freedreno/ir3: large const support

Signed-off-by: Rob Clark <[email protected]>

3 files changed, 20 insertions, 7 deletions

diff --git a/src/gallium/drivers/freedreno/ir3/ir3.c b/src/gallium/drivers/freedreno/ir3/ir3.c
index 3da10fb81e3..70d37ffdeb5 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3.c
@@ -104,7 +104,7 @@ static uint32_t reg(struct ir3_register *reg, struct ir3_info *info,
 		val.iim_val = reg->iim_val;
 	} else {
 		int8_t components = util_last_bit(reg->wrmask);
-		int8_t max = (reg->num + repeat + components - 1) >> 2;
+		int16_t max = (reg->num + repeat + components - 1) >> 2;
 
 		val.comp = reg->num & 0x3;
 		val.num  = reg->num >> 2;
diff --git a/src/gallium/drivers/freedreno/ir3/ir3.h b/src/gallium/drivers/freedreno/ir3/ir3.h
index b92a57a43f8..d2d3dcaadb9 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3.h
+++ b/src/gallium/drivers/freedreno/ir3/ir3.h
@@ -47,7 +47,7 @@ struct ir3_info {
 	 */
 	int8_t   max_reg;   /* highest GPR # used by shader */
 	int8_t   max_half_reg;
-	int8_t   max_const;
+	int16_t  max_const;
 };
 
 struct ir3_register {
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_compiler.c b/src/gallium/drivers/freedreno/ir3/ir3_compiler.c
index 1a5119c074f..8c4ec88ccc0 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_compiler.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_compiler.c
@@ -600,11 +600,6 @@ add_src_reg_wrmask(struct ir3_compile_context *ctx,
 	struct ir3_register *reg;
 	struct ir3_instruction *orig = NULL;
 
-	/* TODO we need to use a mov to temp for const >= 64.. or maybe
-	 * we could use relative addressing..
-	 */
-	compile_assert(ctx, src->Index < 64);
-
 	switch (src->File) {
 	case TGSI_FILE_IMMEDIATE:
 		/* TODO if possible, use actual immediate instead of const.. but
@@ -632,6 +627,24 @@ add_src_reg_wrmask(struct ir3_compile_context *ctx,
 		break;
 	}
 
+	/* We seem to have 8 bits (6.2) for dst register always, so I think
+	 * it is safe to assume GPR cannot be >=64
+	 *
+	 * cat3 instructions only have 8 bits for src2, but cannot take a
+	 * const for src2
+	 *
+	 * cat5 and cat6 in some cases only has 8 bits, but cannot take a
+	 * const for any src.
+	 *
+	 * Other than that we seem to have 12 bits to encode const src,
+	 * except for cat1 which may only have 11 bits (but that seems like
+	 * a bug)
+	 */
+	if (flags & IR3_REG_CONST)
+		compile_assert(ctx, src->Index < (1 << 9));
+	else
+		compile_assert(ctx, src->Index < (1 << 6));
+
 	if (src->Absolute)
 		flags |= IR3_REG_ABS;
 	if (src->Negate)