[Mesa-dev] [PATCH] r600g: Fix access to constants > 31.

[Tilman Sauerbeck]

Signed-off-by: Tilman Sauerbeck <tilman at code-monkey.de>
---

This is pretty ugly, but hopefully it will make a proper fix easier
to implement :)

 src/gallium/drivers/r600/r600_asm.c    |   30 +++++++++++++++++++++++++++++-
 src/gallium/drivers/r600/r600_asm.h    |    1 +
 src/gallium/drivers/r600/r600_shader.c |    7 ++++++-
 3 files changed, 36 insertions(+), 2 deletions(-)

diff --git a/src/gallium/drivers/r600/r600_asm.c b/src/gallium/drivers/r600/r600_asm.c
index ba1471e..fc85432 100644
--- a/src/gallium/drivers/r600/r600_asm.c
+++ b/src/gallium/drivers/r600/r600_asm.c
@@ -410,6 +410,7 @@ int r600_bc_add_alu_type(struct r600_bc *bc, const struct r600_bc_alu *alu, int
 {
 	struct r600_bc_alu *nalu = r600_bc_alu();
 	struct r600_bc_alu *lalu;
+	unsigned kcache1_addr = 0;
 	int i, r;
 
 	if (nalu == NULL)
@@ -440,13 +441,37 @@ int r600_bc_add_alu_type(struct r600_bc *bc, const struct r600_bc_alu *alu, int
 	}
 	/* number of gpr == the last gpr used in any alu */
 	for (i = 0; i < 3; i++) {
+		unsigned const_index;
+
 		if (alu->src[i].sel >= bc->ngpr && alu->src[i].sel < 128) {
 			bc->ngpr = alu->src[i].sel + 1;
 		}
+
+		if (alu->src[i].is_const) {
+			const_index = alu->src[i].sel - 128;
+			unsigned want_addr;
+
+			/* Constants 0-31 will be read from the first
+			 * set of constants (kcache0_*).
+			 *
+			 * Constants > 31 will be read from the second set
+			 * of constants (kcache1_*).
+			 */
+			if (const_index > 31) {
+				want_addr = (const_index / 32) * 2;
+
+				if (!kcache1_addr || kcache1_addr == want_addr) {
+					kcache1_addr = want_addr;
+					nalu->src[i].sel = 160 + (const_index % 32);
+				} else
+					fprintf(stderr, "kcache1_addr already in use (want_addr %i, is %i)\n", want_addr, kcache1_addr);
+			}
+		}
+
 		/* compute how many literal are needed
 		 * either 2 or 4 literals
 		 */
-		if (alu->src[i].sel == 253) {
+		if (nalu->src[i].sel == 253) {
 			if (((alu->src[i].chan + 2) & 0x6) > nalu->nliteral) {
 				nalu->nliteral = (alu->src[i].chan + 2) & 0x6;
 			}
@@ -467,6 +492,9 @@ int r600_bc_add_alu_type(struct r600_bc *bc, const struct r600_bc_alu *alu, int
 	bc->ndw += 2;
 
 	bc->cf_last->kcache0_mode = 2;
+	bc->cf_last->kcache1_mode = 2;
+	bc->cf_last->kcache0_addr = 0;
+	bc->cf_last->kcache1_addr = kcache1_addr;
 
 	/* process cur ALU instructions for bank swizzle */
 	if (alu->last) {
diff --git a/src/gallium/drivers/r600/r600_asm.h b/src/gallium/drivers/r600/r600_asm.h
index f2016af..50e5052 100644
--- a/src/gallium/drivers/r600/r600_asm.h
+++ b/src/gallium/drivers/r600/r600_asm.h
@@ -34,6 +34,7 @@ struct r600_bc_alu_src {
 	unsigned			neg;
 	unsigned			abs;
 	unsigned			rel;
+	unsigned			is_const;
 };
 
 struct r600_bc_alu_dst {
diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
index c46658a..b733ddb 100644
--- a/src/gallium/drivers/r600/r600_shader.c
+++ b/src/gallium/drivers/r600/r600_shader.c
@@ -880,7 +880,10 @@ static int tgsi_src(struct r600_shader_ctx *ctx,
 	int index;
 	memset(r600_src, 0, sizeof(struct r600_bc_alu_src));
 	r600_src->sel = tgsi_src->Register.Index;
-	if (tgsi_src->Register.File == TGSI_FILE_IMMEDIATE) {
+
+	if (tgsi_src->Register.File == TGSI_FILE_CONSTANT) {
+		r600_src->is_const = 1;
+	} else if (tgsi_src->Register.File == TGSI_FILE_IMMEDIATE) {
 		r600_src->sel = 0;
 		index = tgsi_src->Register.Index;
 		ctx->value[0] = ctx->literals[index * 4 + 0];
@@ -953,6 +956,7 @@ static int tgsi_split_constant(struct r600_shader_ctx *ctx, struct r600_bc_alu_s
 				memset(&alu, 0, sizeof(struct r600_bc_alu));
 				alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV);
 				alu.src[0].sel = r600_src[i].sel;
+				alu.src[0].is_const = 1;
 				alu.src[0].chan = k;
 				alu.src[0].rel = r600_src[i].rel;
 				alu.dst.sel = treg;
@@ -965,6 +969,7 @@ static int tgsi_split_constant(struct r600_shader_ctx *ctx, struct r600_bc_alu_s
 					return r;
 			}
 			r600_src[i].sel = treg;
+			r600_src[i].is_const = 0;
 			r600_src[i].rel =0;
 			j--;
 		}
-- 
1.7.3.1