freedreno/ir3/cp: support to swap mad src's

For a normal MAD (ie. not MADSH), if first source is gpr and second
source is const, we can swap the first two sources to avoid needing a
mov instruction.

This gives back the biggest advantage TGSI f/e had over NIR f/e for
common shaders, since TGSI f/e had this logic in the f/e.  Note that
doing this in copy-prop step has the advantage that it will also work
for cases like:

   MOV TEMP[b], CONST[x]
   MAD TEMP[d], TEMP[a], TEMP[b], TEMP[c]

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

4 files changed, 43 insertions, 9 deletions

diff --git a/src/gallium/drivers/freedreno/ir3/instr-a3xx.h b/src/gallium/drivers/freedreno/ir3/instr-a3xx.h
index 4d75d77..98637c7 100644
--- a/src/gallium/drivers/freedreno/ir3/instr-a3xx.h
+++ b/src/gallium/drivers/freedreno/ir3/instr-a3xx.h
@@ -676,9 +676,7 @@ static inline bool is_mad(opc_t opc)
 {
 	switch (opc) {
 	case OPC_MAD_U16:
-	case OPC_MADSH_U16:
 	case OPC_MAD_S16:
-	case OPC_MADSH_M16:
 	case OPC_MAD_U24:
 	case OPC_MAD_S24:
 	case OPC_MAD_F16:
@@ -689,4 +687,15 @@ static inline bool is_mad(opc_t opc)
 	}
 }
 
+static inline bool is_madsh(opc_t opc)
+{
+	switch (opc) {
+	case OPC_MADSH_U16:
+	case OPC_MADSH_M16:
+		return true;
+	default:
+		return false;
+	}
+}
+
 #endif /* INSTR_A3XX_H_ */
diff --git a/src/gallium/drivers/freedreno/ir3/ir3.h b/src/gallium/drivers/freedreno/ir3/ir3.h
index f424f73..1a8bead 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3.h
+++ b/src/gallium/drivers/freedreno/ir3/ir3.h
@@ -540,10 +540,10 @@ static inline bool reg_gpr(struct ir3_register *r)
 	return true;
 }
 
-/* some cat2 instructions (ie. those which are not float can embed an
+/* some cat2 instructions (ie. those which are not float) can embed an
  * immediate:
  */
-static inline bool ir3_cat2_immed(opc_t opc)
+static inline bool ir3_cat2_int(opc_t opc)
 {
 	switch (opc) {
 	case OPC_ADD_U:
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_cp.c b/src/gallium/drivers/freedreno/ir3/ir3_cp.c
index 3eb85f6..77bfbc5 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_cp.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_cp.c
@@ -115,7 +115,7 @@ static bool valid_flags(struct ir3_instruction *instr, unsigned n,
 	case 2:
 		valid_flags = ir3_cat2_absneg(instr->opc) | IR3_REG_CONST;
 
-		if (ir3_cat2_immed(instr->opc))
+		if (ir3_cat2_int(instr->opc))
 			valid_flags |= IR3_REG_IMMED;
 
 		if (flags & ~valid_flags)
@@ -199,6 +199,15 @@ static void combine_flags(unsigned *dstflags, unsigned srcflags)
 
 static struct ir3_instruction * instr_cp(struct ir3_instruction *instr, unsigned *flags);
 
+/* the "plain" MAD's (ie. the ones that don't shift first src prior to
+ * multiply) can swap their first two srcs if src[0] is !CONST and
+ * src[1] is CONST:
+ */
+static bool is_valid_mad(struct ir3_instruction *instr)
+{
+	return (instr->category == 3) && is_mad(instr->opc);
+}
+
 /**
  * Handle cp for a given src register.  This additionally handles
  * the cases of collapsing immedate/const (which replace the src
@@ -255,8 +264,23 @@ reg_cp(struct ir3_instruction *instr, struct ir3_register *reg, unsigned n)
 
 		combine_flags(&new_flags, reg->flags);
 
-		if (!valid_flags(instr, n, new_flags))
-			return;
+		if (!valid_flags(instr, n, new_flags)) {
+			/* special case for "normal" mad instructions, we can
+			 * try swapping the first two args if that fits better.
+			 */
+			if ((n == 1) && is_valid_mad(instr) &&
+					!(instr->regs[0 + 1]->flags & IR3_REG_CONST) &&
+					valid_flags(instr, 0, new_flags)) {
+				/* swap src[0] and src[1]: */
+				struct ir3_register *tmp;
+				tmp = instr->regs[0 + 1];
+				instr->regs[0 + 1] = instr->regs[1 + 1];
+				instr->regs[1 + 1] = tmp;
+				n = 0;
+			} else {
+				return;
+			}
+		}
 
 		/* Here we handle the special case of mov from
 		 * CONST and/or RELATIV.  These need to be handled
@@ -305,7 +329,7 @@ reg_cp(struct ir3_instruction *instr, struct ir3_register *reg, unsigned n)
 
 			debug_assert((instr->category == 6) ||
 					((instr->category == 2) &&
-						ir3_cat2_immed(instr->opc)));
+						ir3_cat2_int(instr->opc)));
 
 			if (new_flags & IR3_REG_SABS)
 				iim_val = abs(iim_val);
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_depth.c b/src/gallium/drivers/freedreno/ir3/ir3_depth.c
index 0cda62b..9e1f45d 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_depth.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_depth.c
@@ -74,7 +74,8 @@ int ir3_delayslots(struct ir3_instruction *assigner,
 	if (is_flow(consumer) || is_sfu(consumer) || is_tex(consumer)) {
 		return 6;
 	} else if ((consumer->category == 3) &&
-			is_mad(consumer->opc) && (n == 2)) {
+			(is_mad(consumer->opc) || is_madsh(consumer->opc)) &&
+			(n == 2)) {
 		/* special case, 3rd src to cat3 not required on first cycle */
 		return 1;
 	} else {