diff options
Diffstat (limited to 'compiler/rustc_codegen_cranelift/src/intrinsics')
3 files changed, 122 insertions, 15 deletions
diff --git a/compiler/rustc_codegen_cranelift/src/intrinsics/llvm_x86.rs b/compiler/rustc_codegen_cranelift/src/intrinsics/llvm_x86.rs index 7bc161fbe55..d2ae6978ca2 100644 --- a/compiler/rustc_codegen_cranelift/src/intrinsics/llvm_x86.rs +++ b/compiler/rustc_codegen_cranelift/src/intrinsics/llvm_x86.rs @@ -33,8 +33,8 @@ pub(crate) fn codegen_x86_llvm_intrinsic_call<'tcx>( // cast float to int let a_lane = match lane_ty { - types::F32 => fx.bcx.ins().bitcast(types::I32, a_lane), - types::F64 => fx.bcx.ins().bitcast(types::I64, a_lane), + types::F32 => codegen_bitcast(fx, types::I32, a_lane), + types::F64 => codegen_bitcast(fx, types::I64, a_lane), _ => a_lane, }; diff --git a/compiler/rustc_codegen_cranelift/src/intrinsics/mod.rs b/compiler/rustc_codegen_cranelift/src/intrinsics/mod.rs index e4ac89a7bec..52720daac6f 100644 --- a/compiler/rustc_codegen_cranelift/src/intrinsics/mod.rs +++ b/compiler/rustc_codegen_cranelift/src/intrinsics/mod.rs @@ -200,7 +200,7 @@ fn bool_to_zero_or_max_uint<'tcx>( let mut res = fx.bcx.ins().bmask(int_ty, val); if ty.is_float() { - res = fx.bcx.ins().bitcast(ty, res); + res = codegen_bitcast(fx, ty, res); } res @@ -240,10 +240,9 @@ pub(crate) fn codegen_intrinsic_call<'tcx>( substs, args, destination, + target, source_info.span, ); - let ret_block = fx.get_block(target); - fx.bcx.ins().jump(ret_block, &[]); } else if codegen_float_intrinsic_call(fx, intrinsic, args, destination) { let ret_block = fx.get_block(target); fx.bcx.ins().jump(ret_block, &[]); @@ -650,7 +649,7 @@ fn codegen_regular_intrinsic_call<'tcx>( let layout = fx.layout_of(substs.type_at(0)); if layout.abi.is_uninhabited() { with_no_trimmed_paths!({ - crate::base::codegen_panic( + crate::base::codegen_panic_nounwind( fx, &format!("attempted to instantiate uninhabited type `{}`", layout.ty), source_info, @@ -661,7 +660,7 @@ fn codegen_regular_intrinsic_call<'tcx>( if intrinsic == sym::assert_zero_valid && !fx.tcx.permits_zero_init(layout) { with_no_trimmed_paths!({ - crate::base::codegen_panic( + crate::base::codegen_panic_nounwind( fx, &format!( "attempted to zero-initialize type `{}`, which is invalid", @@ -677,7 +676,7 @@ fn codegen_regular_intrinsic_call<'tcx>( && !fx.tcx.permits_uninit_init(layout) { with_no_trimmed_paths!({ - crate::base::codegen_panic( + crate::base::codegen_panic_nounwind( fx, &format!( "attempted to leave type `{}` uninitialized, which is invalid", diff --git a/compiler/rustc_codegen_cranelift/src/intrinsics/simd.rs b/compiler/rustc_codegen_cranelift/src/intrinsics/simd.rs index 14f5e918739..b33eb29754a 100644 --- a/compiler/rustc_codegen_cranelift/src/intrinsics/simd.rs +++ b/compiler/rustc_codegen_cranelift/src/intrinsics/simd.rs @@ -24,6 +24,7 @@ pub(super) fn codegen_simd_intrinsic_call<'tcx>( _substs: SubstsRef<'tcx>, args: &[mir::Operand<'tcx>], ret: CPlace<'tcx>, + target: BasicBlock, span: Span, ) { match intrinsic { @@ -277,16 +278,15 @@ pub(super) fn codegen_simd_intrinsic_call<'tcx>( } else { fx.tcx.sess.span_warn(span, "Index argument for `simd_extract` is not a constant"); let trap_block = fx.bcx.create_block(); - let dummy_block = fx.bcx.create_block(); let true_ = fx.bcx.ins().iconst(types::I8, 1); fx.bcx.ins().brnz(true_, trap_block, &[]); - fx.bcx.ins().jump(dummy_block, &[]); + let ret_block = fx.get_block(target); + fx.bcx.ins().jump(ret_block, &[]); fx.bcx.switch_to_block(trap_block); crate::trap::trap_unimplemented( fx, "Index argument for `simd_extract` is not a constant", ); - fx.bcx.switch_to_block(dummy_block); return; }; @@ -770,11 +770,119 @@ pub(super) fn codegen_simd_intrinsic_call<'tcx>( }); } - // simd_arith_offset - // simd_scatter - // simd_gather + sym::simd_expose_addr | sym::simd_from_exposed_addr | sym::simd_cast_ptr => { + intrinsic_args!(fx, args => (arg); intrinsic); + ret.write_cvalue_transmute(fx, arg); + } + + sym::simd_arith_offset => { + intrinsic_args!