Merge pull request #13251 from Vexu/c-abi

implement ARM C ABI, separate C ABI tests from standalone tests

13 files changed, 694 insertions, 201 deletions

diff --git a/build.zig b/build.zig
index 2dd047a1e3..a1ba2b20bc 100644
--- a/build.zig
+++ b/build.zig
@@ -508,6 +508,7 @@ pub fn build(b: *Builder) !void {
         b.enable_wasmtime,
         b.enable_wine,
     ));
+    test_step.dependOn(tests.addCAbiTests(b, skip_non_native));
     test_step.dependOn(tests.addLinkTests(b, test_filter, modes, enable_macos_sdk, skip_stage2_tests));
     test_step.dependOn(tests.addStackTraceTests(b, test_filter, modes));
     test_step.dependOn(tests.addCliTests(b, test_filter, modes));
diff --git a/src/arch/aarch64/abi.zig b/src/arch/aarch64/abi.zig
index 7c92d4e91c..e83cc0444a 100644
--- a/src/arch/aarch64/abi.zig
+++ b/src/arch/aarch64/abi.zig
@@ -5,41 +5,54 @@ const Register = bits.Register;
 const RegisterManagerFn = @import("../../register_manager.zig").RegisterManager;
 const Type = @import("../../type.zig").Type;
 
-pub const Class = enum(u8) { memory, integer, none, float_array, _ };
+pub const Class = union(enum) {
+    memory,
+    byval,
+    integer,
+    double_integer,
+    float_array: u8,
+};
 
 /// For `float_array` the second element will be the amount of floats.
-pub fn classifyType(ty: Type, target: std.Target) [2]Class {
-    var maybe_float_bits: ?u16 = null;
-    const float_count = countFloats(ty, target, &maybe_float_bits);
-    if (float_count <= sret_float_count) return .{ .float_array, @intToEnum(Class, float_count) };
-    return classifyTypeInner(ty, target);
-}
+pub fn classifyType(ty: Type, target: std.Target) Class {
+    std.debug.assert(ty.hasRuntimeBitsIgnoreComptime());
 
-fn classifyTypeInner(ty: Type, target: std.Target) [2]Class {
-    if (!ty.hasRuntimeBitsIgnoreComptime()) return .{ .none, .none };
+    var maybe_float_bits: ?u16 = null;
     switch (ty.zigTypeTag()) {
         .Struct => {
-            if (ty.containerLayout() == .Packed) return .{ .integer, .none };
+            if (ty.containerLayout() == .Packed) return .byval;
+            const float_count = countFloats(ty, target, &maybe_float_bits);
+            if (float_count <= sret_float_count) return .{ .float_array = float_count };
+
             const bit_size = ty.bitSize(target);
-            if (bit_size > 128) return .{ .memory, .none };
-            if (bit_size > 64) return .{ .integer, .integer };
-            return .{ .integer, .none };
+            if (bit_size > 128) return .memory;
+            if (bit_size > 64) return .double_integer;
+            return .integer;
         },
         .Union => {
+            if (ty.containerLayout() == .Packed) return .byval;
+            const float_count = countFloats(ty, target, &maybe_float_bits);
+            if (float_count <= sret_float_count) return .{ .float_array = float_count };
+
+            const bit_size = ty.bitSize(target);
+            if (bit_size > 128) return .memory;
+            if (bit_size > 64) return .double_integer;
+            return .integer;
+        },
+        .Int, .Enum, .ErrorSet, .Float, .Bool => return .byval,
+        .Vector => {
             const bit_size = ty.bitSize(target);
-            if (bit_size > 128) return .{ .memory, .none };
-            if (bit_size > 64) return .{ .integer, .integer };
-            return .{ .integer, .none };
+            // TODO is this controlled by a cpu feature?
+            if (bit_size > 128) return .memory;
+            return .byval;
         },
-        .Int, .Enum, .ErrorSet, .Vector, .Float, .Bool => return .{ .integer, .none },
-        .Array => return .{ .memory, .none },
         .Optional => {
             std.debug.assert(ty.isPtrLikeOptional());
-            return .{ .integer, .none };
+            return .byval;
         },
         .Pointer => {
             std.debug.assert(!ty.isSlice());
-            return .{ .integer, .none };
+            return .byval;
         },
         .ErrorUnion,
         .Frame,
@@ -55,17 +68,18 @@ fn classifyTypeInner(ty: Type, target: std.Target) [2]Class {
         .Fn,
         .Opaque,
         .EnumLiteral,
+        .Array,
         => unreachable,
     }
 }
 
