commit e611bc623c1526d5c806964287edf31b4b346d0d
Author: chr <chr@ECHELON.(none)>
Date:   Sat May 16 16:50:23 2009 +0200

    - Unify moving result from temporary to destination registers.
    - Don't do mov and cvt.sat for MOV_SAT, just cvt.sat suffices.

diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c
index ed9e20b..4a03cf5 100644
--- a/src/gallium/drivers/nv50/nv50_program.c
+++ b/src/gallium/drivers/nv50/nv50_program.c
@@ -221,7 +221,7 @@ alloc_preferred_temp(struct nv50_pc *pc, int hw)
 static void
 assimilate_temp(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src)
 {
-	assert(dst->index != -1 && src->index == -1 && src->hw != -1);
+	assert(src->index == -1 && src->hw != -1);
 
 	if (dst->hw != -1)
 		pc->r_temp[dst->hw] = NULL;
@@ -1311,22 +1311,12 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok)
 		temp = alloc_temp(pc, NULL);
 		emit_precossin(pc, temp, src[0][0]);
 		emit_flop(pc, 5, temp, temp);
-		for (c = 0; c < 4; c++) {
-			if (!(mask & (1 << c)))
-				continue;
-			emit_mov(pc, dst[c], temp);
-		}
 		break;
 	case TGSI_OPCODE_DP3:
 		temp = alloc_temp(pc, NULL);
 		emit_mul(pc, temp, src[0][0], src[1][0]);
 		emit_mad(pc, temp, src[0][1], src[1][1], temp);
 		emit_mad(pc, temp, src[0][2], src[1][2], temp);
-		for (c = 0; c < 4; c++) {
-			if (!(mask & (1 << c)))
-				continue;
-			emit_mov(pc, dst[c], temp);
-		}
 		break;
 	case TGSI_OPCODE_DP4:
 		temp = alloc_temp(pc, NULL);
@@ -1334,11 +1324,6 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok)
 		emit_mad(pc, temp, src[0][1], src[1][1], temp);
 		emit_mad(pc, temp, src[0][2], src[1][2], temp);
 		emit_mad(pc, temp, src[0][3], src[1][3], temp);
-		for (c = 0; c < 4; c++) {
-			if (!(mask & (1 << c)))
-				continue;
-			emit_mov(pc, dst[c], temp);
-		}
 		break;
 	case TGSI_OPCODE_DPH:
 		temp = alloc_temp(pc, NULL);
@@ -1346,11 +1331,6 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok)
 		emit_mad(pc, temp, src[0][1], src[1][1], temp);
 		emit_mad(pc, temp, src[0][2], src[1][2], temp);
 		emit_add(pc, temp, src[1][3], temp);
-		for (c = 0; c < 4; c++) {
-			if (!(mask & (1 << c)))
-				continue;
-			emit_mov(pc, dst[c], temp);
-		}
 		break;
 	case TGSI_OPCODE_DST:
 	{
@@ -1370,11 +1350,6 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok)
 		temp = alloc_temp(pc, NULL);
 		emit_preex2(pc, temp, src[0][0]);
 		emit_flop(pc, 6, temp, temp);
-		for (c = 0; c < 4; c++) {
-			if (!(mask & (1 << c)))
-				continue;
-			emit_mov(pc, dst[c], temp);
-		}
 		break;
 	case TGSI_OPCODE_FLR:
 		for (c = 0; c < 4; c++) {
@@ -1405,11 +1380,6 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok)
 	case TGSI_OPCODE_LG2:
 		temp = alloc_temp(pc, NULL);
 		emit_flop(pc, 3, temp, src[0][0]);
-		for (c = 0; c < 4; c++) {
-			if (!(mask & (1 << c)))
-				continue;
-			emit_mov(pc, dst[c], temp);
-		}
 		break;
 	case TGSI_OPCODE_LRP:
 		temp = alloc_temp(pc, NULL);
@@ -1419,6 +1389,8 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok)
 			emit_sub(pc, temp, src[1][c], src[2][c]);
 			emit_mad(pc, dst[c], temp, src[0][c], src[2][c]);
 		}
+		free_temp(pc, temp);
+		temp = NULL;
 		break;
 	case TGSI_OPCODE_MAD:
 		for (c = 0; c < 4; c++) {
@@ -1442,6 +1414,13 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok)
 		}
 		break;
 	case TGSI_OPCODE_MOV:
+		if (sat) {
+			dst[0] = src[0][0];
+			dst[1] = src[0][1];
+			dst[2] = src[0][2];
+			dst[3] = src[0][3];
+			break;
+		}
 		for (c = 0; c < 4; c++) {
 			if (!(mask & (1 << c)))
 				continue;
@@ -1458,11 +1437,6 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok)
 	case TGSI_OPCODE_POW:
 		temp = alloc_temp(pc, NULL);
 		emit_pow(pc, temp, src[0][0], src[1][0]);
-		for (c = 0; c < 4; c++) {
-			if (!(mask & (1 << c)))
-				continue;
-			emit_mov(pc, dst[c], temp);
-		}
 		break;
 	case TGSI_OPCODE_RCP:
 		for (c = 0; c < 4; c++) {
@@ -1493,6 +1467,8 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok)
 			emit_flop(pc, 5, dst[0], temp);
 		if (mask & (1 << 1))
 			emit_flop(pc, 4, dst[1], temp);
+		free_temp(pc, temp);
+		temp = NULL;
 		break;
 	case TGSI_OPCODE_SGE:
 		for (c = 0; c < 4; c++) {
@@ -1505,11 +1481,6 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok)
 		temp = alloc_temp(pc, NULL);
 		emit_precossin(pc, temp, src[0][0]);
 		emit_flop(pc, 4, temp, temp);
-		for (c = 0; c < 4; c++) {
-			if (!(mask & (1 << c)))
-				continue;
-			emit_mov(pc, dst[c], temp);
-		}
 		break;
 	case TGSI_OPCODE_SLT:
 		for (c = 0; c < 4; c++) {
@@ -1573,8 +1544,24 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok)
 		return FALSE;
 	}
 
-	if (temp)
-		free_temp(pc, temp);
+	i = -1;
+	if (temp) {
+		for (c = 0; c < 4; c++) {
+			if (!(mask & (1 << c)))
+				continue;
+			if (i >= 0)
+				emit_mov(pc, dst[c], dst[i]);
+			else if (dst[c]->type == P_TEMP) {
+				assimilate_temp(pc, dst[c], temp);
+				i = c;
+				temp = NULL;
+			} else
+				emit_mov(pc, dst[c], temp);
+		}
+
+		if (temp)
+			free_temp(pc, temp);
+	}
 
 	if (sat) {
 		for (c = 0; c < 4; c++) {