Optimize if of br_if (#2216)

An if whose body is a br_if can be turned into a br_if of a combined condition (if side effects allow it).

The naive size in bytes is identical between the patterns, but the select may avoid a hardware branch, and also the select may be further optimized. On the benchmark suite this helps every single benchmark, but by quite small amounts (e.g. 100 bytes on sqlite, which is 1MB).

This was noticed in emscripten-core/emscripten#8941

12 files changed, 195 insertions, 77 deletions

diff --git a/test/emcc_hello_world.fromasm b/test/emcc_hello_world.fromasm
index a9b17068e..25cbe7176 100644
--- a/test/emcc_hello_world.fromasm
+++ b/test/emcc_hello_world.fromasm
@@ -10327,13 +10327,18 @@
              )
             )
             (block
-             (if
-              (local.tee $3
-               (i32.load
-                (i32.const 616)
+             (br_if $label$break$L279
+              (select
+               (i32.eqz
+                (i32.eqz
+                 (local.tee $3
+                  (i32.load
+                   (i32.const 616)
+                  )
+                 )
+                )
                )
-              )
-              (br_if $label$break$L279
+               (i32.const 0)
                (i32.or
                 (i32.le_u
                  (local.get $12)
diff --git a/test/emcc_hello_world.fromasm.clamp b/test/emcc_hello_world.fromasm.clamp
index 212e5e8d3..8feba2827 100644
--- a/test/emcc_hello_world.fromasm.clamp
+++ b/test/emcc_hello_world.fromasm.clamp
@@ -10377,13 +10377,18 @@
              )
             )
             (block
-             (if
-              (local.tee $3
-               (i32.load
-                (i32.const 616)
+             (br_if $label$break$L279
+              (select
+               (i32.eqz
+                (i32.eqz
+                 (local.tee $3
+                  (i32.load
+                   (i32.const 616)
+                  )
+                 )
+                )
                )
-              )
-              (br_if $label$break$L279
+               (i32.const 0)
                (i32.or
                 (i32.le_u
                  (local.get $12)
diff --git a/test/emcc_hello_world.fromasm.imprecise b/test/emcc_hello_world.fromasm.imprecise
index f6fe8aca1..8bb8c257b 100644
--- a/test/emcc_hello_world.fromasm.imprecise
+++ b/test/emcc_hello_world.fromasm.imprecise
@@ -10223,13 +10223,18 @@
              )
             )
             (block
-             (if
-              (local.tee $3
-               (i32.load
-                (i32.const 616)
+             (br_if $label$break$L279
+              (select
+               (i32.eqz
+                (i32.eqz
+                 (local.tee $3
+                  (i32.load
+                   (i32.const 616)
+                  )
+                 )
+                )
                )
-              )
-              (br_if $label$break$L279
+               (i32.const 0)
                (i32.or
                 (i32.le_u
                  (local.get $12)
diff --git a/test/memorygrowth.fromasm b/test/memorygrowth.fromasm
index 13231d482..f1858e724 100644
--- a/test/memorygrowth.fromasm
+++ b/test/memorygrowth.fromasm
@@ -3152,13 +3152,14 @@
               )
              )
              (block
-              (if
-               (local.tee $12
-                (i32.load
-                 (i32.const 1648)
+              (br_if $do-once33
+               (select
+                (local.tee $12
+                 (i32.load
+                  (i32.const 1648)
+                 )
                 )
-               )
-               (br_if $do-once33
+                (i32.const 0)
                 (i32.or
                  (i32.le_u
                   (local.get $8)
diff --git a/test/memorygrowth.fromasm.clamp b/test/memorygrowth.fromasm.clamp
index 13231d482..f1858e724 100644
