cranelift: Port most of simple_preopt.rs over to the peepmatic DSL

This ports all of the identity, no-op, simplification, and canonicalization
related optimizations over from being hand-coded to the `peepmatic` DSL. This
does not handle the branch-to-branch optimizations or most of the
divide-by-constant optimizations.

cranelift/filetests/filetests/isa/x86/isub_imm-i8.clif

@@ -6,9 +6,9 @@ function u0:0(i8) -> i8 fast {
 block0(v0: i8):
     v1 = iconst.i8 0
     v2 = isub v1, v0
-    ; check: v3 = uextend.i32 v0
-    ; nextln: v5 = iconst.i32 0
-    ; nextln = isub v5, v3
-    ; nextln = ireduce.i8 v4
+    ; check:  v4 = uextend.i32 v0
+    ; nextln: v6 = iconst.i32 0
+    ; nextln: v5 = isub v6, v4
+    ; nextln: v2 = ireduce.i8 v5
     return v2
 }

cranelift/filetests/filetests/simple_preopt/div_by_const_indirect.clif

@@ -7,14 +7,13 @@ function %indir_udiv32(i32) -> i32 {
 block0(v0: i32):
     v1 = iconst.i32 7
     v2 = udiv v0, v1
-    ; check: iconst.i32 7
-    ; check: iconst.i32 0x2492_4925
-    ; check: umulhi v0, v3
-    ; check: isub v0, v4
-    ; check: ushr_imm v5, 1
-    ; check: iadd v6, v4
-    ; check: v8 = ushr_imm v7, 2
-    ; check: v2 -> v8
+    ; check:  v4 = iconst.i32 0x2492_4925
+    ; nextln: v5 = umulhi v0, v4
+    ; nextln: v6 = isub v0, v5
+    ; nextln: v7 = ushr_imm v6, 1
+    ; nextln: v8 = iadd v7, v5
+    ; nextln: v9 = ushr_imm v8, 2
+    ; nextln: v2 -> v9
     return v2
 }
 
@@ -22,13 +21,12 @@ function %indir_sdiv32(i32) -> i32 {
 block0(v0: i32):
     v1 = iconst.i32 -17
     v2 = sdiv v0, v1
-    ; check: iconst.i32 -17
-    ; check: iconst.i32 0xffff_ffff_8787_8787
-    ; check: smulhi v0, v3
-    ; check: sshr_imm v4, 3
-    ; check: ushr_imm v5, 31
-    ; check: v7 = iadd v5, v6
-    ; check: v2 -> v7
+    ; check:  v4 = iconst.i32 0xffff_ffff_8787_8787
+    ; nextln: v5 = smulhi v0, v4
+    ; nextln: v6 = sshr_imm v5, 3
+    ; nextln: v7 = ushr_imm v6, 31
+    ; nextln: v8 = iadd v6, v7
+    ; nextln: v2 -> v8
     return v2
 }
 
@@ -36,11 +34,10 @@ function %indir_udiv64(i64) -> i64 {
 block0(v0: i64):
     v1 = iconst.i64 1337
     v2 = udiv v0, v1
-    ; check: iconst.i64 1337
-    ; check: iconst.i64 0xc411_9d95_2866_a139
-    ; check: umulhi v0, v3
-    ; check: v5 = ushr_imm v4, 10
-    ; check: v2 -> v5
+    ; check:  v4 = iconst.i64 0xc411_9d95_2866_a139
+    ; nextln: v5 = umulhi v0, v4
+    ; nextln: v6 = ushr_imm v5, 10
+    ; nextln: v2 -> v6
     return v2
 }
 
@@ -48,12 +45,11 @@ function %indir_sdiv64(i64) -> i64 {
 block0(v0: i64):
     v1 = iconst.i64 -90210
     v2 = sdiv v0, v1
-    ; check: iconst.i64 0xffff_ffff_fffe_9f9e
-    ; check: iconst.i64 0xd181_4ee8_939c_b8bb
-    ; check: smulhi v0, v3
-    ; check: sshr_imm v4, 14
-    ; check: ushr_imm v5, 63
-    ; check: v7 = iadd v5, v6
-    ; check: v2 -> v7
+    ; check:  v4 = iconst.i64 0xd181_4ee8_939c_b8bb
+    ; nextln: v5 = smulhi v0, v4
+    ; nextln: v6 = sshr_imm v5, 14
+    ; nextln: v7 = ushr_imm v6, 63
+    ; nextln: v8 = iadd v6, v7
+    ; nextln: v2 -> v8
     return v2
 }

cranelift/filetests/filetests/simple_preopt/replace_branching_instructions_and_cfg_predecessors.clif

@@ -0,0 +1,22 @@
+test simple_preopt
+target x86_64
+
+function u0:2(i64 , i64) {
+    gv1 = load.i64 notrap aligned gv0
+    heap0 = static gv1
+    block0(v0: i64, v1: i64):
+        v16 = iconst.i32 6
+        v17 = heap_addr.i64 heap0, v16, 1
+        v18 = load.i32 v17
+        v19 = iconst.i32 4
+        v20 = icmp ne v18, v19
+        v21 = bint.i32 v20
+        brnz v21, block2
+        jump block4
+    block4:
+        jump block1
+    block2:
+        jump block1
+    block1:
+        return
+}

cranelift/filetests/filetests/simple_preopt/simplify32.clif

@@ -58,4 +58,3 @@ block0(v0: i64):
 ; nextln:     v2 = iadd v0, v1
 ; nextln:     return v2
 ; nextln: }
-

cranelift/filetests/filetests/simple_preopt/simplify64.clif

@@ -44,6 +44,37 @@ block0(v0: i32):
 ; nextln:     return v3
 ; nextln: }
 
+function %ifcmp_imm(i32) -> i32 {
+block0(v0: i32):
+    v1 = iconst.i32 2
+    v2 = ifcmp v0, v1
+    brif eq v2, block1
+    jump block2
+
+block1:
+    v3 = iconst.i32 1
+    return v3
+
+block2:
+    v4 = iconst.i32 2
+    return v4
+}
+; sameln: function %ifcmp_imm
+; nextln: block0(v0: i32):
+; nextln:     v1 = iconst.i32 2
+; nextln:     v2 = ifcmp_imm v0, 2
+; nextln:     brif eq v2, block1
+; nextln:     jump block2
+; nextln: 
+; nextln: block1:
+; nextln:     v3 = iconst.i32 1
+; nextln:     return v3
+; nextln: 
+; nextln: block2:
+; nextln:     v4 = iconst.i32 2
+; nextln:     return v4
+; nextln: }
+
 function %brz_bint(i32) {
 block0(v0: i32):
     v3 = icmp_imm slt v0, 0

cranelift/filetests/filetests/simple_preopt/simplify_instruction_into_alias_of_value.clif

@@ -0,0 +1,17 @@
+test simple_preopt
+target x86_64
+
+;; The `isub` is a no-op, but we can't replace the whole `isub` instruction with
+;; its `v2` operand's instruction because `v2` is one of many results. Instead,
+;; we need to make an alias `v3 -> v2`.
+
+function %replace_inst_with_alias() -> i32 {
+block0:
+    v0 = iconst.i32 0
+    v1, v2 = x86_smulx v0, v0
+    v3 = isub v2, v0
+    ; check:  v0 = iconst.i32 0
+    ; nextln: v1, v2 = x86_smulx v0, v0
+    ; nextln: v3 -> v2
+    return v3
+}