 const sret_float_count = 4;
-fn countFloats(ty: Type, target: std.Target, maybe_float_bits: *?u16) u32 {
-    const invalid = std.math.maxInt(u32);
+fn countFloats(ty: Type, target: std.Target, maybe_float_bits: *?u16) u8 {
+    const invalid = std.math.maxInt(u8);
     switch (ty.zigTypeTag()) {
         .Union => {
             const fields = ty.unionFields();
-            var max_count: u32 = 0;
+            var max_count: u8 = 0;
             for (fields.values()) |field| {
                 const field_count = countFloats(field.ty, target, maybe_float_bits);
                 if (field_count == invalid) return invalid;
@@ -76,7 +90,7 @@ fn countFloats(ty: Type, target: std.Target, maybe_float_bits: *?u16) u32 {
         },
         .Struct => {
             const fields_len = ty.structFieldCount();
-            var count: u32 = 0;
+            var count: u8 = 0;
             var i: u32 = 0;
             while (i < fields_len) : (i += 1) {
                 const field_ty = ty.structFieldType(i);
diff --git a/src/arch/arm/abi.zig b/src/arch/arm/abi.zig
index 3fcfb0e561..9659ca13d7 100644
--- a/src/arch/arm/abi.zig
+++ b/src/arch/arm/abi.zig
@@ -2,6 +2,160 @@ const std = @import("std");
 const bits = @import("bits.zig");
 const Register = bits.Register;
 const RegisterManagerFn = @import("../../register_manager.zig").RegisterManager;
+const Type = @import("../../type.zig").Type;
+
+pub const Class = union(enum) {
+    memory,
+    byval,
+    i32_array: u8,
+    i64_array: u8,
+
+    fn arrSize(total_size: u64, arr_size: u64) Class {
+        const count = @intCast(u8, std.mem.alignForward(total_size, arr_size) / arr_size);
+        if (arr_size == 32) {
+            return .{ .i32_array = count };
+        } else {
+            return .{ .i64_array = count };
+        }
+    }
+};
+
+pub const Context = enum { ret, arg };
+
+pub fn classifyType(ty: Type, target: std.Target, ctx: Context) Class {
+    std.debug.assert(ty.hasRuntimeBitsIgnoreComptime());
+
+    var maybe_float_bits: ?u16 = null;
+    const max_byval_size = 512;
+    switch (ty.zigTypeTag()) {
+        .Struct => {
+            const bit_size = ty.bitSize(target);
+            if (ty.containerLayout() == .Packed) {
+                if (bit_size > 64) return .memory;
+                return .byval;
+            }
+            if (bit_size > max_byval_size) return .memory;
+            const float_count = countFloats(ty, target, &maybe_float_bits);
+            if (float_count <= byval_float_count) return .byval;
+
+            const fields = ty.structFieldCount();
+            var i: u32 = 0;
+            while (i < fields) : (i += 1) {
+                const field_ty = ty.structFieldType(i);
+                const field_alignment = ty.structFieldAlign(i, target);
+                const field_size = field_ty.bitSize(target);
+                if (field_size > 32 or field_alignment > 32) {
+                    return Class.arrSize(bit_size, 64);
+                }
+            }
+            return Class.arrSize(bit_size, 32);
+        },
+        .Union => {
+            const bit_size = ty.bitSize(target);
+            if (ty.containerLayout() == .Packed) {
+                if (bit_size > 64) return .memory;
+                return .byval;
+            }
+            if (bit_size > max_byval_size) return .memory;
+            const float_count = countFloats(ty, target, &maybe_float_bits);
+            if (float_count <= byval_float_count) return .byval;
+
+            for (ty.unionFields().values()) |field| {
+                if (field.ty.bitSize(target) > 32 or field.normalAlignment(target) > 32) {
+                    return Class.arrSize(bit_size, 64);
+                }
+            }
+            return Class.arrSize(bit_size, 32);
+        },
+        .Bool, .Float => return .byval,
+        .Int => {
+            // TODO this is incorrect for _BitInt(128) but implementing
+            // this correctly makes implementing compiler-rt impossible.
+            // const bit_size = ty.bitSize(target);
+            // if (bit_size > 64) return .memory;
+            return .byval;
+        },
+        .Enum, .ErrorSet => {
+            const bit_size = ty.bitSize(target);
+            if (bit_size > 64) return .memory;
+            return .byval;
+        },
+        .Vector => {
+            const bit_size = ty.bitSize(target);
+            // TODO is this controlled by a cpu feature?
+            if (ctx == .ret and bit_size > 128) return .memory;
+            if (bit_size > 512) return .memory;
+            return .byval;
+        },
+        .Optional => {
+            std.debug.assert(ty.isPtrLikeOptional());
+            return .byval;
+        },
+        .Pointer => {
+            std.debug.assert(!ty.isSlice());
+            return .byval;
+        },
+        .ErrorUnion,
+        .Frame,
+        .AnyFrame,
+        .NoReturn,
+        .Void,
+        .Type,
+        .ComptimeFloat,
+        .ComptimeInt,
+        .Undefined,
+        .Null,
+        .BoundFn,
+        .Fn,
+        .Opaque,
+        .EnumLiteral,
+        .Array,
+        => unreachable,
+    }
+}
+
+const byval_float_count = 4;
+fn countFloats(ty: Type, target: std.Target, maybe_float_bits: *?u16) u32 {
+    const invalid = std.math.maxInt(u32);
+    switch (ty.zigTypeTag()) {
+        .Union => {
+            const fields = ty.unionFields();
+            var max_count: u32 = 0;
+            for (fields.values()) |field| {
+                const field_count = countFloats(field.ty, target, maybe_float_bits);
+                if (field_count == invalid) return invalid;
+                if (field_count > max_count) max_count = field_count;
+                if (max_count > byval_float_count) return invalid;
+            }
+            return max_count;
+        },
+        .Struct => {
+            const fields_len = ty.structFieldCount();
+            var count: u32 = 0;
+            var i: u32 = 0;
+            while (i < fields_len) : (i += 1) {
+                const field_ty = ty.structFieldType(i);
+                const field_count = countFloats(field_ty, target, maybe_float_bits);
+                if (field_count == invalid) return invalid;
+                count += field_count;
+                if (count > byval_float_count) return invalid;
+            }
+            return count;
+        },
+        .Float => {
+            const float_bits = maybe_float_bits.* orelse {
+                const float_bits = ty.floatBits(target);
+                if (float_bits != 32 and float_bits != 64) return invalid;
+                maybe_float_bits.* = float_bits;
+                return 1;
+            };
+            if (ty.floatBits(target) == float_bits) return 1;
+            return invalid;
+        },
+        .Void => return 0,
+        else => return invalid,
+    }
+}
 
 pub const callee_preserved_regs = [_]Register{ .r4, .r5, .r6, .r7, .r8, .r10 };
 pub const caller_preserved_regs = [_]Register{ .r0, .r1, .r2, .r3 };
diff --git a/src/arch/riscv64/abi.zig b/src/arch/riscv64/abi.zig
index 3792c4ab18..8a560f4596 100644
--- a/src/arch/riscv64/abi.zig
+++ b/src/arch/riscv64/abi.zig
@@ -2,6 +2,75 @@ const std = @import("std");
 const bits = @import("bits.zig");
 const Register = bits.Register;
 const RegisterManagerFn = @import("../../register_manager.zig").RegisterManager;
+const Type = @import("../../type.zig").Type;
+
+pub const Class = enum { memory, byval, integer, double_integer };
+
+pub fn classifyType(ty: Type, target: std.Target) Class {
+    std.debug.assert(ty.hasRuntimeBitsIgnoreComptime());
+
+    const max_byval_size = target.cpu.arch.ptrBitWidth() * 2;
+    switch (ty.zigTypeTag()) {
+        .Struct => {
+            const bit_size = ty.bitSize(target);
+            if (ty.containerLayout() == .Packed) {
+                if (bit_size > max_byval_size) return .memory;
+                return .byval;
+            }
+            // TODO this doesn't exactly match what clang produces but its better than nothing
+            if (bit_size > max_byval_size) return .memory;
+            if (bit_size > max_byval_size / 2) return .double_integer;
+            return .integer;
+        },
+        .Union => {
+            const bit_size = ty.bitSize(target);
+            if (ty.containerLayout() == .Packed) {
+                if (bit_size > max_byval_size) return .memory;
+                return .byval;
+            }
+            // TODO this doesn't exactly match what clang produces but its better than nothing
+            if (bit_size > max_byval_size) return .memory;
+            if (bit_size > max_byval_size / 2) return .double_integer;
+            return .integer;
+        },
+        .Bool => return .integer,
+        .Float => return .byval,
+        .Int, .Enum, .ErrorSet => {
+            const bit_size = ty.bitSize(target);
+            if (bit_size > max_byval_size) return .memory;
+            return .byval;
+        },
+        .Vector => {
+            const bit_size = ty.bitSize(target);
+            if (bit_size > max_byval_size) return .memory;
+            return .integer;
+        },
+        .Optional => {
+            std.debug.assert(ty.isPtrLikeOptional());
+            return .byval;
+        },
+        .Pointer => {
+            std.debug.assert(!ty.isSlice());
+            return .byval;
+        },
+        .ErrorUnion,
+        .Frame,
+        .AnyFrame,
+        .NoReturn,
+        .Void,
+        .Type,
+        .ComptimeFloat,
+        .ComptimeInt,
+        .Undefined,
+        .Null,
+        .BoundFn,
+        .Fn,
+        .Opaque,
+        .EnumLiteral,
+        .Array,
+        => unreachable,
+    }
+}
 
 pub const callee_preserved_regs = [_]Register{
     .s0, .s1, .s2, .s3, .s4, .s5, .s6, .s7, .s8, .s9, .s10, .s11,
diff --git a/src/arch/x86_64/CodeGen.zig b/src/arch/x86_64/CodeGen.zig
index a1b354482b..5f793aaeb9 100644
--- a/src/arch/x86_64/CodeGen.zig
+++ b/src/arch/x86_64/CodeGen.zig
@@ -7149,7 +7149,7 @@ fn resolveCallingConventionValues(self: *Self, fn_ty: Type) !CallMCValues {
 
                 const classes: []const abi.Class = switch (self.target.os.tag) {
                     .windows => &[1]abi.Class{abi.classifyWindows(ty, self.target.*)},
-                    else => mem.sliceTo(&abi.classifySystemV(ty, self.target.*), .none),
+                    else => mem.sliceTo(&abi.classifySystemV(ty, self.target.*, .arg), .none),
                 };
                 if (classes.len > 1) {
                     return self.fail("TODO handle multiple classes per type", .{});
diff --git a/src/arch/x86_64/abi.zig b/src/arch/x86_64/abi.zig
index 45c5760540..a428bcacdd 100644
--- a/src/arch/x86_64/abi.zig
+++ b/src/arch/x86_64/abi.zig
@@ -60,9 +60,11 @@ pub fn classifyWindows(ty: Type, target: Target) Class {
     }
 }
 
