More i8 legalizations (#1253)

* Legalize stack_{load,store}.i8, fixes #433
* Legalize select.i8, cc: #466
* Legalize brz.i8 and brnz.i8, cc: #1117

cranelift/codegen/meta/src/shared/legalize.rs

@@ -139,6 +139,7 @@ pub(crate) fn define(insts: &InstructionGroup, imm: &Immediates) -> TransformGro
     expand.custom_legalize(trapnz, "expand_cond_trap");
     expand.custom_legalize(br_table, "expand_br_table");
     expand.custom_legalize(select, "expand_select");
+    widen.custom_legalize(select, "expand_select"); // small ints
 
     // Custom expansions for floating point constants.
     // These expansions require bit-casting or creating constant pool entries.
@@ -149,6 +150,10 @@ pub(crate) fn define(insts: &InstructionGroup, imm: &Immediates) -> TransformGro
     expand.custom_legalize(insts.by_name("stack_load"), "expand_stack_load");
     expand.custom_legalize(insts.by_name("stack_store"), "expand_stack_store");
 
+    // Custom expansions for small stack memory acccess.
+    widen.custom_legalize(insts.by_name("stack_load"), "expand_stack_load");
+    widen.custom_legalize(insts.by_name("stack_store"), "expand_stack_store");
+
     // List of variables to reuse in patterns.
     let x = var("x");
     let y = var("y");
@@ -612,6 +617,18 @@ pub(crate) fn define(insts: &InstructionGroup, imm: &Immediates) -> TransformGro
         }
     }
 
+    for &ty in &[I8, I16] {
+        widen.legalize(
+            def!(brz.ty(x, ebb, vararg)),
+            vec![def!(a = uextend.I32(x)), def!(brz(a, ebb, vararg))],
+        );
+
+        widen.legalize(
+            def!(brnz.ty(x, ebb, vararg)),
+            vec![def!(a = uextend.I32(x)), def!(brnz(a, ebb, vararg))],
+        );
+    }
+
     // Expand integer operations with carry for RISC architectures that don't have
     // the flags.
     let intcc_ult = Literal::enumerator_for(&imm.intcc, "ult");

cranelift/filetests/filetests/isa/x86/brz-i8-run.clif

@@ -0,0 +1,34 @@
+test run
+target x86_64
+
+function u0:0() -> b1 {
+ebb0:
+    v0 = iconst.i8 0
+    brz v0, ebb1
+    jump ebb2
+
+ebb1:
+    v1 = bconst.b1 true
+    return v1
+
+ebb2:
+    v2 = bconst.b1 false
+    return v2
+}
+; run
+
+function u0:1() -> b1 {
+ebb0:
+    v0 = iconst.i8 0
+    brnz v0, ebb1
+    jump ebb2
+
+ebb1:
+    v1 = bconst.b1 false
+    return v1
+
+ebb2:
+    v2 = bconst.b1 true
+    return v2
+}
+; run

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

@@ -0,0 +1,38 @@
+test compile
+target x86_64
+
+function u0:0() -> b1 {
+ebb0:
+    v0 = iconst.i8 0
+    ; check: v0 = iconst.i8 0
+    brz v0, ebb1
+    ; nextln: v3 = uextend.i32 v0
+    ; nextln: brz v3, ebb1
+    jump ebb2
+
+ebb1:
+    v1 = bconst.b1 true
+    return v1
+
+ebb2:
+    v2 = bconst.b1 false
+    return v2
+}
+
+function u0:1() -> b1 {
+ebb0:
+    v0 = iconst.i8 0
+    ; check: v0 = iconst.i8 0
+    brnz v0, ebb1
+    ; nextln: v3 = uextend.i32 v0
+    ; nextln: brnz v3, ebb1
+    jump ebb2
+
+ebb1:
+    v1 = bconst.b1 false
+    return v1
+
+ebb2:
+    v2 = bconst.b1 true
+    return v2
+}

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

@@ -0,0 +1,8 @@
+test compile
+target x86_64
+
+function u0:0(b1, i8, i8) -> i8 {
+ebb0(v0: b1, v1: i8, v2: i8):
+    v3 = select v0, v1, v2
+    return v3
+}

cranelift/filetests/filetests/isa/x86/stack-load-store8.clif

@@ -0,0 +1,19 @@
+test compile
+target x86_64
+
+function u0:0(i8) -> i8 {
+    ss0 = explicit_slot 1
+
+ebb0(v0: i8):
+    stack_store v0, ss0
+    ; check: v2 = stack_addr.i64 ss0
+    ; nextln: v3 = uextend.i32 v0
+    ; nextln: istore8 notrap aligned v3, v2
+
+    v1 = stack_load.i8 ss0
+    ; check: v4 = stack_addr.i64 ss0
+    ; nextln: v5 = uload8.i32 notrap aligned v4
+    ; nextln: v1 = ireduce.i8 v5
+
+    return v1
+}