Mesa (master): freedreno/ir3: Apply the max upload limit to initial range setup

[GitLab Mirror]

Module: Mesa
Branch: master
Commit: 41b5aafef341f8086be16901c15d4325a3be3359
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=41b5aafef341f8086be16901c15d4325a3be3359

Author: Eric Anholt <eric at anholt.net>
Date:   Fri Sep  4 10:59:37 2020 -0700

freedreno/ir3: Apply the max upload limit to initial range setup

There's no sense in planning out an upload that we won't be able to
actually upload due to the limit.  This means that we can keep making
other loads pushable, even after we find one that wouldn't be, and we
don't fill the const file with UBO data for a load we couldn't promote.

total instructions in shared programs: 8096655 -> 8044344 (-0.65%)
total constlen in shared programs: 1447824 -> 1411384 (-2.52%)
total cat6 in shared programs: 97824 -> 89983 (-8.02%)

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6359>

---

 src/freedreno/ir3/ir3_nir_analyze_ubo_ranges.c | 52 +++++++++++++++-----------
 1 file changed, 31 insertions(+), 21 deletions(-)

diff --git a/src/freedreno/ir3/ir3_nir_analyze_ubo_ranges.c b/src/freedreno/ir3/ir3_nir_analyze_ubo_ranges.c
index 57c00779465..1a40058261e 100644
--- a/src/freedreno/ir3/ir3_nir_analyze_ubo_ranges.c
+++ b/src/freedreno/ir3/ir3_nir_analyze_ubo_ranges.c
@@ -124,7 +124,8 @@ get_range(nir_intrinsic_instr *instr, struct ir3_ubo_analysis_state *state)
 
 static void
 gather_ubo_ranges(nir_shader *nir, nir_intrinsic_instr *instr,
-				  struct ir3_ubo_analysis_state *state, uint32_t alignment)
+		struct ir3_ubo_analysis_state *state, uint32_t alignment,
+		uint32_t *upload_remaining)
 {
 	if (ir3_shader_debug & IR3_DBG_NOUBOOPT)
 		return;
@@ -137,10 +138,20 @@ gather_ubo_ranges(nir_shader *nir, nir_intrinsic_instr *instr,
 	if (!get_ubo_load_range(nir, instr, alignment, &r))
 		return;
 
-	if (r.start < old_r->start)
-		old_r->start = r.start;
-	if (old_r->end < r.end)
-		old_r->end = r.end;
+	r.start = MIN2(r.start, old_r->start);
+	r.end = MAX2(r.end, old_r->end);
+
+	/* If adding this range would definitely put us over the limit, don't try.
+	 * Prevents sparse access or large indirects from occupying all the
+	 * planned UBO space
+	 */
+	uint32_t added = (old_r->start - r.start) + (r.end - old_r->end);
+	if (added >= *upload_remaining)
+		return;
+
+	*upload_remaining -= added;
+	old_r->start = r.start;
+	old_r->end = r.end;
 }
 
 /* For indirect offset, it is common to see a pattern of multiple
@@ -329,18 +340,31 @@ ir3_nir_analyze_ubo_ranges(nir_shader *nir, struct ir3_shader_variant *v)
 	struct ir3_ubo_analysis_state *state = &const_state->ubo_state;
 	struct ir3_compiler *compiler = v->shader->compiler;
 
+	/* Limit our uploads to the amount of constant buffer space available in
+	 * the hardware, minus what the shader compiler may need for various
+	 * driver params.  We do this UBO-to-push-constant before the real
+	 * allocation of the driver params' const space, because UBO pointers can
+	 * be driver params but this pass usually eliminatings them.
+	 */
+	struct ir3_const_state worst_case_const_state = { };
+	ir3_setup_const_state(nir, v, &worst_case_const_state);
+	const uint32_t max_upload = (ir3_max_const(v) -
+			worst_case_const_state.offsets.immediate) * 16;
+
 	memset(state, 0, sizeof(*state));
 	for (int i = 0; i < IR3_MAX_UBO_PUSH_RANGES; i++) {
 		state->range[i].start = UINT32_MAX;
 	}
 
+	uint32_t upload_remaining = max_upload;
 	nir_foreach_function (function, nir) {
 		if (function->impl) {
 			nir_foreach_block (block, function->impl) {
 				nir_foreach_instr (instr, block) {
 					if (instr_is_load_ubo(instr))
 						gather_ubo_ranges(nir, nir_instr_as_intrinsic(instr),
-								state, compiler->const_upload_unit);
+								state, compiler->const_upload_unit,
+								&upload_remaining);
 				}
 			}
 		}
@@ -355,17 +379,6 @@ ir3_nir_analyze_ubo_ranges(nir_shader *nir, struct ir3_shader_variant *v)
 	 * first.
 	 */
 
-	/* Limit our uploads to the amount of constant buffer space available in
-	 * the hardware, minus what the shader compiler may need for various
-	 * driver params.  We do this UBO-to-push-constant before the real
-	 * allocation of the driver params' const space, because UBO pointers can
-	 * be driver params but this pass usually eliminatings them.
-	 */
-	struct ir3_const_state worst_case_const_state = { };
-	ir3_setup_const_state(nir, v, &worst_case_const_state);
-	const uint32_t max_upload = (ir3_max_const(v) -
-			worst_case_const_state.offsets.immediate) * 16;
-
 	uint32_t offset = v->shader->num_reserved_user_consts * 16;
 	state->num_enabled = ARRAY_SIZE(state->range);
 	for (uint32_t i = 0; i < ARRAY_SIZE(state->range); i++) {
@@ -378,10 +391,7 @@ ir3_nir_analyze_ubo_ranges(nir_shader *nir, struct ir3_shader_variant *v)
 
 		debug_assert(offset <= max_upload);
 		state->range[i].offset = offset;
-		if (offset + range_size > max_upload) {
-			range_size = max_upload - offset;
-			state->range[i].end = state->range[i].start + range_size;
-		}
+		assert(offset <= max_upload);
 		offset += range_size;
 
 	}