+pub const Context = enum { ret, arg };
+
 /// There are a maximum of 8 possible return slots. Returned values are in
 /// the beginning of the array; unused slots are filled with .none.
-pub fn classifySystemV(ty: Type, target: Target) [8]Class {
+pub fn classifySystemV(ty: Type, target: Target, ctx: Context) [8]Class {
     const memory_class = [_]Class{
         .memory, .none, .none, .none,
         .none,   .none, .none, .none,
@@ -134,6 +136,22 @@ pub fn classifySystemV(ty: Type, target: Target) [8]Class {
         },
         .Vector => {
             const elem_ty = ty.childType();
+            if (ctx == .arg) {
+                const bit_size = ty.bitSize(target);
+                if (bit_size > 128) return memory_class;
+                if (bit_size > 80) return .{
+                    .integer, .integer, .none, .none,
+                    .none,    .none,    .none, .none,
+                };
+                if (bit_size > 64) return .{
+                    .x87,  .none, .none, .none,
+                    .none, .none, .none, .none,
+                };
+                return .{
+                    .integer, .none, .none, .none,
+                    .none,    .none, .none, .none,
+                };
+            }
             const bits = elem_ty.bitSize(target) * ty.arrayLen();
             if (bits <= 64) return .{
                 .sse,  .none, .none, .none,
@@ -201,7 +219,7 @@ pub fn classifySystemV(ty: Type, target: Target) [8]Class {
                     }
                 }
                 const field_size = field.ty.abiSize(target);
-                const field_class_array = classifySystemV(field.ty, target);
+                const field_class_array = classifySystemV(field.ty, target, .arg);
                 const field_class = std.mem.sliceTo(&field_class_array, .none);
                 if (byte_i + field_size <= 8) {
                     // Combine this field with the previous one.
@@ -315,7 +333,7 @@ pub fn classifySystemV(ty: Type, target: Target) [8]Class {
                     }
                 }
                 // Combine this field with the previous one.
-                const field_class = classifySystemV(field.ty, target);
+                const field_class = classifySystemV(field.ty, target, .arg);
                 for (result) |*result_item, i| {
                     const field_item = field_class[i];
                     // "If both classes are equal, this is the resulting class."
diff --git a/src/codegen/llvm.zig b/src/codegen/llvm.zig
index 3ebca13c20..ffc19cb6f6 100644
--- a/src/codegen/llvm.zig
+++ b/src/codegen/llvm.zig
@@ -24,6 +24,8 @@ const CType = @import("../type.zig").CType;
 const x86_64_abi = @import("../arch/x86_64/abi.zig");
 const wasm_c_abi = @import("../arch/wasm/abi.zig");
 const aarch64_c_abi = @import("../arch/aarch64/abi.zig");
+const arm_c_abi = @import("../arch/arm/abi.zig");
+const riscv_c_abi = @import("../arch/riscv64/abi.zig");
 
 const Error = error{ OutOfMemory, CodegenFail };
 
@@ -1138,6 +1140,25 @@ pub const Object = struct {
                         try args.append(load_inst);
                     }
                 },
+                .i32_array, .i64_array => {
+                    const param_ty = fn_info.param_types[it.zig_index - 1];
+                    const param_llvm_ty = try dg.lowerType(param_ty);
+                    const param = llvm_func.getParam(llvm_arg_i);
+                    llvm_arg_i += 1;
+
+                    const alignment = param_ty.abiAlignment(target);
+                    const arg_ptr = buildAllocaInner(builder, llvm_func, false, param_llvm_ty, alignment, target);
+                    const casted_ptr = builder.buildBitCast(arg_ptr, param.typeOf().pointerType(0), "");
+                    _ = builder.buildStore(param, casted_ptr);
+
+                    if (isByRef(param_ty)) {
+                        try args.append(arg_ptr);
+                    } else {
+                        const load_inst = builder.buildLoad(param_llvm_ty, arg_ptr, "");
+                        load_inst.setAlignment(alignment);
+                        try args.append(load_inst);
+                    }
+                },
             };
         }
 
@@ -2578,6 +2599,8 @@ pub const DeclGen = struct {
                 .multiple_llvm_float,
                 .as_u16,
                 .float_array,
+                .i32_array,
+                .i64_array,
                 => continue,
 
                 .slice => unreachable, // extern functions do not support slice types.
@@ -3132,6 +3155,11 @@ pub const DeclGen = struct {
                 const arr_ty = float_ty.arrayType(field_count);
                 try llvm_params.append(arr_ty);
             },
+            .i32_array, .i64_array => |arr_len| {
+                const elem_size: u8 = if (lowering == .i32_array) 32 else 64;
+                const arr_ty = dg.context.intType(elem_size).arrayType(arr_len);
+                try llvm_params.append(arr_ty);
+            },
         };
 
         return llvm.functionType(
@@ -4822,6 +4850,25 @@ pub const FuncGen = struct {
                 load_inst.setAlignment(alignment);
                 try llvm_args.append(load_inst);
             },
+            .i32_array, .i64_array => |arr_len| {
+                const elem_size: u8 = if (lowering == .i32_array) 32 else 64;
+                const arg = args[it.zig_index - 1];
+                const arg_ty = self.air.typeOf(arg);
+                var llvm_arg = try self.resolveInst(arg);
+                if (!isByRef(arg_ty)) {
+                    const p = self.buildAlloca(llvm_arg.typeOf(), null);
+                    const store_inst = self.builder.buildStore(llvm_arg, p);
+                    store_inst.setAlignment(arg_ty.abiAlignment(target));
+                    llvm_arg = store_inst;
+                }
+
+                const array_llvm_ty = self.dg.context.intType(elem_size).arrayType(arr_len);
+                const casted = self.builder.buildBitCast(llvm_arg, array_llvm_ty.pointerType(0), "");
+                const alignment = arg_ty.abiAlignment(target);
+                const load_inst = self.builder.buildLoad(array_llvm_ty, casted, "");
+                load_inst.setAlignment(alignment);
+                try llvm_args.append(load_inst);
+            },
         };
 