--- a/test/memorygrowth.fromasm.clamp
+++ b/test/memorygrowth.fromasm.clamp
@@ -3152,13 +3152,14 @@
               )
              )
              (block
-              (if
-               (local.tee $12
-                (i32.load
-                 (i32.const 1648)
+              (br_if $do-once33
+               (select
+                (local.tee $12
+                 (i32.load
+                  (i32.const 1648)
+                 )
                 )
-               )
-               (br_if $do-once33
+                (i32.const 0)
                 (i32.or
                  (i32.le_u
                   (local.get $8)
diff --git a/test/memorygrowth.fromasm.imprecise b/test/memorygrowth.fromasm.imprecise
index d6011b1c7..61c455725 100644
--- a/test/memorygrowth.fromasm.imprecise
+++ b/test/memorygrowth.fromasm.imprecise
@@ -3150,13 +3150,14 @@
               )
              )
              (block
-              (if
-               (local.tee $12
-                (i32.load
-                 (i32.const 1648)
+              (br_if $do-once33
+               (select
+                (local.tee $12
+                 (i32.load
+                  (i32.const 1648)
+                 )
                 )
-               )
-               (br_if $do-once33
+                (i32.const 0)
                 (i32.or
                  (i32.le_u
                   (local.get $8)
diff --git a/test/passes/bysyncify_optimize-level=1.txt b/test/passes/bysyncify_optimize-level=1.txt
index 7b1093964..b2bc87f85 100644
--- a/test/passes/bysyncify_optimize-level=1.txt
+++ b/test/passes/bysyncify_optimize-level=1.txt
@@ -1244,11 +1244,12 @@
        )
       )
      )
-     (if
-      (i32.eqz
-       (global.get $__bysyncify_state)
-      )
-      (br_if $l
+     (br_if $l
+      (select
+       (i32.eqz
+        (global.get $__bysyncify_state)
+       )
+       (i32.const 0)
        (local.get $0)
       )
      )
diff --git a/test/passes/remove-unused-brs.txt b/test/passes/remove-unused-brs.txt
index 944828d5c..790d943c3 100644
--- a/test/passes/remove-unused-brs.txt
+++ b/test/passes/remove-unused-brs.txt
@@ -2431,4 +2431,55 @@
    )
   )
  )
+ (func $if_br_if (; 115 ;) (type $1)
+  (local $0 i32)
+  (block $label$1
+   (br_if $label$1
+    (select
+     (local.tee $0
+      (i32.const 1024)
+     )
+     (i32.const 0)
+     (i32.eqz
+      (i32.const -4)
+     )
+    )
+   )
+   (br_if $label$1
+    (select
+     (i32.const 1025)
+     (i32.const 0)
+     (i32.eqz
+      (i32.const -5)
+     )
+    )
+   )
+   (br_if $label$1
+    (select
+     (local.tee $0
+      (i32.const 1025)
+     )
+     (i32.const 0)
+     (i32.eqz
+      (i32.const -6)
+     )
+    )
+   )
+   (if
+    (i32.const 1026)
+    (br_if $label$1
+     (local.tee $0
+      (i32.const -7)
+     )
+    )
+   )
+   (i32.store
+    (i32.const 1024)
+    (i32.add
+     (local.get $0)
+     (i32.const 1)
+    )
+   )
+  )
+ )
 )
diff --git a/test/passes/remove-unused-brs.wast b/test/passes/remove-unused-brs.wast
index 15a56b2d9..c20e352f0 100644
--- a/test/passes/remove-unused-brs.wast
+++ b/test/passes/remove-unused-brs.wast
@@ -2039,5 +2039,53 @@
     )
    )
   )
+ (func $if_br_if
+  (local $0 i32)
+  (block $label$1
+   (if
+    (local.tee $0 ;; note side effect; it's ok
+     (i32.const 1024)
+    )
+    (br_if $label$1
+     (i32.eqz
+      (i32.const -4)
+     )
+    )
+   )
+   (if
+    (i32.const 1025)
+    (br_if $label$1
+     (i32.eqz
+      (i32.const -5)
+     )
+    )
+   )
+   (if
+    (local.tee $0 ;; note side effect; it's ok
+     (i32.const 1025)
+    )
+    (br_if $label$1
+     (i32.eqz
+      (i32.const -6)
+     )
+    )
+   )
+   (if
+    (i32.const 1026)
+    (br_if $label$1
+     (local.tee $0 ;; but here it is *not* ok
+      (i32.const -7)
+     )
+    )
+   )
+   (i32.store
+    (i32.const 1024)
+    (i32.add
+     (local.get $0)
+     (i32.const 1)
+    )
+   )
+  )
+ )
 )
 
