freedreno/ir3: add Sethi–Ullman numbering pass

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

7 files changed, 141 insertions, 4 deletions

diff --git a/src/freedreno/ir3/ir3.h b/src/freedreno/ir3/ir3.h
index 53e56edb3c4..e38ef9fcb66 100644
--- a/src/freedreno/ir3/ir3.h
+++ b/src/freedreno/ir3/ir3.h
@@ -29,8 +29,9 @@
 
 #include "compiler/shader_enums.h"
 
-#include "util/u_debug.h"
+#include "util/bitscan.h"
 #include "util/list.h"
+#include "util/u_debug.h"
 
 #include "instr-a3xx.h"
 
@@ -292,6 +293,9 @@ struct ir3_instruction {
 	 */
 	void *data;
 
+	int sun;            /* Sethi–Ullman number, used by sched */
+	int use_count;      /* currently just updated/used by cp */
+
 	/* Used during CP and RA stages.  For fanin and shader inputs/
 	 * outputs where we need a sequence of consecutive registers,
 	 * keep track of each src instructions left (ie 'n-1') and right
@@ -363,8 +367,6 @@ struct ir3_instruction {
 	/* Entry in ir3_block's instruction list: */
 	struct list_head node;
 
-	int use_count;      /* currently just updated/used by cp */
-
 #ifdef DEBUG
 	uint32_t serialno;
 #endif
@@ -443,6 +445,8 @@ struct ir3 {
 	/* List of ir3_array's: */
 	struct list_head array_list;
 
+	unsigned max_sun;   /* max Sethi–Ullman number */
+
 #ifdef DEBUG
 	unsigned block_count, instr_count;
 #endif
@@ -739,6 +743,14 @@ static inline bool is_meta(struct ir3_instruction *instr)
 	return (opc_cat(instr->opc) == -1);
 }
 
+static inline unsigned dest_regs(struct ir3_instruction *instr)
+{
+	if ((instr->regs_count == 0) || is_store(instr))
+		return 0;
+
+	return util_last_bit(instr->regs[0]->wrmask);
+}
+
 static inline bool writes_addr(struct ir3_instruction *instr)
 {
 	if (instr->regs_count > 0) {
@@ -999,6 +1011,9 @@ void ir3_cp(struct ir3 *ir, struct ir3_shader_variant *so);
 /* group neighbors and insert mov's to resolve conflicts: */
 void ir3_group(struct ir3 *ir);
 
+/* Sethi–Ullman numbering: */
+void ir3_sun(struct ir3 *ir);
+
 /* scheduling: */
 void ir3_sched_add_deps(struct ir3 *ir);
 int ir3_sched(struct ir3 *ir);
diff --git a/src/freedreno/ir3/ir3_compiler_nir.c b/src/freedreno/ir3/ir3_compiler_nir.c
index e6fe45daa12..1e3fbeb3117 100644
--- a/src/freedreno/ir3/ir3_compiler_nir.c
+++ b/src/freedreno/ir3/ir3_compiler_nir.c
@@ -2610,6 +2610,9 @@ ir3_compile_shader_nir(struct ir3_compiler *compiler,
 		ir3_print(ir);
 	}
 
+	/* do Sethi–Ullman numbering before scheduling: */
+	ir3_sun(ir);
+
 	ret = ir3_sched(ir);
 	if (ret) {
 		DBG("SCHED failed!");
@@ -2708,6 +2711,8 @@ ir3_compile_shader_nir(struct ir3_compiler *compiler,
 	else
 		so->total_in = max_bary + 1;
 
+	so->max_sun = ir->max_sun;
+
 out:
 	if (ret) {
 		if (so->ir)
diff --git a/src/freedreno/ir3/ir3_print.c b/src/freedreno/ir3/ir3_print.c
index 415024b985f..b69941da29e 100644
--- a/src/freedreno/ir3/ir3_print.c
+++ b/src/freedreno/ir3/ir3_print.c
@@ -40,7 +40,8 @@ static void print_instr_name(struct ir3_instruction *instr)
 #endif
 	printf("%04u:", instr->name);
 	printf("%04u:", instr->ip);
-	printf("%03u: ", instr->depth);
+	printf("%03u:", instr->depth);
+	printf("%03u: ", instr->sun);
 
 	if (instr->flags & IR3_INSTR_SY)
 		printf("(sy)");
diff --git a/src/freedreno/ir3/ir3_shader.c b/src/freedreno/ir3/ir3_shader.c
index 3dcc3945ffe..7ebc0b4c853 100644
--- a/src/freedreno/ir3/ir3_shader.c
+++ b/src/freedreno/ir3/ir3_shader.c
@@ -393,6 +393,8 @@ ir3_shader_disasm(struct ir3_shader_variant *so, uint32_t *bin, FILE *out)
 
 	fprintf(out, "; %u (ss), %u (sy)\n", so->info.ss, so->info.sy);
 