         const call = self.builder.buildCall(
@@ -10083,10 +10130,16 @@ fn firstParamSRet(fn_info: Type.Payload.Function.Data, target: std.Target) bool
             .mips, .mipsel => return false,
             .x86_64 => switch (target.os.tag) {
                 .windows => return x86_64_abi.classifyWindows(fn_info.return_type, target) == .memory,
-                else => return x86_64_abi.classifySystemV(fn_info.return_type, target)[0] == .memory,
+                else => return x86_64_abi.classifySystemV(fn_info.return_type, target, .ret)[0] == .memory,
             },
             .wasm32 => return wasm_c_abi.classifyType(fn_info.return_type, target)[0] == .indirect,
-            .aarch64, .aarch64_be => return aarch64_c_abi.classifyType(fn_info.return_type, target)[0] == .memory,
+            .aarch64, .aarch64_be => return aarch64_c_abi.classifyType(fn_info.return_type, target) == .memory,
+            .arm, .armeb => switch (arm_c_abi.classifyType(fn_info.return_type, target, .ret)) {
+                .memory, .i64_array => return true,
+                .i32_array => |size| return size != 1,
+                .byval => return false,
+            },
+            .riscv32, .riscv64 => return riscv_c_abi.classifyType(fn_info.return_type, target) == .memory,
             else => return false, // TODO investigate C ABI for other architectures
         },
         else => return false,
@@ -10139,7 +10192,7 @@ fn lowerFnRetTy(dg: *DeclGen, fn_info: Type.Payload.Function.Data) !*llvm.Type {
                         if (is_scalar) {
                             return dg.lowerType(fn_info.return_type);
                         }
-                        const classes = x86_64_abi.classifySystemV(fn_info.return_type, target);
+                        const classes = x86_64_abi.classifySystemV(fn_info.return_type, target, .ret);
                         if (classes[0] == .memory) {
                             return dg.context.voidType();
                         }
@@ -10197,22 +10250,44 @@ fn lowerFnRetTy(dg: *DeclGen, fn_info: Type.Payload.Function.Data) !*llvm.Type {
                     return dg.context.intType(@intCast(c_uint, abi_size * 8));
                 },
                 .aarch64, .aarch64_be => {
-                    if (is_scalar) {
-                        return dg.lowerType(fn_info.return_type);
-                    }
-                    const classes = aarch64_c_abi.classifyType(fn_info.return_type, target);
-                    if (classes[0] == .memory or classes[0] == .none) {
-                        return dg.context.voidType();
+                    switch (aarch64_c_abi.classifyType(fn_info.return_type, target)) {
+                        .memory => return dg.context.voidType(),
+                        .float_array => return dg.lowerType(fn_info.return_type),
+                        .byval => return dg.lowerType(fn_info.return_type),
+                        .integer => {
+                            const bit_size = fn_info.return_type.bitSize(target);
+                            return dg.context.intType(@intCast(c_uint, bit_size));
+                        },
+                        .double_integer => return dg.context.intType(64).arrayType(2),
                     }
-                    if (classes[0] == .float_array) {
-                        return dg.lowerType(fn_info.return_type);
+                },
+                .arm, .armeb => {
+                    switch (arm_c_abi.classifyType(fn_info.return_type, target, .ret)) {
+                        .memory, .i64_array => return dg.context.voidType(),
+                        .i32_array => |len| if (len == 1) {
+                            return dg.context.intType(32);
+                        } else {
+                            return dg.context.voidType();
+                        },
+                        .byval => return dg.lowerType(fn_info.return_type),
                     }
-                    if (classes[1] == .none) {
-                        const bit_size = fn_info.return_type.bitSize(target);
-                        return dg.context.intType(@intCast(c_uint, bit_size));
+                },
+                .riscv32, .riscv64 => {
+                    switch (riscv_c_abi.classifyType(fn_info.return_type, target)) {
+                        .memory => return dg.context.voidType(),
+                        .integer => {
+                            const bit_size = fn_info.return_type.bitSize(target);
+                            return dg.context.intType(@intCast(c_uint, bit_size));
+                        },
+                        .double_integer => {
+                            var llvm_types_buffer: [2]*llvm.Type = .{
+                                dg.context.intType(64),
+                                dg.context.intType(64),
+                            };
+                            return dg.context.structType(&llvm_types_buffer, 2, .False);
+                        },
+                        .byval => return dg.lowerType(fn_info.return_type),
                     }
-
-                    return dg.context.intType(64).arrayType(2);
                 },
                 // TODO investigate C ABI for other architectures
                 else => return dg.lowerType(fn_info.return_type),
@@ -10242,6 +10317,8 @@ const ParamTypeIterator = struct {
         slice,
         as_u16,
         float_array: u8,
+        i32_array: u8,
+        i64_array: u8,
     };
 
     pub fn next(it: *ParamTypeIterator) ?Lowering {
@@ -10288,15 +10365,6 @@ const ParamTypeIterator = struct {
             .C => {
                 const is_scalar = isScalar(ty);
                 switch (it.target.cpu.arch) {
-                    .riscv32, .riscv64 => {
-                        it.zig_index += 1;
-                        it.llvm_index += 1;
-                        if (ty.tag() == .f16) {
-                            return .as_u16;
-                        } else {
-                            return .byval;
-                        }
-                    },
                     .mips, .mipsel => {
                         it.zig_index += 1;
                         it.llvm_index += 1;
@@ -10334,18 +10402,18 @@ const ParamTypeIterator = struct {
                             else => unreachable,
                         },
                         else => {
-                            if (is_scalar) {
-                                it.zig_index += 1;
-                                it.llvm_index += 1;
-                                return .byval;
-                            }
-                            const classes = x86_64_abi.classifySystemV(ty, it.target);
+                            const classes = x86_64_abi.classifySystemV(ty, it.target, .arg);
                             if (classes[0] == .memory) {
                                 it.zig_index += 1;
                                 it.llvm_index += 1;
                                 it.byval_attr = true;
                                 return .byref;
                             }
+                            if (is_scalar) {
+                                it.zig_index += 1;
+                                it.llvm_index += 1;
+                                return .byval;
+                            }
                             var llvm_types_buffer: [8]u16 = undefined;
                             var llvm_types_index: u32 = 0;
                             for (classes) |class| {
@@ -10383,11 +10451,6 @@ const ParamTypeIterator = struct {
                                 it.llvm_index += 1;
                                 return .abi_sized_int;
                             }
-                            if (classes[0] == .sse and classes[1] == .none) {
-                                it.zig_index += 1;
-                                it.llvm_index += 1;
-                                return .byval;
-                            }
                             it.llvm_types_buffer = llvm_types_buffer;
                             it.llvm_types_len = llvm_types_index;
                             it.llvm_index += llvm_types_index;
@@ -10410,24 +10473,45 @@ const ParamTypeIterator = struct {
                     .aarch64, .aarch64_be => {
                         it.zig_index += 1;
                         it.llvm_index += 1;
-                        if (is_scalar) {
-                            return .byval;
+                        switch (aarch64_c_abi.classifyType(ty, it.target)) {
+                            .memory => return .byref,
+                            .float_array => |len| return Lowering{ .float_array = len },
+                            .byval => return .byval,
+                            .integer => {
+                                it.llvm_types_len = 1;
+                                it.llvm_types_buffer[0] = 64;
+                                return .multiple_llvm_ints;
+                            },
+                            .double_integer => return Lowering{ .i64_array = 2 },
                         }
-                        const classes = aarch64_c_abi.classifyType(ty, it.target);
-                        if (classes[0] == .memory) {
-                            return .byref;
+                    },
+                    .arm, .armeb => {
+                        it.zig_index += 1;
+                        it.llvm_index += 1;
+                        switch (arm_c_abi.classifyType(ty, it.target, .arg)) {
+                            .memory => {
+                                it.byval_attr = true;
+                                return .byref;
+                            },
+                            .byval => return .byval,
+                            .i32_array => |size| return Lowering{ .i32_array = size },
+                            .i64_array => |size| return Lowering{ .i64_array = size },
                         }
-                        if (classes[0] == .float_array) {
-                            return Lowering{ .float_array = @enumToInt(classes[1]) };
+                    },
+                    .riscv32, .riscv64 => {
+                        it.zig_index += 1;
+                        it.llvm_index += 1;
+                        if (ty.tag() == .f16) {
+                            return .as_u16;
                         }
-                        if (classes[1] == .none) {
-                            it.llvm_types_len = 1;
-                        } else {
-                            it.llvm_types_len = 2;
+                        switch (riscv_c_abi.classifyType(ty, it.target)) {
+                            .memory => {
+                                return .byref;
+                            },
+                            .byval => return .byval,
+                            .integer => return .abi_sized_int,
+                            .double_integer => return Lowering{ .i64_array = 2 },
                         }
-                        it.llvm_types_buffer[0] = 64;
-                        it.llvm_types_buffer[1] = 64;
-                        return .multiple_llvm_ints;
                     },
                     // TODO investigate C ABI for other architectures
                     else => {
@@ -10475,8 +10559,16 @@ fn ccAbiPromoteInt(
     };
     if (int_info.bits <= 16) return int_info.signedness;
     switch (target.cpu.arch) {
+        .riscv64 => {
+            if (int_info.bits == 32) {
+                // LLVM always signextends 32 bit ints, unsure if bug.
+                return .signed;
+            }
+            if (int_info.bits < 64) {
+                return int_info.signedness;
+            }
+        },
         .sparc64,
-        .riscv64,
         .powerpc64,
         .powerpc64le,
         => {
diff --git a/test/c_abi/build.zig b/test/c_abi/build.zig
deleted file mode 100644
index b9151f6dda..0000000000
--- a/test/c_abi/build.zig
+++ /dev/null
@@ -1,22 +0,0 @@
-const Builder = @import("std").build.Builder;
-
-pub fn build(b: *Builder) void {
-    const rel_opts = b.standardReleaseOptions();
-    const target = b.standardTargetOptions(.{});
-
-    const c_obj = b.addObject("cfuncs", null);
-    c_obj.addCSourceFile("cfuncs.c", &[_][]const u8{"-std=c99"});
-    c_obj.setBuildMode(rel_opts);
-    c_obj.linkSystemLibrary("c");
-    c_obj.target = target;
-
-    const main = b.addTest("main.zig");
-    main.setBuildMode(rel_opts);
-    main.addObject(c_obj);
-    main.target = target;
-
-    const test_step = b.step("test", "Test the program");
-    test_step.dependOn(&main.step);
-
-    b.default_step.dependOn(test_step);
-}
diff --git a/test/c_abi/build_wasm.zig b/test/c_abi/build_wasm.zig
deleted file mode 100644
index b288650855..0000000000
--- a/test/c_abi/build_wasm.zig
+++ /dev/null
@@ -1,24 +0,0 @@
-const std = @import("std");
-const Builder = std.build.Builder;
-
-pub fn build(b: *Builder) void {
-    const rel_opts = b.standardReleaseOptions();
-    const target: std.zig.CrossTarget = .{ .cpu_arch = .wasm32, .os_tag = .wasi };
-    b.use_stage1 = false;
-
-    const c_obj = b.addObject("cfuncs", null);
-    c_obj.addCSourceFile("cfuncs.c", &[_][]const u8{"-std=c99"});
-    c_obj.setBuildMode(rel_opts);
-    c_obj.linkSystemLibrary("c");
-    c_obj.setTarget(target);
-
-    const main = b.addTest("main.zig");
-    main.setBuildMode(rel_opts);
-    main.addObject(c_obj);
-    main.setTarget(target);
-
-    const test_step = b.step("test", "Test the program");
-    test_step.dependOn(&main.step);
-
-    b.default_step.dependOn(test_step);
-}
diff --git a/test/c_abi/cfuncs.c b/test/c_abi/cfuncs.c
index 3d169badc2..5891ee29d7 100644
--- a/test/c_abi/cfuncs.c
+++ b/test/c_abi/cfuncs.c
@@ -1,8 +1,8 @@
+#include <complex.h>
 #include <inttypes.h>
-#include <stdlib.h>
 #include <stdbool.h>
+#include <stdlib.h>
 #include <string.h>
-#include <complex.h>
 
