freedreno/ir3: add support for store instructions

For store instructions, the "dst" register is a read register, not a
written register.  (Ie. it is the address to store to.)  Lets not
confuse register allocation, scheduling, etc, with these details.
Instead just leave a dummy instr->regs[0], and take "dst" from
instr->regs[1] and srcs following.

Signed-off-by: Rob Clark <robclark@freedesktop.org>

3 files changed, 43 insertions, 6 deletions

diff --git a/src/gallium/drivers/freedreno/ir3/ir3.c b/src/gallium/drivers/freedreno/ir3/ir3.c
index 6d19a29275b..b24825cff85 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3.c
@@ -499,12 +499,28 @@ static int emit_cat5(struct ir3_instruction *instr, void *ptr,
 static int emit_cat6(struct ir3_instruction *instr, void *ptr,
 		struct ir3_info *info)
 {
-	struct ir3_register *dst  = instr->regs[0];
-	struct ir3_register *src1 = instr->regs[1];
-	struct ir3_register *src2 = (instr->regs_count >= 3) ? instr->regs[2] : NULL;
+	struct ir3_register *dst, *src1, *src2;
 	instr_cat6_t *cat6 = ptr;
 
-	iassert(instr->regs_count >= 2);
+	/* the "dst" for a store instruction is (from the perspective
+	 * of data flow in the shader, ie. register use/def, etc) in
+	 * fact a register that is read by the instruction, rather
+	 * than written:
+	 */
+	if (is_store(instr)) {
+		iassert(instr->regs_count >= 3);
+
+		dst  = instr->regs[1];
+		src1 = instr->regs[2];
+		src2 = (instr->regs_count >= 4) ? instr->regs[3] : NULL;
+	} else {
+		iassert(instr->regs_count >= 2);
+
+		dst  = instr->regs[0];
+		src1 = instr->regs[1];
+		src2 = (instr->regs_count >= 3) ? instr->regs[2] : NULL;
+	}
+
 
 	/* TODO we need a more comprehensive list about which instructions
 	 * can be encoded which way.  Or possibly use IR3_INSTR_0 flag to
diff --git a/src/gallium/drivers/freedreno/ir3/ir3.h b/src/gallium/drivers/freedreno/ir3/ir3.h
index c3b61a0fe01..e68170dec58 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3.h
+++ b/src/gallium/drivers/freedreno/ir3/ir3.h
@@ -554,6 +554,26 @@ is_store(struct ir3_instruction *instr)
 	return false;
 }
 
+static inline bool is_load(struct ir3_instruction *instr)
+{
+	if (is_mem(instr)) {
+		switch (instr->opc) {
+		case OPC_LDG:
+		case OPC_LDL:
+		case OPC_LDP:
+		case OPC_L2G:
+		case OPC_LDLW:
+		case OPC_LDC_4:
+		case OPC_LDLV:
+		/* probably some others too.. */
+			return true;
+		default:
+			break;
+		}
+	}
+	return false;
+}
+
 static inline bool is_input(struct ir3_instruction *instr)
 {
 	/* in some cases, ldlv is used to fetch varying without
@@ -1043,6 +1063,7 @@ ir3_SAM(struct ir3_block *block, opc_t opc, type_t type,
 /* cat6 instructions: */
 INSTR2(6, LDLV)
 INSTR2(6, LDG)
+INSTR3(6, STG)
 
 /* ************************************************************************* */
 /* split this out or find some helper to use.. like main/bitset.h.. */
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_legalize.c b/src/gallium/drivers/freedreno/ir3/ir3_legalize.c
index f4a4223ae17..e94293f6d6b 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_legalize.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_legalize.c
@@ -182,14 +182,14 @@ legalize_block(struct ir3_legalize_ctx *ctx, struct ir3_block *block)
 			 */
 			ctx->has_samp = true;
 			regmask_set(&needs_sy, n->regs[0]);
-		} else if (is_mem(n)) {
+		} else if (is_load(n)) {
 			regmask_set(&needs_sy, n->regs[0]);
 		}
 
 		/* both tex/sfu appear to not always immediately consume
 		 * their src register(s):
 		 */
-		if (is_tex(n) || is_sfu(n) || is_mem(n)) {
+		if (is_tex(n) || is_sfu(n) || is_load(n)) {
 			foreach_src(reg, n) {
 				if (reg_gpr(reg))
 					regmask_set(&needs_ss_war, reg);