Mesa (8.0): r600g: fix instance divisor on Cayman

[Marek Olšák]

Module: Mesa
Branch: 8.0
Commit: 1c66286bd6915352eb747cb479b5d651c936c8fe
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=1c66286bd6915352eb747cb479b5d651c936c8fe

Author: Marek Olšák <maraeo at gmail.com>
Date:   Fri Oct 19 21:53:31 2012 +0200

r600g: fix instance divisor on Cayman

NOTE: This is a candidate for the stable branches.
(cherry picked from commit 836325bf7edd797e02ab0717a05ca5c51aa1ac93)

Conflicts:

	src/gallium/drivers/r600/r600_asm.c

---

 src/gallium/drivers/r600/r600_asm.c |   54 ++++++++++++++++++++++------------
 1 files changed, 35 insertions(+), 19 deletions(-)

diff --git a/src/gallium/drivers/r600/r600_asm.c b/src/gallium/drivers/r600/r600_asm.c
index f4977fb..b12a8a3 100644
--- a/src/gallium/drivers/r600/r600_asm.c
+++ b/src/gallium/drivers/r600/r600_asm.c
@@ -2689,7 +2689,7 @@ int r600_vertex_elements_build_fetch_shader(struct r600_pipe_context *rctx, stru
 	unsigned fetch_resource_start = rctx->chip_class >= EVERGREEN ? 0 : 160;
 	unsigned format, num_format, format_comp, endian;
 	u32 *bytecode;
-	int i, r;
+	int i, j, r;
 
 	/* Vertex element offsets need special handling. If the offset is
 	 * bigger than what we can put in the fetch instruction we need to
@@ -2708,24 +2708,40 @@ int r600_vertex_elements_build_fetch_shader(struct r600_pipe_context *rctx, stru
 
 	for (i = 0; i < ve->count; i++) {
 		if (elements[i].instance_divisor > 1) {
-			struct r600_bytecode_alu alu;
-
-			memset(&alu, 0, sizeof(alu));
-			alu.inst = BC_INST(&bc, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MULHI_UINT);
-			alu.src[0].sel = 0;
-			alu.src[0].chan = 3;
-
-			alu.src[1].sel = V_SQ_ALU_SRC_LITERAL;
-			alu.src[1].value = (1ll << 32) / elements[i].instance_divisor + 1;
-
-			alu.dst.sel = i + 1;
-			alu.dst.chan = 3;
-			alu.dst.write = 1;
-			alu.last = 1;
-
-			if ((r = r600_bytecode_add_alu(&bc, &alu))) {
-				r600_bytecode_clear(&bc);
-				return r;
+			if (rctx->chip_class == CAYMAN) {
+				for (j = 0; j < 4; j++) {
+					struct r600_bytecode_alu alu;
+					memset(&alu, 0, sizeof(alu));
+					alu.inst = BC_INST(&bc, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MULHI_UINT);
+					alu.src[0].sel = 0;
+					alu.src[0].chan = 3;
+					alu.src[1].sel = V_SQ_ALU_SRC_LITERAL;
+					alu.src[1].value = (1ll << 32) / elements[i].instance_divisor + 1;
+					alu.dst.sel = i + 1;
+					alu.dst.chan = j;
+					alu.dst.write = j == 3;
+					alu.last = j == 3;
+					if ((r = r600_bytecode_add_alu(&bc, &alu))) {
+						r600_bytecode_clear(&bc);
+						return r;
+					}
+				}
+			} else {
+				struct r600_bytecode_alu alu;
+				memset(&alu, 0, sizeof(alu));
+				alu.inst = BC_INST(&bc, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MULHI_UINT);
+				alu.src[0].sel = 0;
+				alu.src[0].chan = 3;
+				alu.src[1].sel = V_SQ_ALU_SRC_LITERAL;
+				alu.src[1].value = (1ll << 32) / elements[i].instance_divisor + 1;
+				alu.dst.sel = i + 1;
+				alu.dst.chan = 3;
+				alu.dst.write = 1;
+				alu.last = 1;
+				if ((r = r600_bytecode_add_alu(&bc, &alu))) {
+					r600_bytecode_clear(&bc);
+					return r;
+				}
 			}
 		}
 	}