 void zig_panic();
 
@@ -12,6 +12,14 @@ static void assert_or_panic(bool ok) {
     }
 }
 
+#if defined __powerpc__ && !defined _ARCH_PPC64
+#  define ZIG_PPC32
+#endif
+
+#if defined __riscv && defined _ILP32
+#  define ZIG_RISCV32
+#endif
+
 #ifdef __i386__
 #  define ZIG_NO_I128
 #endif
@@ -24,6 +32,14 @@ static void assert_or_panic(bool ok) {
 #  define ZIG_NO_I128
 #endif
 
+#ifdef ZIG_PPC32
+#  define ZIG_NO_I128
+#endif
+
+#ifdef ZIG_RISCV32
+#  define ZIG_NO_I128
+#endif
+
 #ifdef __i386__
 #  define ZIG_NO_COMPLEX
 #endif
@@ -32,6 +48,18 @@ static void assert_or_panic(bool ok) {
 #  define ZIG_NO_COMPLEX
 #endif
 
+#ifdef __arm__
+#  define ZIG_NO_COMPLEX
+#endif
+
+#ifdef __powerpc__
+#  define ZIG_NO_COMPLEX
+#endif
+
+#ifdef __riscv
+#  define ZIG_NO_COMPLEX
+#endif
+
 #ifndef ZIG_NO_I128
 struct i128 {
     __int128 value;
@@ -206,7 +234,7 @@ void run_c_tests(void) {
     zig_longdouble(12.34l);
     zig_five_floats(1.0f, 2.0f, 3.0f, 4.0f, 5.0f);
 
-    zig_ptr((void*)0xdeadbeefL);
+    zig_ptr((void *)0xdeadbeefL);
 
     zig_bool(true);
 
@@ -249,14 +277,15 @@ void run_c_tests(void) {
     }
 #endif
 
-#if !defined __mips__ && !defined __riscv
+#if !defined __mips__ && !defined ZIG_PPC32
     {
         struct BigStruct s = {1, 2, 3, 4, 5};
         zig_big_struct(s);
     }
 #endif
 
-#if !defined __i386__ && !defined __arm__ && !defined __mips__ && !defined __riscv
+#if !defined __i386__ && !defined __arm__ && !defined __mips__ && \
+    !defined ZIG_PPC32 && !defined _ARCH_PPC64
     {
         struct SmallStructInts s = {1, 2, 3, 4};
         zig_small_struct_ints(s);
@@ -281,28 +310,30 @@ void run_c_tests(void) {
         zig_small_packed_struct(s);
     }
 
-#if !defined __i386__ && !defined __arm__ && !defined __mips__ && !defined __riscv
+#if !defined __i386__ && !defined __arm__ && !defined __mips__ && \
+    !defined ZIG_PPC32 && !defined _ARCH_PPC64
     {
         struct SplitStructInts s = {1234, 100, 1337};
         zig_split_struct_ints(s);
     }
 #endif
 
-#if !defined __arm__ && !defined __riscv
+#if !defined __arm__ && !defined ZIG_PPC32 && !defined _ARCH_PPC64
     {
         struct MedStructMixed s = {1234, 100.0f, 1337.0f};
         zig_med_struct_mixed(s);
     }
 #endif
 
-#if !defined __i386__ && !defined __arm__ && !defined __mips__ && !defined __riscv
+#if !defined __i386__ && !defined __arm__ && !defined __mips__ && \
+    !defined ZIG_PPC32 && !defined _ARCH_PPC64
     {
         struct SplitStructMixed s = {1234, 100, 1337.0f};
         zig_split_struct_mixed(s);
     }
 #endif
 
-#if !defined __mips__ && !defined __riscv
+#if !defined __mips__ && !defined ZIG_PPC32
     {
         struct BigStruct s = {30, 31, 32, 33, 34};
         struct BigStruct res = zig_big_struct_both(s);
@@ -314,7 +345,7 @@ void run_c_tests(void) {
     }
 #endif
 
-#ifndef __riscv
+#if !defined ZIG_PPC32 && !defined _ARCH_PPC64
     {
         struct Rect r1 = {1, 21, 16, 4};
         struct Rect r2 = {178, 189, 21, 15};
@@ -322,7 +353,7 @@ void run_c_tests(void) {
     }
 #endif
 