(fx, args => (ptr, offset); intrinsic); + + let (lane_count, ptr_lane_ty) = ptr.layout().ty.simd_size_and_type(fx.tcx); + let pointee_ty = ptr_lane_ty.builtin_deref(true).unwrap().ty; + let pointee_size = fx.layout_of(pointee_ty).size.bytes(); + let (ret_lane_count, ret_lane_ty) = ret.layout().ty.simd_size_and_type(fx.tcx); + let ret_lane_layout = fx.layout_of(ret_lane_ty); + assert_eq!(lane_count, ret_lane_count); + + for lane_idx in 0..lane_count { + let ptr_lane = ptr.value_lane(fx, lane_idx).load_scalar(fx); + let offset_lane = offset.value_lane(fx, lane_idx).load_scalar(fx); + + let ptr_diff = if pointee_size != 1 { + fx.bcx.ins().imul_imm(offset_lane, pointee_size as i64) + } else { + offset_lane + }; + let res_lane = fx.bcx.ins().iadd(ptr_lane, ptr_diff); + let res_lane = CValue::by_val(res_lane, ret_lane_layout); + + ret.place_lane(fx, lane_idx).write_cvalue(fx, res_lane); + } + } + + sym::simd_gather => { + intrinsic_args!(fx, args => (val, ptr, mask); intrinsic); + + let (val_lane_count, val_lane_ty) = val.layout().ty.simd_size_and_type(fx.tcx); + let (ptr_lane_count, _ptr_lane_ty) = ptr.layout().ty.simd_size_and_type(fx.tcx); + let (mask_lane_count, _mask_lane_ty) = mask.layout().ty.simd_size_and_type(fx.tcx); + let (ret_lane_count, ret_lane_ty) = ret.layout().ty.simd_size_and_type(fx.tcx); + assert_eq!(val_lane_count, ptr_lane_count); + assert_eq!(val_lane_count, mask_lane_count); + assert_eq!(val_lane_count, ret_lane_count); + + let lane_clif_ty = fx.clif_type(val_lane_ty).unwrap(); + let ret_lane_layout = fx.layout_of(ret_lane_ty); + + for lane_idx in 0..ptr_lane_count { + let val_lane = val.value_lane(fx, lane_idx).load_scalar(fx); + let ptr_lane = ptr.value_lane(fx, lane_idx).load_scalar(fx); + let mask_lane = mask.value_lane(fx, lane_idx).load_scalar(fx); + + let if_enabled = fx.bcx.create_block(); + let if_disabled = fx.bcx.create_block(); + let next = fx.bcx.create_block(); + let res_lane = fx.bcx.append_block_param(next, lane_clif_ty); + + fx.bcx.ins().brnz(mask_lane, if_enabled, &[]); + fx.bcx.ins().jump(if_disabled, &[]); + fx.bcx.seal_block(if_enabled); + fx.bcx.seal_block(if_disabled); + + fx.bcx.switch_to_block(if_enabled); + let res = fx.bcx.ins().load(lane_clif_ty, MemFlags::trusted(), ptr_lane, 0); + fx.bcx.ins().jump(next, &[res]); + + fx.bcx.switch_to_block(if_disabled); + fx.bcx.ins().jump(next, &[val_lane]); + + fx.bcx.seal_block(next); + fx.bcx.switch_to_block(next); + + fx.bcx.ins().nop(); + + ret.place_lane(fx, lane_idx) + .write_cvalue(fx, CValue::by_val(res_lane, ret_lane_layout)); + } + } + + sym::simd_scatter => { + intrinsic_args!(fx, args => (val, ptr, mask); intrinsic); + + let (val_lane_count, _val_lane_ty) = val.layout().ty.simd_size_and_type(fx.tcx); + let (ptr_lane_count, _ptr_lane_ty) = ptr.layout().ty.simd_size_and_type(fx.tcx); + let (mask_lane_count, _mask_lane_ty) = mask.layout().ty.simd_size_and_type(fx.tcx); + assert_eq!(val_lane_count, ptr_lane_count); + assert_eq!(val_lane_count, mask_lane_count); + + for lane_idx in 0..ptr_lane_count { + let val_lane = val.value_lane(fx, lane_idx).load_scalar(fx); + let ptr_lane = ptr.value_lane(fx, lane_idx).load_scalar(fx); + let mask_lane = mask.value_lane(fx, lane_idx).load_scalar(fx); + + let if_enabled = fx.bcx.create_block(); + let next = fx.bcx.create_block(); + + fx.bcx.ins().brnz(mask_lane, if_enabled, &[]); + fx.bcx.ins().jump(next, &[]); + fx.bcx.seal_block(if_enabled); + + fx.bcx.switch_to_block(if_enabled); + fx.bcx.ins().store(MemFlags::trusted(), val_lane, ptr_lane, 0); + fx.bcx.ins().jump(next, &[]); + + fx.bcx.seal_block(next); + fx.bcx.switch_to_block(next); + } + } + _ => { - fx.tcx.sess.span_fatal(span, &format!("Unknown SIMD intrinsic {}", intrinsic)); + fx.tcx.sess.span_err(span, &format!("Unknown SIMD intrinsic {}", intrinsic)); + // Prevent verifier error + fx.bcx.ins().trap(TrapCode::UnreachableCodeReached); } } + let ret_block = fx.get_block(target); + fx.bcx.ins().jump(ret_block, &[]); } |