
@@ -99,19 +99,19 @@ nv50_vbo_size_to_hw(unsigned size, unsigned nr_c)
 {
 	static const uint32_t hw_values[] = {
 		0, 0, 0, 0,
-		NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_8,
-		NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_8_8,
-		NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_8_8_8,
-		NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_8_8_8_8,
-		NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_16,
-		NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_16_16,
-		NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_16_16_16,
-		NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_16_16_16_16,
+		NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_8,
+		NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_8_8,
+		NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_8_8_8,
+		NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_8_8_8_8,
+		NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_16,
+		NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_16_16,
+		NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_16_16_16,
+		NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_16_16_16_16,
 		0, 0, 0, 0,
-		NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_32,
-		NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_32_32,
-		NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_32_32_32,
-		NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_32_32_32_32 };
+		NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_32,
+		NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_32_32,
+		NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_32_32_32,
+		NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_32_32_32_32 };
 
 	/* we'd also have R11G11B10 and R10G10B10A2 */
 
@@ -198,7 +198,7 @@ nv50_draw_elements_inline_u08(struct nv50_context *nv50, uint8_t *map,
 		return nv50_push_elements_u08(nv50, map, count);
 
 	if (count & 1) {
-		BEGIN_RING(chan, tesla, 0x15e8, 1);
+		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U32, 1);
 		OUT_RING  (chan, map[0]);
 		map++;
 		count--;
@@ -208,7 +208,7 @@ nv50_draw_elements_inline_u08(struct nv50_context *nv50, uint8_t *map,
 		unsigned nr = count > 2046 ? 2046 : count;
 		int i;
 
-		BEGIN_RING(chan, tesla, 0x400015f0, nr >> 1);
+		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16 | 0x40000000, nr >> 1);
 		for (i = 0; i < nr; i += 2)
 			OUT_RING  (chan, (map[i + 1] << 16) | map[i]);
 
@@ -231,7 +231,7 @@ nv50_draw_elements_inline_u16(struct nv50_context *nv50, uint16_t *map,
 		return nv50_push_elements_u16(nv50, map, count);
 
 	if (count & 1) {
-		BEGIN_RING(chan, tesla, 0x15e8, 1);
+		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U32, 1);
 		OUT_RING  (chan, map[0]);
 		map++;
 		count--;
@@ -241,7 +241,7 @@ nv50_draw_elements_inline_u16(struct nv50_context *nv50, uint16_t *map,
 		unsigned nr = count > 2046 ? 2046 : count;
 		int i;
 
-		BEGIN_RING(chan, tesla, 0x400015f0, nr >> 1);
+		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16 | 0x40000000, nr >> 1);
 		for (i = 0; i < nr; i += 2)
 			OUT_RING  (chan, (map[i + 1] << 16) | map[i]);
 
@@ -266,7 +266,7 @@ nv50_draw_elements_inline_u32(struct nv50_context *nv50, uint32_t *map,
 	while (count) {
 		unsigned nr = count > 2047 ? 2047 : count;
 
-		BEGIN_RING(chan, tesla, 0x400015e8, nr);
+		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U32 | 0x40000000, nr);
 		OUT_RINGp (chan, map, nr);
 
 		count -= nr;
@@ -372,6 +372,10 @@ nv50_vbo_static_attrib(struct nv50_context *nv50, unsigned attrib,
 		so_data  (so, fui(v[1]));
 		break;
 	case 1:
+		if (attrib == nv50->vertprog->cfg.edgeflag_in) {
+			so_method(so, tesla, NV50TCL_EDGEFLAG_ENABLE, 1);
+			so_data  (so, v[0] ? 1 : 0);
+		}
 		so_method(so, tesla, NV50TCL_VTX_ATTR_1F(attrib), 1);
 		so_data  (so, fui(v[0]));
 		break;
@@ -401,6 +405,9 @@ nv50_vbo_validate(struct nv50_context *nv50)
 		    !(nv50->vtxbuf[i].buffer->usage & PIPE_BUFFER_USAGE_VERTEX))
 			nv50->vbo_fifo = 0xffff;
 
+	if (nv50->vertprog->cfg.edgeflag_in < 16)
+		nv50->vbo_fifo = 0xffff; /* vertprog can't set edgeflag */
+
 	n_ve = MAX2(nv50->vtxelt_nr, nv50->state.vtxelt_nr);
 
 	vtxattr = NULL;
@@ -445,7 +452,7 @@ nv50_vbo_validate(struct nv50_context *nv50)
 			  NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0);
 