-#if !defined __mips__ && !defined __riscv
+#if !defined __mips__ && !defined ZIG_PPC32
     {
         struct FloatRect r1 = {1, 21, 16, 4};
         struct FloatRect r2 = {178, 189, 21, 15};
@@ -335,9 +366,7 @@ void run_c_tests(void) {
 
         assert_or_panic(zig_ret_u8() == 0xff);
         assert_or_panic(zig_ret_u16() == 0xffff);
-#ifndef __riscv
         assert_or_panic(zig_ret_u32() == 0xffffffff);
-#endif
         assert_or_panic(zig_ret_u64() == 0xffffffffffffffff);
 
         assert_or_panic(zig_ret_i8() == -1);
@@ -404,7 +433,7 @@ void c_long_double(long double x) {
 }
 
 void c_ptr(void *x) {
-    assert_or_panic(x == (void*)0xdeadbeefL);
+    assert_or_panic(x == (void *)0xdeadbeefL);
 }
 
 void c_bool(bool x) {
@@ -672,7 +701,7 @@ void c_struct_with_array(StructWithArray x) {
 }
 
 StructWithArray c_ret_struct_with_array() {
-    return (StructWithArray) { 4, {}, 155 };
+    return (StructWithArray){4, {}, 155};
 }
 
 typedef struct {
@@ -701,3 +730,43 @@ FloatArrayStruct c_ret_float_array_struct() {
     x.size.height = 4;
     return x;
 }
+
+typedef uint32_t SmallVec __attribute__((vector_size(2 * sizeof(uint32_t))));
+
+void c_small_vec(SmallVec vec) {
+    assert_or_panic(vec[0] == 1);
+    assert_or_panic(vec[1] == 2);
+}
+
+SmallVec c_ret_small_vec(void) {
+    return (SmallVec){3, 4};
+}
+
+typedef size_t BigVec __attribute__((vector_size(8 * sizeof(size_t))));
+
+void c_big_vec(BigVec vec) {
+    assert_or_panic(vec[0] == 1);
+    assert_or_panic(vec[1] == 2);
+    assert_or_panic(vec[2] == 3);
+    assert_or_panic(vec[3] == 4);
+    assert_or_panic(vec[4] == 5);
+    assert_or_panic(vec[5] == 6);
+    assert_or_panic(vec[6] == 7);
+    assert_or_panic(vec[7] == 8);
+}
+
+BigVec c_ret_big_vec(void) {
+    return (BigVec){9, 10, 11, 12, 13, 14, 15, 16};
+}
+
+typedef struct {
+    float x, y;
+} Vector2;
+
+void c_ptr_size_float_struct(Vector2 vec) {
+    assert_or_panic(vec.x == 1);
+    assert_or_panic(vec.y == 2);
+}
+Vector2 c_ret_ptr_size_float_struct(void) {
+    return (Vector2){3, 4};
+}
diff --git a/test/c_abi/main.zig b/test/c_abi/main.zig
index b239fd6ce3..d9fb25f4b3 100644
--- a/test/c_abi/main.zig
+++ b/test/c_abi/main.zig
@@ -3,7 +3,7 @@ const builtin = @import("builtin");
 const print = std.debug.print;
 const expect = std.testing.expect;
 const has_i128 = builtin.cpu.arch != .i386 and !builtin.cpu.arch.isARM() and
-    !builtin.cpu.arch.isMIPS();
+    !builtin.cpu.arch.isMIPS() and !builtin.cpu.arch.isPPC();
 
 extern fn run_c_tests() void;
 
@@ -112,6 +112,9 @@ test "C ABI floats" {
 }
 
 test "C ABI long double" {
+    if (comptime builtin.cpu.arch.isPPC()) return error.SkipZigTest;
+    if (comptime builtin.cpu.arch.isPPC64()) return error.SkipZigTest;
+
     c_long_double(12.34);
 }
 
@@ -167,7 +170,8 @@ extern fn c_cmultd_comp(a_r: f64, a_i: f64, b_r: f64, b_i: f64) ComplexDouble;
 extern fn c_cmultf(a: ComplexFloat, b: ComplexFloat) ComplexFloat;
 extern fn c_cmultd(a: ComplexDouble, b: ComplexDouble) ComplexDouble;
 
-const complex_abi_compatible = builtin.cpu.arch != .i386 and !builtin.cpu.arch.isMIPS();
+const complex_abi_compatible = builtin.cpu.arch != .i386 and !builtin.cpu.arch.isMIPS() and
+    !builtin.cpu.arch.isARM() and !builtin.cpu.arch.isPPC() and !builtin.cpu.arch.isRISCV();
 
 test "C ABI complex float" {
     if (!complex_abi_compatible) return error.SkipZigTest;
@@ -177,8 +181,8 @@ test "C ABI complex float" {
     const b = ComplexFloat{ .real = 11.3, .imag = -1.5 };
 
     const z = c_cmultf(a, b);
-    expect(z.real == 1.5) catch @panic("test failure: zig_complex_float 1");
-    expect(z.imag == 13.5) catch @panic("test failure: zig_complex_float 2");
+    try expect(z.real == 1.5);
+    try expect(z.imag == 13.5);
 }
 
 test "C ABI complex float by component" {
@@ -188,8 +192,8 @@ test "C ABI complex float by component" {
     const b = ComplexFloat{ .real = 11.3, .imag = -1.5 };
 
     const z2 = c_cmultf_comp(a.real, a.imag, b.real, b.imag);
-    expect(z2.real == 1.5) catch @panic("test failure: zig_complex_float 3");
-    expect(z2.imag == 13.5) catch @panic("test failure: zig_complex_float 4");
+    try expect(z2.real == 1.5);
+    try expect(z2.imag == 13.5);
 }
 
 test "C ABI complex double" {
@@ -199,8 +203,8 @@ test "C ABI complex double" {
     const b = ComplexDouble{ .real = 11.3, .imag = -1.5 };
 
     const z = c_cmultd(a, b);
-    expect(z.real == 1.5) catch @panic("test failure: zig_complex_double 1");
-    expect(z.imag == 13.5) catch @panic("test failure: zig_complex_double 2");
+    try expect(z.real == 1.5);
+    try expect(z.imag == 13.5);
 }
 
 test "C ABI complex double by component" {
@@ -210,8 +214,8 @@ test "C ABI complex double by component" {
     const b = ComplexDouble{ .real = 11.3, .imag = -1.5 };
 
     const z = c_cmultd_comp(a.real, a.imag, b.real, b.imag);
-    expect(z.real == 1.5) catch @panic("test failure: zig_complex_double 3");
-    expect(z.imag == 13.5) catch @panic("test failure: zig_complex_double 4");
+    try expect(z.real == 1.5);
+    try expect(z.imag == 13.5);
 }
 
 export fn zig_cmultf(a: ComplexFloat, b: ComplexFloat) ComplexFloat {
@@ -261,7 +265,7 @@ extern fn c_big_struct(BigStruct) void;
 
 test "C ABI big struct" {
     if (comptime builtin.cpu.arch.isMIPS()) return error.SkipZigTest;
-    if (comptime builtin.cpu.arch.isRISCV()) return error.SkipZigTest;
+    if (comptime builtin.cpu.arch.isPPC()) return error.SkipZigTest;
 
     var s = BigStruct{
         .a = 1,
@@ -287,7 +291,7 @@ const BigUnion = extern union {
 extern fn c_big_union(BigUnion) void;
 
 test "C ABI big union" {
-    if (comptime builtin.cpu.arch.isRISCV()) return error.SkipZigTest;
+    if (comptime builtin.cpu.arch.isPPC()) return error.SkipZigTest;
 