+	fprintf(out, "; max_sun=%u\n", ir->max_sun);
+
 	/* print shader type specific info: */
 	switch (so->type) {
 	case MESA_SHADER_VERTEX:
diff --git a/src/freedreno/ir3/ir3_shader.h b/src/freedreno/ir3/ir3_shader.h
index 5fffaf9fb85..f8aa8522a55 100644
--- a/src/freedreno/ir3/ir3_shader.h
+++ b/src/freedreno/ir3/ir3_shader.h
@@ -344,6 +344,8 @@ struct ir3_shader_variant {
 	 */
 	unsigned branchstack;
 
+	unsigned max_sun;
+
 	/* the instructions length is in units of instruction groups
 	 * (4 instructions for a3xx, 16 instructions for a4xx.. each
 	 * instruction is 2 dwords):
diff --git a/src/freedreno/ir3/ir3_sun.c b/src/freedreno/ir3/ir3_sun.c
new file mode 100644
index 00000000000..7fea9a073e5
--- /dev/null
+++ b/src/freedreno/ir3/ir3_sun.c
@@ -0,0 +1,111 @@
+/*
+ * Copyright (C) 2018 Rob Clark <[email protected]>
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * Authors:
+ *    Rob Clark <[email protected]>
+ */
+
+
+#include "util/u_math.h"
+
+#include "ir3.h"
+
+/*
+ * A simple pass to do Sethi–Ullman numbering, as described in "Generalizations
+ * of the Sethi-Ullman algorithm for register allocation"[1].  This is used by
+ * the scheduler pass.
+ *
+ * TODO this could probably be more clever about flow control, ie. if a src
+ * is computed in multiple paths into a block, I think we should only have to
+ * consider the worst-case.
+ *
+ * [1] https://pdfs.semanticscholar.org/ae53/6010b214612c2571f483354c264b0b39c545.pdf
+ */
+
+static unsigned
+number_instr(struct ir3_instruction *instr)
+{
+	if (ir3_instr_check_mark(instr))
+		return instr->sun;
+
+	struct ir3_instruction *src;
+	const unsigned n = __ssa_src_cnt(instr);
+	unsigned a[n];
+	unsigned b[n];
+	unsigned i = 0;
+
+	/* TODO I think including false-deps in the calculation is the right
+	 * thing to do:
+	 */
+	foreach_ssa_src_n(src, n, instr) {
+		if (__is_false_dep(instr, n))
+			continue;
+		if (src->block != instr->block) {
+			a[i] = 1;
+		} else {
+			a[i] = number_instr(src);
+		}
+		b[i] = dest_regs(src);
+		i++;
+	}
+
+	/*
+	 * Rπ = max(aπ(1), bπ(1) + max(aπ(2), bπ(2) + max(..., bπ(k−1) + max(aπ(k), bπ(k)))...):
+	 */
+	unsigned last_r = 0;
+
+	for (int k = i - 1; k >= 0; k--) {
+		unsigned r = MAX2(a[k], b[k] + last_r);
+
+		if (k > 0)
+			r += b[k-1];
+
+		last_r = r;
+	}
+
+	last_r = MAX2(last_r, dest_regs(instr));
+
+	instr->sun = last_r;
+
+	return instr->sun;
+}
+
+void
+ir3_sun(struct ir3 *ir)
+{
+	unsigned max = 0;
+
+	ir3_clear_mark(ir);
+
+	for (unsigned i = 0; i < ir->noutputs; i++)
+		if (ir->outputs[i])
+			max = MAX2(max, number_instr(ir->outputs[i]));
+
+	list_for_each_entry (struct ir3_block, block, &ir->block_list, node) {
+		for (unsigned i = 0; i < block->keeps_count; i++)
+			max = MAX2(max, number_instr(block->keeps[i]));
+		if (block->condition)
+			max = MAX2(max, number_instr(block->condition));
+	}
+
+	ir->max_sun = max;
+}
diff --git a/src/freedreno/ir3/meson.build b/src/freedreno/ir3/meson.build
index bf99476311d..5f87ca6185b 100644
--- a/src/freedreno/ir3/meson.build
+++ b/src/freedreno/ir3/meson.build
@@ -56,6 +56,7 @@ libfreedreno_ir3_files = files(
   'ir3_sched.c',
   'ir3_shader.c',
   'ir3_shader.h',
+  'ir3_sun.c',
 )
 
 libfreedreno_ir3 = static_library(