diff options
Diffstat (limited to 'src')
| -rw-r--r-- | src/Sema.zig | 85 | ||||
| -rw-r--r-- | src/codegen/llvm.zig | 23 |
2 files changed, 77 insertions, 31 deletions
diff --git a/src/Sema.zig b/src/Sema.zig index 7805d7f095..8fda67b652 100644 --- a/src/Sema.zig +++ b/src/Sema.zig @@ -3205,14 +3205,11 @@ fn zirValidateArrayInit( // instruction after it within the same block. // Possible performance enhancement: save the `block_index` between iterations // of the for loop. - const next_air_inst = inst: { - var block_index = block.instructions.items.len - 1; - while (block.instructions.items[block_index] != elem_ptr_air_inst) { - block_index -= 1; - } - first_block_index = @minimum(first_block_index, block_index); - break :inst block.instructions.items[block_index + 1]; - }; + var block_index = block.instructions.items.len - 1; + while (block.instructions.items[block_index] != elem_ptr_air_inst) { + block_index -= 1; + } + first_block_index = @minimum(first_block_index, block_index); // Array has one possible value, so value is always comptime-known if (opt_opv) |opv| { @@ -3222,6 +3219,7 @@ fn zirValidateArrayInit( // If the next instructon is a store with a comptime operand, this element // is comptime. + const next_air_inst = block.instructions.items[block_index + 1]; switch (air_tags[next_air_inst]) { .store => { const bin_op = air_datas[next_air_inst].bin_op; @@ -13491,30 +13489,61 @@ fn zirByteSwap(sema: *Sema, block: *Block, inst: Zir.Inst.Index) CompileError!Ai const operand_src: LazySrcLoc = .{ .node_offset_builtin_call_arg1 = inst_data.src_node }; const operand = sema.resolveInst(inst_data.operand); const operand_ty = sema.typeOf(operand); - // TODO implement support for vectors - if (operand_ty.zigTypeTag() != .Int) { - return sema.fail(block, ty_src, "expected integer type, found '{}'", .{ - operand_ty, - }); - } + const scalar_ty = try sema.checkIntOrVectorAllowComptime(block, operand, operand_src); const target = sema.mod.getTarget(); - const bits = operand_ty.intInfo(target).bits; - if (bits == 0) return Air.Inst.Ref.zero; - if (operand_ty.intInfo(target).bits % 8 != 0) { - return sema.fail(block, ty_src, "@byteSwap requires the number of bits to be evenly divisible by 8, but {} has {} bits", .{ - operand_ty, - operand_ty.intInfo(target).bits, - }); + const bits = scalar_ty.intInfo(target).bits; + if (bits % 8 != 0) { + return sema.fail( + block, + ty_src, + "@byteSwap requires the number of bits to be evenly divisible by 8, but {} has {} bits", + .{ scalar_ty, bits }, + ); } - const runtime_src = if (try sema.resolveMaybeUndefVal(block, operand_src, operand)) |val| { - if (val.isUndef()) return sema.addConstUndef(operand_ty); - const result_val = try val.byteSwap(operand_ty, target, sema.arena); - return sema.addConstant(operand_ty, result_val); - } else operand_src; + switch (operand_ty.zigTypeTag()) { + .Int, .ComptimeInt => { + if (bits == 0) return Air.Inst.Ref.zero; - try sema.requireRuntimeBlock(block, runtime_src); - return block.addTyOp(.byte_swap, operand_ty, operand); + const runtime_src = if (try sema.resolveMaybeUndefVal(block, operand_src, operand)) |val| { + if (val.isUndef()) return sema.addConstUndef(operand_ty); + const result_val = try val.byteSwap(operand_ty, target, sema.arena); + return sema.addConstant(operand_ty, result_val); + } else operand_src; + + try sema.requireRuntimeBlock(block, runtime_src); + return block.addTyOp(.byte_swap, operand_ty, operand); + }, + .Vector => { + if (bits == 0) { + return sema.addConstant( + operand_ty, + try Value.Tag.repeated.create(sema.arena, Value.zero), + ); + } + + const runtime_src = if (try sema.resolveMaybeUndefVal(block, operand_src, operand)) |val| { + if (val.isUndef()) + return sema.addConstUndef(operand_ty); + + const vec_len = operand_ty.vectorLen(); + var elem_buf: Value.ElemValueBuffer = undefined; + const elems = try sema.arena.alloc(Value, vec_len); + for (elems) |*elem, i| { + const elem_val = val.elemValueBuffer(i, &elem_buf); + elem.* = try elem_val.byteSwap(operand_ty, target, sema.arena); + } + return sema.addConstant( + operand_ty, + try Value.Tag.aggregate.create(sema.arena, elems), + ); + } else operand_src; + + try sema.requireRuntimeBlock(block, runtime_src); + return block.addTyOp(.byte_swap, operand_ty, operand); + }, + else => unreachable, + } } fn zirBitReverse(sema: *Sema, block: *Block, inst: Zir.Inst.Index) CompileError!Air.Inst.Ref { diff --git a/src/codegen/llvm.zig b/src/codegen/llvm.zig index 481730452c..5a231ddc6e 100644 --- a/src/codegen/llvm.zig +++ b/src/codegen/llvm.zig @@ -6078,9 +6078,26 @@ pub const FuncGen = struct { if (bits % 16 == 8) { // If not an even byte-multiple, we need zero-extend + shift-left 1 byte // The truncated result at the end will be the correct bswap - operand_llvm_ty = self.context.intType(bits + 8); - const extended = self.builder.buildZExt(operand, operand_llvm_ty, ""); - operand = self.builder.buildShl(extended, operand_llvm_ty.constInt(8, .False), ""); + const scalar_llvm_ty = self.context.intType(bits + 8); + if (operand_ty.zigTypeTag() == .Vector) { + const vec_len = operand_ty.vectorLen(); + operand_llvm_ty = scalar_llvm_ty.vectorType(vec_len); + + const shifts = try self.gpa.alloc(*const llvm.Value, vec_len); + defer self.gpa.free(shifts); + + for (shifts) |*elem| { + elem.* = scalar_llvm_ty.constInt(8, .False); + } + const shift_vec = llvm.constVector(shifts.ptr, vec_len); + + const extended = self.builder.buildZExt(operand, operand_llvm_ty, ""); + operand = self.builder.buildShl(extended, shift_vec, ""); + } else { + const extended = self.builder.buildZExt(operand, scalar_llvm_ty, ""); + operand = self.builder.buildShl(extended, scalar_llvm_ty.constInt(8, .False), ""); + operand_llvm_ty = scalar_llvm_ty; + } bits = bits + 8; } |