     var x = BigUnion{
         .a = BigStruct{
@@ -320,9 +324,9 @@ extern fn c_ret_med_struct_mixed() MedStructMixed;
 
 test "C ABI medium struct of ints and floats" {
     if (builtin.cpu.arch == .i386) return error.SkipZigTest;
-    if (comptime builtin.cpu.arch.isARM()) return error.SkipZigTest;
     if (comptime builtin.cpu.arch.isMIPS()) return error.SkipZigTest;
-    if (comptime builtin.cpu.arch.isRISCV()) return error.SkipZigTest;
+    if (comptime builtin.cpu.arch.isPPC()) return error.SkipZigTest;
+    if (comptime builtin.cpu.arch.isPPC64()) return error.SkipZigTest;
 
     var s = MedStructMixed{
         .a = 1234,
@@ -331,9 +335,9 @@ test "C ABI medium struct of ints and floats" {
     };
     c_med_struct_mixed(s);
     var s2 = c_ret_med_struct_mixed();
-    expect(s2.a == 1234) catch @panic("test failure");
-    expect(s2.b == 100.0) catch @panic("test failure");
-    expect(s2.c == 1337.0) catch @panic("test failure");
+    try expect(s2.a == 1234);
+    try expect(s2.b == 100.0);
+    try expect(s2.c == 1337.0);
 }
 
 export fn zig_med_struct_mixed(x: MedStructMixed) void {
@@ -353,9 +357,9 @@ extern fn c_ret_small_struct_ints() SmallStructInts;
 
 test "C ABI small struct of ints" {
     if (builtin.cpu.arch == .i386) return error.SkipZigTest;
-    if (comptime builtin.cpu.arch.isARM()) return error.SkipZigTest;
     if (comptime builtin.cpu.arch.isMIPS()) return error.SkipZigTest;
-    if (comptime builtin.cpu.arch.isRISCV()) return error.SkipZigTest;
+    if (comptime builtin.cpu.arch.isPPC()) return error.SkipZigTest;
+    if (comptime builtin.cpu.arch.isPPC64()) return error.SkipZigTest;
 
     var s = SmallStructInts{
         .a = 1,
@@ -365,10 +369,10 @@ test "C ABI small struct of ints" {
     };
     c_small_struct_ints(s);
     var s2 = c_ret_small_struct_ints();
-    expect(s2.a == 1) catch @panic("test failure");
-    expect(s2.b == 2) catch @panic("test failure");
-    expect(s2.c == 3) catch @panic("test failure");
-    expect(s2.d == 4) catch @panic("test failure");
+    try expect(s2.a == 1);
+    try expect(s2.b == 2);
+    try expect(s2.c == 3);
+    try expect(s2.d == 4);
 }
 
 export fn zig_small_struct_ints(x: SmallStructInts) void {
@@ -435,9 +439,9 @@ extern fn c_split_struct_ints(SplitStructInt) void;
 
 test "C ABI split struct of ints" {
     if (builtin.cpu.arch == .i386) return error.SkipZigTest;
-    if (comptime builtin.cpu.arch.isARM()) return error.SkipZigTest;
     if (comptime builtin.cpu.arch.isMIPS()) return error.SkipZigTest;
-    if (comptime builtin.cpu.arch.isRISCV()) return error.SkipZigTest;
+    if (comptime builtin.cpu.arch.isPPC()) return error.SkipZigTest;
+    if (comptime builtin.cpu.arch.isPPC64()) return error.SkipZigTest;
 
     var s = SplitStructInt{
         .a = 1234,
@@ -463,9 +467,9 @@ extern fn c_ret_split_struct_mixed() SplitStructMixed;
 
 test "C ABI split struct of ints and floats" {
     if (builtin.cpu.arch == .i386) return error.SkipZigTest;
-    if (comptime builtin.cpu.arch.isARM()) return error.SkipZigTest;
     if (comptime builtin.cpu.arch.isMIPS()) return error.SkipZigTest;
-    if (comptime builtin.cpu.arch.isRISCV()) return error.SkipZigTest;
+    if (comptime builtin.cpu.arch.isPPC()) return error.SkipZigTest;
+    if (comptime builtin.cpu.arch.isPPC64()) return error.SkipZigTest;
 
     var s = SplitStructMixed{
         .a = 1234,
@@ -474,9 +478,9 @@ test "C ABI split struct of ints and floats" {
     };
     c_split_struct_mixed(s);
     var s2 = c_ret_split_struct_mixed();
-    expect(s2.a == 1234) catch @panic("test failure");
-    expect(s2.b == 100) catch @panic("test failure");
-    expect(s2.c == 1337.0) catch @panic("test failure");
+    try expect(s2.a == 1234);
+    try expect(s2.b == 100);
+    try expect(s2.c == 1337.0);
 }
 
 export fn zig_split_struct_mixed(x: SplitStructMixed) void {
@@ -492,7 +496,7 @@ extern fn c_multiple_struct_floats(FloatRect, FloatRect) void;
 
 test "C ABI sret and byval together" {
     if (comptime builtin.cpu.arch.isMIPS()) return error.SkipZigTest;
-    if (comptime builtin.cpu.arch.isRISCV()) return error.SkipZigTest;
+    if (comptime builtin.cpu.arch.isPPC()) return error.SkipZigTest;
 
     var s = BigStruct{
         .a = 1,
@@ -543,9 +547,9 @@ const Vector5 = extern struct {
 extern fn c_big_struct_floats(Vector5) void;
 
 test "C ABI structs of floats as parameter" {
-    if (comptime builtin.cpu.arch.isARM()) return error.SkipZigTest;
     if (comptime builtin.cpu.arch.isMIPS()) return error.SkipZigTest;
-    if (comptime builtin.cpu.arch.isRISCV()) return error.SkipZigTest;
+    if (comptime builtin.cpu.arch.isPPC()) return error.SkipZigTest;
+    if (comptime builtin.cpu.arch.isPPC64()) return error.SkipZigTest;
 
     var v3 = Vector3{
         .x = 3.0,
@@ -584,7 +588,8 @@ export fn zig_multiple_struct_ints(x: Rect, y: Rect) void {
 }
 
 test "C ABI structs of ints as multiple parameters" {
-    if (comptime builtin.cpu.arch.isRISCV()) return error.SkipZigTest;
+    if (comptime builtin.cpu.arch.isPPC()) return error.SkipZigTest;
+    if (comptime builtin.cpu.arch.isPPC64()) return error.SkipZigTest;
 
     var r1 = Rect{
         .left = 1,
@@ -621,7 +626,7 @@ export fn zig_multiple_struct_floats(x: FloatRect, y: FloatRect) void {
 
 test "C ABI structs of floats as multiple parameters" {
     if (comptime builtin.cpu.arch.isMIPS()) return error.SkipZigTest;
-    if (comptime builtin.cpu.arch.isRISCV()) return error.SkipZigTest;
+    if (comptime builtin.cpu.arch.isPPC()) return error.SkipZigTest;
 
     var r1 = FloatRect{
         .left = 1,
@@ -725,15 +730,15 @@ extern fn c_ret_struct_with_array() StructWithArray;
 
 test "Struct with array as padding." {
     if (builtin.cpu.arch == .i386) return error.SkipZigTest;
-    if (comptime builtin.cpu.arch.isARM()) return error.SkipZigTest;
     if (comptime builtin.cpu.arch.isMIPS()) return error.SkipZigTest;
-    if (comptime builtin.cpu.arch.isRISCV()) return error.SkipZigTest;
+    if (comptime builtin.cpu.arch.isPPC()) return error.SkipZigTest;
+    if (comptime builtin.cpu.arch.isPPC64()) return error.SkipZigTest;
 
