Fix arm64 ABI: '#simd' vectors and vector aggregates not passed in registers (fixes #7010)

src/llvm_abi.cpp

@@ -1147,6 +1147,10 @@ namespace lbAbiArm64 {
 		case LLVMDoubleTypeKind:
 		case LLVMPointerTypeKind:
 			return true;
+		case LLVMVectorTypeKind:{
+				i32 sz = lb_sizeof(type);
+				return sz == 8 || sz == 16;
+			}
 		}
 		return false;
 	}
@@ -1228,6 +1232,15 @@ namespace lbAbiArm64 {
 			if (base_type_) *base_type_ = type;
 			if (member_count_) *member_count_ = 1;
 			return true;
+		case LLVMVectorTypeKind:{
+			i64 sz = lb_sizeof(type);
+			if (sz == 8 || sz == 16){
+				if (base_type_) *base_type_ = type;
+				if (member_count_) *member_count_ = 1;
+				return true;
+			}
+			return false;
+		}
 		case LLVMArrayTypeKind:
 			return is_homogenous_array(c, type, base_type_, member_count_);
 		case LLVMStructTypeKind:

tests/issues/run.sh

@@ -81,6 +81,10 @@ else
 fi
 $ODIN check ../test_issue_6979.odin -no-entry-point $COMMON
 
+
+clang -c ../test_issue_7010.c -o test_issue_7010_c.o
+$ODIN test ../test_issue_7010.odin $COMMON
+
 set +x
 
 popd

tests/issues/test_issue_7010.c

@@ -0,0 +1,51 @@
+// Tests issue: https://github.com/odin-lang/Odin/issues/7010
+
+// #include <simd/types.h>
+// #ifdef __APPLE__
+// #include <simd/simd.h>
+// #else
+typedef __attribute__((__ext_vector_type__(2))) float simd_float2;
+typedef __attribute__((__ext_vector_type__(4))) float simd_float4;
+typedef __attribute__((__ext_vector_type__(2))) unsigned int simd_uint2;
+typedef struct { simd_float4 columns[4]; } simd_float4x4;
+typedef struct { simd_float2 columns[4]; } simd_float4x2;
+// typedef simd_float4x4 matrix_float4x4;
+// #endif
+
+typedef struct { simd_float4 columns[5]; } simd_float5x4;
+
+simd_float4 c_add_vec4f32(simd_float4 a, simd_float4 b) {
+  	return a + b;
+}
+
+simd_uint2 c_add_vec2u32(simd_uint2 a, simd_uint2 b) {
+    return a + b;
+}
+
+simd_float4x4 c_add_matrix4x4(simd_float4x4 a, simd_float4x4 b) {
+  simd_float4x4 copy_mat = a;
+  copy_mat.columns[0] += b.columns[0];
+  copy_mat.columns[1] += b.columns[1];
+  copy_mat.columns[2] += b.columns[2];
+  copy_mat.columns[3] += b.columns[3];
+  return copy_mat;
+}
+
+simd_float5x4 c_add_matrix5x4(simd_float5x4 a, simd_float5x4 b) {
+  simd_float5x4 copy_mat = a;
+  copy_mat.columns[0] += b.columns[0];
+  copy_mat.columns[1] += b.columns[1];
+  copy_mat.columns[2] += b.columns[2];
+  copy_mat.columns[3] += b.columns[3];
+  copy_mat.columns[4] += b.columns[4];
+  return copy_mat;
+}
+
+simd_float4x2 c_add_matrix4x2(simd_float4x2 a, simd_float4x2 b) {
+  simd_float4x2 copy_mat = a;
+  copy_mat.columns[0] += b.columns[0];
+  copy_mat.columns[1] += b.columns[1];
+  copy_mat.columns[2] += b.columns[2];
+  copy_mat.columns[3] += b.columns[3];
+  return copy_mat;
+}