diff --git a/test/unit.fromasm b/test/unit.fromasm
index b6065c905..748be4b8e 100644
--- a/test/unit.fromasm
+++ b/test/unit.fromasm
@@ -705,19 +705,17 @@
   )
  )
  (func $loophi (; 45 ;) (; has Stack IR ;) (param $0 i32) (param $1 i32)
-  (local $2 i32)
   (loop $while-in
    (block $while-out
     (call $loophi
      (local.get $0)
      (i32.const 0)
     )
-    (if
-     (local.tee $2
+    (br_if $while-out
+     (select
+      (local.get $0)
+      (i32.const 0)
       (local.get $0)
-     )
-     (br_if $while-out
-      (local.get $2)
      )
     )
     (br_if $while-in
@@ -747,9 +745,10 @@
      (local.set $2
       (local.get $0)
      )
-     (if
-      (call $return_int)
-      (br_if $label$break$L7
+     (br_if $label$break$L7
+      (select
+       (call $return_int)
+       (i32.const 0)
        (local.get $2)
       )
      )
@@ -779,9 +778,10 @@
      (local.set $1
       (local.get $0)
      )
-     (if
-      (call $return_int)
-      (br_if $label$break$L7
+     (br_if $label$break$L7
+      (select
+       (call $return_int)
+       (i32.const 0)
        (local.get $1)
       )
      )
diff --git a/test/unit.fromasm.clamp b/test/unit.fromasm.clamp
index 1a905fee5..0b57f2077 100644
--- a/test/unit.fromasm.clamp
+++ b/test/unit.fromasm.clamp
@@ -753,19 +753,17 @@
   )
  )
  (func $loophi (; 46 ;) (; has Stack IR ;) (param $0 i32) (param $1 i32)
-  (local $2 i32)
   (loop $while-in
    (block $while-out
     (call $loophi
      (local.get $0)
      (i32.const 0)
     )
-    (if
-     (local.tee $2
+    (br_if $while-out
+     (select
+      (local.get $0)
+      (i32.const 0)
       (local.get $0)
-     )
-     (br_if $while-out
-      (local.get $2)
      )
     )
     (br_if $while-in
@@ -795,9 +793,10 @@
      (local.set $2
       (local.get $0)
      )
-     (if
-      (call $return_int)
-      (br_if $label$break$L7
+     (br_if $label$break$L7
+      (select
+       (call $return_int)
+       (i32.const 0)
        (local.get $2)
       )
      )
@@ -827,9 +826,10 @@
      (local.set $1
       (local.get $0)
      )
-     (if
-      (call $return_int)
-      (br_if $label$break$L7
+     (br_if $label$break$L7
+      (select
+       (call $return_int)
+       (i32.const 0)
        (local.get $1)
       )
      )
diff --git a/test/unit.fromasm.imprecise b/test/unit.fromasm.imprecise
index bc2546b6e..f9fc294c5 100644
--- a/test/unit.fromasm.imprecise
+++ b/test/unit.fromasm.imprecise
@@ -693,19 +693,17 @@
   )
  )
  (func $loophi (; 44 ;) (; has Stack IR ;) (param $0 i32) (param $1 i32)
-  (local $2 i32)
   (loop $while-in
    (block $while-out
     (call $loophi
      (local.get $0)
      (i32.const 0)
     )
-    (if
-     (local.tee $2
+    (br_if $while-out
+     (select
+      (local.get $0)
+      (i32.const 0)
       (local.get $0)
-     )
-     (br_if $while-out
-      (local.get $2)
      )
     )
     (br_if $while-in
@@ -735,9 +733,10 @@
      (local.set $2
       (local.get $0)
      )
-     (if
-      (call $return_int)
-      (br_if $label$break$L7
+     (br_if $label$break$L7
+      (select
+       (call $return_int)
+       (i32.const 0)
        (local.get $2)
       )
      )
@@ -767,9 +766,10 @@
      (local.set $1
       (local.get $0)
      )
-     (if
-      (call $return_int)
-      (br_if $label$break$L7
+     (br_if $label$break$L7
+      (select
+       (call $return_int)
+       (i32.const 0)
        (local.get $1)
       )
      )