     c_struct_with_array(.{ .a = 1, .padding = undefined, .b = 2 });
 
     var x = c_ret_struct_with_array();
-    try std.testing.expect(x.a == 4);
-    try std.testing.expect(x.b == 155);
+    try expect(x.a == 4);
+    try expect(x.b == 155);
 }
 
 const FloatArrayStruct = extern struct {
@@ -752,7 +757,7 @@ extern fn c_ret_float_array_struct() FloatArrayStruct;
 
 test "Float array like struct" {
     if (comptime builtin.cpu.arch.isMIPS()) return error.SkipZigTest;
-    if (comptime builtin.cpu.arch.isRISCV()) return error.SkipZigTest;
+    if (comptime builtin.cpu.arch.isPPC()) return error.SkipZigTest;
 
     c_float_array_struct(.{
         .origin = .{
@@ -766,8 +771,63 @@ test "Float array like struct" {
     });
 
     var x = c_ret_float_array_struct();
-    try std.testing.expect(x.origin.x == 1);
-    try std.testing.expect(x.origin.y == 2);
-    try std.testing.expect(x.size.width == 3);
-    try std.testing.expect(x.size.height == 4);
+    try expect(x.origin.x == 1);
+    try expect(x.origin.y == 2);
+    try expect(x.size.width == 3);
+    try expect(x.size.height == 4);
+}
+
+const SmallVec = @Vector(2, u32);
+
+extern fn c_small_vec(SmallVec) void;
+extern fn c_ret_small_vec() SmallVec;
+
+test "small simd vector" {
+    if (builtin.cpu.arch == .i386) return error.SkipZigTest;
+    if (comptime builtin.cpu.arch.isPPC64()) return error.SkipZigTest;
+
+    c_small_vec(.{ 1, 2 });
+
+    var x = c_ret_small_vec();
+    try expect(x[0] == 3);
+    try expect(x[1] == 4);
+}
+
+const BigVec = @Vector(8, usize);
+
+extern fn c_big_vec(BigVec) void;
+extern fn c_ret_big_vec() BigVec;
+
+test "big simd vector" {
+    if (comptime builtin.cpu.arch.isPPC64()) return error.SkipZigTest;
+
+    c_big_vec(.{ 1, 2, 3, 4, 5, 6, 7, 8 });
+
+    var x = c_ret_big_vec();
+    try expect(x[0] == 9);
+    try expect(x[1] == 10);
+    try expect(x[2] == 11);
+    try expect(x[3] == 12);
+    try expect(x[4] == 13);
+    try expect(x[5] == 14);
+    try expect(x[6] == 15);
+    try expect(x[7] == 16);
+}
+
+const Vector2 = extern struct { x: f32, y: f32 };
+
+extern fn c_ptr_size_float_struct(Vector2) void;
+extern fn c_ret_ptr_size_float_struct() Vector2;
+
+test "C ABI pointer sized float struct" {
+    if (builtin.cpu.arch == .i386) return error.SkipZigTest;
+    if (comptime builtin.cpu.arch.isMIPS()) return error.SkipZigTest;
+    if (comptime builtin.cpu.arch.isRISCV()) return error.SkipZigTest;
+    if (comptime builtin.cpu.arch.isPPC()) return error.SkipZigTest;
+
+    c_ptr_size_float_struct(.{ .x = 1, .y = 2 });
+
+    var x = c_ret_ptr_size_float_struct();
+    try expect(x.x == 3);
+    try expect(x.y == 4);
 }
diff --git a/test/standalone.zig b/test/standalone.zig
index 16cb61a85d..f0567943b6 100644
--- a/test/standalone.zig
+++ b/test/standalone.zig
@@ -44,23 +44,6 @@ pub fn addCases(cases: *tests.StandaloneContext) void {
     if (builtin.os.tag != .wasi) {
         cases.addBuildFile("test/standalone/load_dynamic_library/build.zig", .{});
     }
-    // C ABI compatibility issue: https://github.com/ziglang/zig/issues/1481
-    if (builtin.cpu.arch == .x86_64) {
-        if (builtin.zig_backend == .stage1 or builtin.zig_backend == .stage2_llvm) {
-            cases.addBuildFile("test/c_abi/build.zig", .{});
-        }
-    }
-    if (builtin.cpu.arch.isAARCH64() and builtin.zig_backend == .stage2_llvm) {
-        cases.addBuildFile("test/c_abi/build.zig", .{});
-    }
-    if (builtin.cpu.arch == .i386 and builtin.zig_backend == .stage2_llvm) {
-        cases.addBuildFile("test/c_abi/build.zig", .{});
-    }
-    // C ABI tests only pass for the Wasm target when using stage2
-    cases.addBuildFile("test/c_abi/build_wasm.zig", .{
-        .requires_stage2 = true,
-        .use_emulation = true,
-    });
 
     cases.addBuildFile("test/standalone/c_compiler/build.zig", .{
         .build_modes = true,
diff --git a/test/tests.zig b/test/tests.zig
index 4dec71bc52..266d2a37d0 100644
--- a/test/tests.zig
+++ b/test/tests.zig
@@ -1268,3 +1268,82 @@ fn printInvocation(args: []const []const u8) void {
     }
     std.debug.print("\n", .{});
 }
+
+const c_abi_targets = [_]CrossTarget{
+    .{},
+    .{
+        .cpu_arch = .x86_64,
+        .os_tag = .linux,
+        .abi = .musl,
+    },
+    .{
+        .cpu_arch = .i386,
+        .os_tag = .linux,
+        .abi = .musl,
+    },
+    .{
+        .cpu_arch = .aarch64,
+        .os_tag = .linux,
+        .abi = .musl,
+    },
+    .{
+        .cpu_arch = .arm,
+        .os_tag = .linux,
+        .abi = .musleabihf,
+    },
+    .{
+        .cpu_arch = .mips,
+        .os_tag = .linux,
+        .abi = .musl,
+    },
+    .{
+        .cpu_arch = .riscv64,
+        .os_tag = .linux,
+        .abi = .musl,
+    },
+    .{
+        .cpu_arch = .wasm32,
+        .os_tag = .wasi,
+        .abi = .musl,
+    },
+    .{
+        .cpu_arch = .powerpc,
+        .os_tag = .linux,
+        .abi = .musl,
+    },
+    .{
+        .cpu_arch = .powerpc64le,
+        .os_tag = .linux,
+        .abi = .musl,
+    },
+};
+
+pub fn addCAbiTests(b: *build.Builder, skip_non_native: bool) *build.Step {
+    const step = b.step("test-c-abi", "Run the C ABI tests");
+
+    for (c_abi_targets) |c_abi_target| {
+        if (skip_non_native and !c_abi_target.isNative())
+            continue;
+
+        const test_step = b.addTest("test/c_abi/main.zig");
+        test_step.setTarget(c_abi_target);
+        if (c_abi_target.abi != null and c_abi_target.abi.?.isMusl()) {
+            // TODO NativeTargetInfo insists on dynamically linking musl
+            // for some reason?
+            test_step.target_info.dynamic_linker.max_byte = null;
+        }
+        test_step.linkLibC();
+        test_step.addCSourceFile("test/c_abi/cfuncs.c", &.{"-std=c99"});
+
+        const triple_prefix = c_abi_target.zigTriple(b.allocator) catch unreachable;
+        test_step.setNamePrefix(b.fmt("{s}-{s} ", .{
+            "test-c-abi",
+            triple_prefix,
+        }));
+
+        test_step.use_stage1 = false;
+
+        step.dependOn(&test_step.step);
+    }
+    return step;
+}