tests/issues/test_issue_7010.odin

@@ -0,0 +1,164 @@
+// Tests issue: https://github.com/odin-lang/Odin/issues/7010
+
+package test_issues
+
+Matrix4x4 :: struct {
+	columns: [4]#simd[4]f32,
+}
+
+Matrix5x4 :: struct {
+	columns: [5]#simd[4]f32,
+}
+
+Matrix4x2 :: struct {
+	columns: [4]#simd[2]f32,
+}
+
+odin_add_vec4f32 :: proc (a: #simd[4]f32, b: #simd[4]f32) -> #simd[4]f32 {
+  	return a + b
+}
+
+odin_add_vec2u32 :: proc (a: #simd[2]u32, b: #simd[2]u32) -> #simd[2]u32 {
+  	return a + b
+}
+
+odin_add_matrix4x4 :: proc (a: Matrix4x4, b: Matrix4x4) -> Matrix4x4 {
+	copy_mat := a
+	copy_mat.columns[0] += b.columns[0]
+	copy_mat.columns[1] += b.columns[1]
+	copy_mat.columns[2] += b.columns[2]
+	copy_mat.columns[3] += b.columns[3]
+	return copy_mat
+}
+
+odin_add_matrix5x4 :: proc (a: Matrix5x4, b: Matrix5x4) -> Matrix5x4 {
+	copy_mat := a
+	copy_mat.columns[0] += b.columns[0]
+	copy_mat.columns[1] += b.columns[1]
+	copy_mat.columns[2] += b.columns[2]
+	copy_mat.columns[3] += b.columns[3]
+	copy_mat.columns[4] += b.columns[4]
+	return copy_mat
+}
+
+odin_add_matrix4x2 :: proc (a: Matrix4x2, b: Matrix4x2) -> Matrix4x2 {
+	copy_mat := a
+	copy_mat.columns[0] += b.columns[0]
+	copy_mat.columns[1] += b.columns[1]
+	copy_mat.columns[2] += b.columns[2]
+	copy_mat.columns[3] += b.columns[3]
+	return copy_mat
+}
+
+foreign import ctest "build/test_issue_7010_c.o"
+foreign ctest {
+	c_add_vec4f32 :: proc "c" (a: #simd[4]f32, b: #simd[4]f32) -> #simd[4]f32 ---
+	c_add_vec2u32 :: proc "c" (a: #simd[2]u32, b: #simd[2]u32) -> #simd[2]u32 ---
+	c_add_matrix4x4 :: proc "c" (a: Matrix4x4, b: Matrix4x4) -> Matrix4x4 ---
+	c_add_matrix5x4 :: proc "c" (a: Matrix5x4, b: Matrix5x4) -> Matrix5x4 ---
+	c_add_matrix4x2 :: proc "c" (a: Matrix4x2, b: Matrix4x2) -> Matrix4x2 ---
+}
+
+import "core:testing"
+
+@test
+test_simd :: proc(t: ^testing.T) {
+	a_f32 :#simd[4]f32 = {1.0, 2.0, 3.0, 4.0}
+	b_f32 :#simd[4]f32 = {4.0, 3.0, 2.0, 1.0}
+
+	a_u32 :#simd[2]u32 = {1, 2}
+	b_u32 :#simd[2]u32 = {4, 3}
+
+	mat_a_4x4: Matrix4x4 = {
+		{{1.0, 0.0, 0.0, 0.0},
+		{0.0, 2.0, 0.0, 0.0},
+		{0.0, 0.0, 3.0, 0.0},
+		{0.0, 0.0, 0.0, 4.0}},
+	}
+
+	mat_b_4x4: Matrix4x4 = {
+		{{0.0, 0.0, 0.0, 0.0},
+		{0.0, 1.0, 2.0, 0.0},
+		{0.0, 3.0, 4.0, 0.0},
