Mesa (master): freedreno/ir3: track "keeps" in ir

[Rob Clark]

Module: Mesa
Branch: master
Commit: 96d4db683f90f02e72d34ece544de7eedfa873ee
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=96d4db683f90f02e72d34ece544de7eedfa873ee

Author: Rob Clark <robclark at freedesktop.org>
Date:   Sat Jul 25 13:51:16 2015 -0400

freedreno/ir3: track "keeps" in ir

Previously we had a fixed array to track kills, since they don't
generate an SSA value, and then cheated by stuffing them in the
outputs array before sending things through depth/sched/etc.  But
store instructions will need similar treatment.  So convert this
over to a more general array of instructions that must be kept
and fix up the places that were previously relying on kills being
in the output array.

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

---

 src/gallium/drivers/freedreno/ir3/ir3.h            |    6 +++++
 .../drivers/freedreno/ir3/ir3_compiler_nir.c       |   27 +++-----------------
 src/gallium/drivers/freedreno/ir3/ir3_cp.c         |    4 +++
 src/gallium/drivers/freedreno/ir3/ir3_depth.c      |    3 +++
 4 files changed, 17 insertions(+), 23 deletions(-)

diff --git a/src/gallium/drivers/freedreno/ir3/ir3.h b/src/gallium/drivers/freedreno/ir3/ir3.h
index e68170d..12f2ebe 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3.h
+++ b/src/gallium/drivers/freedreno/ir3/ir3.h
@@ -369,6 +369,12 @@ struct ir3 {
 	unsigned predicates_count, predicates_sz;
 	struct ir3_instruction **predicates;
 
+	/* Track instructions which do not write a register but other-
+	 * wise must not be discarded (such as kill, stg, etc)
+	 */
+	unsigned keeps_count, keeps_sz;
+	struct ir3_instruction **keeps;
+
 	/* List of blocks: */
 	struct list_head block_list;
 
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c b/src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c
index e013abe..a4b2785 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c
@@ -117,10 +117,6 @@ struct ir3_compile {
 	/* for looking up which system value is which */
 	unsigned sysval_semantics[8];
 
-	/* list of kill instructions: */
-	struct ir3_instruction *kill[16];
-	unsigned int kill_count;
-
 	/* set if we encounter something we can't handle yet, so we
 	 * can bail cleanly and fallback to TGSI compiler f/e
 	 */
@@ -1481,7 +1477,7 @@ emit_intrinisic(struct ir3_compile *ctx, nir_intrinsic_instr *intr)
 		kill = ir3_KILL(b, cond, 0);
 		array_insert(ctx->ir->predicates, kill);
 
-		ctx->kill[ctx->kill_count++] = kill;
+		array_insert(ctx->ir->keeps, kill);
 		ctx->so->has_kill = true;
 
 		break;
@@ -2165,13 +2161,9 @@ emit_instructions(struct ir3_compile *ctx)
 	ninputs  = exec_list_length(&ctx->s->inputs) * 4;
 	noutputs = exec_list_length(&ctx->s->outputs) * 4;
 
-	/* we need to allocate big enough outputs array so that
-	 * we can stuff the kill's at the end.  Likewise for vtx
-	 * shaders, we need to leave room for sysvals:
+	/* or vtx shaders, we need to leave room for sysvals:
 	 */
-	if (ctx->so->type == SHADER_FRAGMENT) {
-		noutputs += ARRAY_SIZE(ctx->kill);
-	} else if (ctx->so->type == SHADER_VERTEX) {
+	if (ctx->so->type == SHADER_VERTEX) {
 		ninputs += 8;
 	}
 
@@ -2182,9 +2174,7 @@ emit_instructions(struct ir3_compile *ctx)
 	ctx->in_block = ctx->block;
 	list_addtail(&ctx->block->node, &ctx->ir->block_list);
 
-	if (ctx->so->type == SHADER_FRAGMENT) {
-		ctx->ir->noutputs -= ARRAY_SIZE(ctx->kill);
-	} else if (ctx->so->type == SHADER_VERTEX) {
+	if (ctx->so->type == SHADER_VERTEX) {
 		ctx->ir->ninputs -= 8;
 	}
 
@@ -2380,15 +2370,6 @@ ir3_compile_shader_nir(struct ir3_compiler *compiler,
 		}
 	}
 
-	/* at this point, we want the kill's in the outputs array too,
-	 * so that they get scheduled (since they have no dst).. we've
-	 * already ensured that the array is big enough in push_block():
-	 */
-	if (so->type == SHADER_FRAGMENT) {
-		for (i = 0; i < ctx->kill_count; i++)
-			ir->outputs[ir->noutputs++] = ctx->kill[i];
-	}
-
 	if (fd_mesa_debug & FD_DBG_OPTMSGS) {
 		printf("BEFORE CP:\n");
 		ir3_print(ir);
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_cp.c b/src/gallium/drivers/freedreno/ir3/ir3_cp.c
index f4c825b..be4e4e8 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_cp.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_cp.c
@@ -408,6 +408,10 @@ ir3_cp(struct ir3 *ir)
 		}
 	}
 
+	for (unsigned i = 0; i < ir->keeps_count; i++) {
+		ir->keeps[i] = instr_cp(ir->keeps[i], NULL);
+	}
+
 	list_for_each_entry (struct ir3_block, block, &ir->block_list, node) {
 		if (block->condition)
 			block->condition = instr_cp(block->condition, NULL);
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_depth.c b/src/gallium/drivers/freedreno/ir3/ir3_depth.c
index 0f346b2..97df0c2 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_depth.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_depth.c
@@ -156,6 +156,9 @@ ir3_depth(struct ir3 *ir)
 		if (ir->outputs[i])
 			ir3_instr_depth(ir->outputs[i]);
 
+	for (i = 0; i < ir->keeps_count; i++)
+		ir3_instr_depth(ir->keeps[i]);
+
 	/* We also need to account for if-condition: */
 	list_for_each_entry (struct ir3_block, block, &ir->block_list, node) {
 		if (block->condition)