r600g: Implement sm5 UBO/sampler indexing

Caveat: Shaders using UBO/sampler indexing will
not be optimized by SB, due to SB not currently
supporting the necessary CF_INDEX_[01] index
registers.

Signed-off-by: Glenn Kennard <[email protected]>

1 files changed, 48 insertions, 4 deletions

diff --git a/src/gallium/drivers/r600/eg_asm.c b/src/gallium/drivers/r600/eg_asm.c
index acb30409428..295cb4d80b7 100644
--- a/src/gallium/drivers/r600/eg_asm.c
+++ b/src/gallium/drivers/r600/eg_asm.c
@@ -43,10 +43,10 @@ int eg_bytecode_cf_build(struct r600_bytecode *bc, struct r600_bytecode_cf *cf)
 			/* prepend ALU_EXTENDED if we need more than 2 kcache sets */
 			if (cf->eg_alu_extended) {
 				bc->bytecode[id++] =
-						S_SQ_CF_ALU_WORD0_EXT_KCACHE_BANK_INDEX_MODE0(V_SQ_CF_INDEX_NONE) |
-						S_SQ_CF_ALU_WORD0_EXT_KCACHE_BANK_INDEX_MODE1(V_SQ_CF_INDEX_NONE) |
-						S_SQ_CF_ALU_WORD0_EXT_KCACHE_BANK_INDEX_MODE2(V_SQ_CF_INDEX_NONE) |
-						S_SQ_CF_ALU_WORD0_EXT_KCACHE_BANK_INDEX_MODE3(V_SQ_CF_INDEX_NONE) |
+						S_SQ_CF_ALU_WORD0_EXT_KCACHE_BANK_INDEX_MODE0(cf->kcache[0].index_mode) |
+						S_SQ_CF_ALU_WORD0_EXT_KCACHE_BANK_INDEX_MODE1(cf->kcache[1].index_mode) |
+						S_SQ_CF_ALU_WORD0_EXT_KCACHE_BANK_INDEX_MODE2(cf->kcache[2].index_mode) |
+						S_SQ_CF_ALU_WORD0_EXT_KCACHE_BANK_INDEX_MODE3(cf->kcache[3].index_mode) |
 						S_SQ_CF_ALU_WORD0_EXT_KCACHE_BANK2(cf->kcache[2].bank) |
 						S_SQ_CF_ALU_WORD0_EXT_KCACHE_BANK3(cf->kcache[3].bank) |
 						S_SQ_CF_ALU_WORD0_EXT_KCACHE_MODE2(cf->kcache[2].mode);
@@ -143,3 +143,47 @@ void eg_bytecode_export_read(struct r600_bytecode *bc,
 	output->comp_mask = G_SQ_CF_ALLOC_EXPORT_WORD1_BUF_COMP_MASK(word1);
 }
 #endif
+
+int egcm_load_index_reg(struct r600_bytecode *bc, unsigned id, bool inside_alu_clause)
+{
+	struct r600_bytecode_alu alu;
+	int r;
+	unsigned type;
+
+	assert(id < 2);
+	assert(bc->chip_class >= EVERGREEN);
+
+	if (bc->index_loaded[id])
+		return 0;
+
+	memset(&alu, 0, sizeof(alu));
+	alu.op = ALU_OP1_MOVA_INT;
+	alu.src[0].sel = bc->index_reg[id];
+	alu.src[0].chan = 0;
+	alu.last = 1;
+	r = r600_bytecode_add_alu(bc, &alu);
+	if (r)
+		return r;
+
+	bc->ar_loaded = 0; /* clobbered */
+
+	memset(&alu, 0, sizeof(alu));
+	alu.op = id == 0 ? ALU_OP0_SET_CF_IDX0 : ALU_OP0_SET_CF_IDX1;
+	alu.last = 1;
+	r = r600_bytecode_add_alu(bc, &alu);
+	if (r)
+		return r;
+
+	/* Must split ALU group as index only applies to following group */
+	if (inside_alu_clause) {
+		type = bc->cf_last->op;
+		if ((r = r600_bytecode_add_cf(bc))) {
+			return r;
+		}
+		bc->cf_last->op = type;
+	}
+
+	bc->index_loaded[id] = 1;
+
+	return 0;
+}