+		{0.0, 0.0, 0.0, 0.0}},
+	}
+
+	mat_a_5x4: Matrix5x4 = {
+		{{1.0, 0.0, 0.0, 0.0},
+		{0.0, 2.0, 0.0, 0.0},
+		{0.0, 0.0, 3.0, 0.0},
+		{0.0, 0.0, 0.0, 4.0},
+		{1.0, 0.0, 0.0, 1.0}},
+	}
+
+	mat_b_5x4: Matrix5x4 = {
+		{{0.0, 0.0, 0.0, 0.0},
+		{0.0, 1.0, 2.0, 0.0},
+		{0.0, 3.0, 4.0, 0.0},
+		{0.0, 0.0, 0.0, 0.0},
+		{0.0, 1.0, 1.0, 0.0}},
+	}
+
+	mat_a_4x2: Matrix4x2 = {
+		{{1.0, 0.0},
+		 {0.0, 2.0},
+		 {0.0, 0.0},
+		 {0.0, 0.0}},
+	}
+
+	mat_b_4x2: Matrix4x2 = {
+		{{0.0, 0.0},
+		 {0.0, 1.0},
+		 {0.0, 3.0},
+		 {1.0, 0.0}},
+	}
+
+	odin_add_vec4_f32_result := odin_add_vec4f32(a_f32, b_f32)
+	testing.expect(t, odin_add_vec4_f32_result == {5.0, 5.0, 5.0, 5.0})
+
+	odin_add_vec2_u32_result := odin_add_vec2u32(a_u32, b_u32)
+	testing.expect(t, odin_add_vec2_u32_result == {5, 5})
+
+	odin_add_matrix4x4_result := odin_add_matrix4x4(mat_a_4x4, mat_b_4x4)
+	testing.expect(t, odin_add_matrix4x4_result == {
+		{{1.0, 0.0, 0.0, 0.0},
+		{0.0, 3.0, 2.0, 0.0},
+		{0.0, 3.0, 7.0, 0.0},
+		{0.0, 0.0, 0.0, 4.0}},
+	})
+
+	odin_add_matrix5x4_result := odin_add_matrix5x4(mat_a_5x4, mat_b_5x4)
+	testing.expect(t, odin_add_matrix5x4_result == {
+		{{1.0, 0.0, 0.0, 0.0},
+		{0.0, 3.0, 2.0, 0.0},
+		{0.0, 3.0, 7.0, 0.0},
+		{0.0, 0.0, 0.0, 4.0},
+		{1.0, 1.0, 1.0, 1.0}},
+	})
+
+	odin_add_matrix4x2_result := odin_add_matrix4x2(mat_a_4x2, mat_b_4x2)
+	testing.expect(t, odin_add_matrix4x2_result == {
+		{{1.0, 0.0},
+		 {0.0, 3.0},
+		 {0.0, 3.0},
+		 {1.0, 0.0}},
+	})
+
+	c_add_vec4_f32_result := c_add_vec4f32(a_f32, b_f32)
+	testing.expect(t, c_add_vec4_f32_result == odin_add_vec4_f32_result)
+
+	c_add_vec2_u32_result := c_add_vec2u32(a_u32, b_u32)
+	testing.expect(t, c_add_vec2_u32_result == odin_add_vec2_u32_result)
+
+	c_add_matrix4x4_result := c_add_matrix4x4(mat_a_4x4, mat_b_4x4)
+	testing.expect(t, c_add_matrix4x4_result == odin_add_matrix4x4_result)
+
+	c_add_matrix5x4_result := c_add_matrix5x4(mat_a_5x4, mat_b_5x4)
+	testing.expect(t, c_add_matrix5x4_result == odin_add_matrix5x4_result)
+
+	c_add_matrix4x2_result := c_add_matrix4x2(mat_a_4x2, mat_b_4x2)
+	testing.expect(t, c_add_matrix4x2_result == odin_add_matrix4x2_result)
+}
+
+// clang -c test_issue_7010.c -o test_issue_7010_c.o
+// odin test test_issue_7010.odin -file