 		/* vertex array limits */
-		so_method(vtxbuf, tesla, 0x1080 + (i * 8), 2);
+		so_method(vtxbuf, tesla, NV50TCL_VERTEX_ARRAY_LIMIT_HIGH(i), 2);
 		so_reloc (vtxbuf, bo, vb->buffer->size - 1,
 			  NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD |
 			  NOUVEAU_BO_HIGH, 0, 0);
@@ -479,6 +486,9 @@ struct nv50_vbo_emitctx
 	unsigned nr_ve;
 	unsigned vtx_dwords;
 	unsigned vtx_max;
+
+	float edgeflag;
+	unsigned ve_edgeflag;
 };
 
 static INLINE void
@@ -622,6 +632,9 @@ emit_prepare(struct nv50_context *nv50, struct nv50_vbo_emitctx *emit,
 	if (nv50_map_vbufs(nv50) == FALSE)
 		return FALSE;
 
+	emit->ve_edgeflag = nv50->vertprog->cfg.edgeflag_in;
+
+	emit->edgeflag = 0.5f;
 	emit->nr_ve = 0;
 	emit->vtx_dwords = 0;
 
@@ -644,7 +657,8 @@ emit_prepare(struct nv50_context *nv50, struct nv50_vbo_emitctx *emit,
 		desc = util_format_description(ve->src_format);
 		assert(desc);
 
-		size = util_format_get_component_bits(ve->src_format, UTIL_FORMAT_COLORSPACE_RGB, 0);
+		size = util_format_get_component_bits(
+			ve->src_format, UTIL_FORMAT_COLORSPACE_RGB, 0);
 
 		assert(ve->nr_components > 0 && ve->nr_components <= 4);
 
@@ -686,10 +700,31 @@ emit_prepare(struct nv50_context *nv50, struct nv50_vbo_emitctx *emit,
 	}
 
 	emit->vtx_max = 512 / emit->vtx_dwords;
+	if (emit->ve_edgeflag < 16)
+		emit->vtx_max = 1;
 
 	return TRUE;
 }
 
+static INLINE void
+set_edgeflag(struct nouveau_channel *chan,
+	     struct nouveau_grobj *tesla,
+	     struct nv50_vbo_emitctx *emit, uint32_t index)
+{
+	unsigned i = emit->ve_edgeflag;
+
+	if (i < 16) {
+		float f = *((float *)(emit->map[i] + index * emit->stride[i]));
+
+		if (emit->edgeflag != f) {
+			emit->edgeflag = f;
+
+			BEGIN_RING(chan, tesla, 0x15e4, 1);
+			OUT_RING  (chan, f ? 1 : 0);
+		}
+	}
+}
+
 static boolean
 nv50_push_arrays(struct nv50_context *nv50, unsigned start, unsigned count)
 {
@@ -704,6 +739,8 @@ nv50_push_arrays(struct nv50_context *nv50, unsigned start, unsigned count)
 		unsigned i, dw, nr = MIN2(count, emit.vtx_max);
 	        dw = nr * emit.vtx_dwords;
 
+		set_edgeflag(chan, tesla, &emit, 0); /* nr will be 1 */
+
 		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_DATA | 0x40000000, dw);
 		for (i = 0; i < nr; ++i)
 			emit_vtx_next(chan, &emit);
@@ -729,6 +766,8 @@ nv50_push_elements_u32(struct nv50_context *nv50, uint32_t *map, unsigned count)
 		unsigned i, dw, nr = MIN2(count, emit.vtx_max);
 	        dw = nr * emit.vtx_dwords;
 
+		set_edgeflag(chan, tesla, &emit, *map);
+
 		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_DATA | 0x40000000, dw);
 		for (i = 0; i < nr; ++i)
 			emit_vtx(chan, &emit, *map++);
@@ -754,6 +793,8 @@ nv50_push_elements_u16(struct nv50_context *nv50, uint16_t *map, unsigned count)
 		unsigned i, dw, nr = MIN2(count, emit.vtx_max);
 	        dw = nr * emit.vtx_dwords;
 
+		set_edgeflag(chan, tesla, &emit, *map);
+
 		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_DATA | 0x40000000, dw);
 		for (i = 0; i < nr; ++i)
 			emit_vtx(chan, &emit, *map++);
@@ -779,6 +820,8 @@ nv50_push_elements_u08(struct nv50_context *nv50, uint8_t *map, unsigned count)
 		unsigned i, dw, nr = MIN2(count, emit.vtx_max);
 	        dw = nr * emit.vtx_dwords;
 
+		set_edgeflag(chan, tesla, &emit, *map);
+
 		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_DATA | 0x40000000, dw);
 		for (i = 0; i < nr; ++i)
 			emit_vtx(chan, &emit, *map++);