From d0277a3d17f738036a951a44f367839597dcaee4 Mon Sep 17 00:00:00 2001 From: Daniele Cocca Date: Sun, 13 Mar 2022 00:43:51 +0000 Subject: [PATCH] CBE: implement popCount, byteSwap, bitReverse for ints <= 128 bits This folds the airCountZeroes() code from 226fcd7c709ec664c5d883042cf7beb3026f66cb back into airBuiltinCall(), since most of these builtins happen to require the same arguments and can be unified under a common function signature. --- src/codegen/c.zig | 31 ++++-------- src/link/C/zig.h | 96 +++++++++++++++++++++++++++++++++++- test/behavior.zig | 2 +- test/behavior/bitreverse.zig | 2 - test/behavior/byteswap.zig | 1 - test/behavior/popcount.zig | 3 ++ 6 files changed, 107 insertions(+), 28 deletions(-) diff --git a/src/codegen/c.zig b/src/codegen/c.zig index 39b84dde3c..cee42b0603 100644 --- a/src/codegen/c.zig +++ b/src/codegen/c.zig @@ -1711,8 +1711,8 @@ fn genBody(f: *Function, body: []const Air.Inst.Index) error{ AnalysisFail, OutO .memcpy => try airMemcpy(f, inst), .set_union_tag => try airSetUnionTag(f, inst), .get_union_tag => try airGetUnionTag(f, inst), - .clz => try airCountZeroes(f, inst, "clz"), - .ctz => try airCountZeroes(f, inst, "ctz"), + .clz => try airBuiltinCall(f, inst, "clz"), + .ctz => try airBuiltinCall(f, inst, "ctz"), .popcount => try airBuiltinCall(f, inst, "popcount"), .byte_swap => try airBuiltinCall(f, inst, "byte_swap"), .bit_reverse => try airBuiltinCall(f, inst, "bit_reverse"), @@ -3352,23 +3352,6 @@ fn airPtrToInt(f: *Function, inst: Air.Inst.Index) !CValue { fn airBuiltinCall(f: *Function, inst: Air.Inst.Index, fn_name: [*:0]const u8) !CValue { if (f.liveness.isUnused(inst)) return CValue.none; - const inst_ty = f.air.typeOfIndex(inst); - const local = try f.allocLocal(inst_ty, .Const); - const ty_op = f.air.instructions.items(.data)[inst].ty_op; - const writer = f.object.writer(); - const operand = try f.resolveInst(ty_op.operand); - - // TODO implement the function in zig.h and call it here - - try writer.print(" = {s}(", .{fn_name}); - try f.writeCValue(writer, operand); - try writer.writeAll(");\n"); - return local; -} - -fn airCountZeroes(f: *Function, inst: Air.Inst.Index, fn_name: [*:0]const u8) !CValue { - if (f.liveness.isUnused(inst)) return CValue.none; - const inst_ty = f.air.typeOfIndex(inst); const local = try f.allocLocal(inst_ty, .Const); const operand = f.air.instructions.items(.data)[inst].ty_op.operand; @@ -3376,13 +3359,17 @@ fn airCountZeroes(f: *Function, inst: Air.Inst.Index, fn_name: [*:0]const u8) !C const target = f.object.dg.module.getTarget(); const writer = f.object.writer(); - const zig_bits = operand_ty.intInfo(target).bits; - _ = toCIntBits(zig_bits) orelse + const int_info = operand_ty.intInfo(target); + _ = toCIntBits(int_info.bits) orelse return f.fail("TODO: C backend: implement integer types larger than 128 bits", .{}); + const signed_type = switch (int_info.signedness) { + .signed => "true", + .unsigned => "false", + }; try writer.print(" = zig_{s}(", .{fn_name}); try f.writeCValue(writer, try f.resolveInst(operand)); - try writer.print(", {d});\n", .{zig_bits}); + try writer.print(", {d}, {s});\n", .{ int_info.bits, signed_type }); return local; } diff --git a/src/link/C/zig.h b/src/link/C/zig.h index a5534f86aa..64bf106f3b 100644 --- a/src/link/C/zig.h +++ b/src/link/C/zig.h @@ -502,8 +502,20 @@ zig_shl_sat_s(int, int, ((sizeof(int )) * CHAR_BIT - 1)) zig_shl_sat_s(long, long, ((sizeof(long )) * CHAR_BIT - 1)) #define zig_bitsizeof(T) (CHAR_BIT * sizeof(T)) +#define zig_bit_mask(T, bit_width) \ + ((bit_width) == 128 \ + ? ((T)(((uint128_t) 0xffffffffffffffff) << 64 | \ + (uint128_t) 0xffffffffffffffff)) \ + : (((T)1 << (T)(bit_width)) - 1)) + +static inline uint128_t zig_sign_extend(uint128_t value, uint128_t zig_type_bit_width) { + const uint128_t m = (uint128_t)1 << (zig_type_bit_width - 1); + return (value ^ m) - m; +} + +static inline int zig_clz(unsigned long long value, uint8_t zig_type_bit_width, bool signed_type) { + (void)signed_type; // unused -static inline int zig_clz(uint64_t value, uint8_t zig_type_bit_width) { if (value == 0) return zig_type_bit_width; if (zig_type_bit_width <= zig_bitsizeof(unsigned int)) return (__builtin_clz(value) - zig_bitsizeof(unsigned int) + zig_type_bit_width); @@ -512,9 +524,89 @@ static inline int zig_clz(uint64_t value, uint8_t zig_type_bit_width) { return (__builtin_clzll(value) - zig_bitsizeof(unsigned long long) + zig_type_bit_width); } -static inline int zig_ctz(uint64_t value, uint8_t zig_type_bit_width) { +static inline int zig_ctz(unsigned long long value, uint8_t zig_type_bit_width, bool signed_type) { + (void)signed_type; // unused + if (value == 0) return zig_type_bit_width; if (zig_type_bit_width <= zig_bitsizeof(unsigned int)) return __builtin_ctz(value); if (zig_type_bit_width <= zig_bitsizeof(unsigned long)) return __builtin_ctzl(value); return __builtin_ctzll(value); } + +static inline int zig_popcount(unsigned long long value, uint8_t zig_type_bit_width, bool signed_type) { + (void)signed_type; // unused + + const unsigned long long mask = zig_bit_mask(unsigned long long, zig_type_bit_width); + if (zig_type_bit_width <= zig_bitsizeof(unsigned int)) + return __builtin_popcount(value & mask); + if (zig_type_bit_width <= zig_bitsizeof(unsigned long)) + return __builtin_popcountl(value & mask); + return __builtin_popcountll(value & mask); +} + +static inline uint128_t zig_byte_swap(uint128_t value, uint8_t zig_type_bit_width, bool signed_type) { + if (zig_type_bit_width <= 8) return value; + if (zig_type_bit_width <= 16) return __builtin_bswap16(value); + if (zig_type_bit_width <= 32) { + const uint32_t swapped = __builtin_bswap32(value) >> (32 - zig_type_bit_width); + return (signed_type ? zig_sign_extend(swapped, zig_type_bit_width) : swapped); + } + if (zig_type_bit_width <= 64) { + const uint64_t swapped = __builtin_bswap64(value) >> (64 - zig_type_bit_width); + return (signed_type ? zig_sign_extend(swapped, zig_type_bit_width) : swapped); + } + const uint128_t mask = zig_bit_mask(uint128_t, zig_type_bit_width); + const uint128_t hi = __builtin_bswap64((uint64_t)(value >> 64)); + const uint128_t lo = __builtin_bswap64((uint64_t)value); + const uint128_t swapped = (((lo << 64 | hi) >> (128 - zig_type_bit_width))) & mask; + return (signed_type ? zig_sign_extend(swapped, zig_type_bit_width) : swapped); +} + +static inline uint128_t zig_bit_reverse(uint128_t value, uint8_t zig_type_bit_width, bool signed_type) { + static const uint128_t reverse_lut[256] = { + 0x00, 0x80, 0x40, 0xc0, 0x20, 0xa0, 0x60, 0xe0, 0x10, 0x90, 0x50, 0xd0, + 0x30, 0xb0, 0x70, 0xf0, 0x08, 0x88, 0x48, 0xc8, 0x28, 0xa8, 0x68, 0xe8, + 0x18, 0x98, 0x58, 0xd8, 0x38, 0xb8, 0x78, 0xf8, 0x04, 0x84, 0x44, 0xc4, + 0x24, 0xa4, 0x64, 0xe4, 0x14, 0x94, 0x54, 0xd4, 0x34, 0xb4, 0x74, 0xf4, + 0x0c, 0x8c, 0x4c, 0xcc, 0x2c, 0xac, 0x6c, 0xec, 0x1c, 0x9c, 0x5c, 0xdc, + 0x3c, 0xbc, 0x7c, 0xfc, 0x02, 0x82, 0x42, 0xc2, 0x22, 0xa2, 0x62, 0xe2, + 0x12, 0x92, 0x52, 0xd2, 0x32, 0xb2, 0x72, 0xf2, 0x0a, 0x8a, 0x4a, 0xca, + 0x2a, 0xaa, 0x6a, 0xea, 0x1a, 0x9a, 0x5a, 0xda, 0x3a, 0xba, 0x7a, 0xfa, + 0x06, 0x86, 0x46, 0xc6, 0x26, 0xa6, 0x66, 0xe6, 0x16, 0x96, 0x56, 0xd6, + 0x36, 0xb6, 0x76, 0xf6, 0x0e, 0x8e, 0x4e, 0xce, 0x2e, 0xae, 0x6e, 0xee, + 0x1e, 0x9e, 0x5e, 0xde, 0x3e, 0xbe, 0x7e, 0xfe, 0x01, 0x81, 0x41, 0xc1, + 0x21, 0xa1, 0x61, 0xe1, 0x11, 0x91, 0x51, 0xd1, 0x31, 0xb1, 0x71, 0xf1, + 0x09, 0x89, 0x49, 0xc9, 0x29, 0xa9, 0x69, 0xe9, 0x19, 0x99, 0x59, 0xd9, + 0x39, 0xb9, 0x79, 0xf9, 0x05, 0x85, 0x45, 0xc5, 0x25, 0xa5, 0x65, 0xe5, + 0x15, 0x95, 0x55, 0xd5, 0x35, 0xb5, 0x75, 0xf5, 0x0d, 0x8d, 0x4d, 0xcd, + 0x2d, 0xad, 0x6d, 0xed, 0x1d, 0x9d, 0x5d, 0xdd, 0x3d, 0xbd, 0x7d, 0xfd, + 0x03, 0x83, 0x43, 0xc3, 0x23, 0xa3, 0x63, 0xe3, 0x13, 0x93, 0x53, 0xd3, + 0x33, 0xb3, 0x73, 0xf3, 0x0b, 0x8b, 0x4b, 0xcb, 0x2b, 0xab, 0x6b, 0xeb, + 0x1b, 0x9b, 0x5b, 0xdb, 0x3b, 0xbb, 0x7b, 0xfb, 0x07, 0x87, 0x47, 0xc7, + 0x27, 0xa7, 0x67, 0xe7, 0x17, 0x97, 0x57, 0xd7, 0x37, 0xb7, 0x77, 0xf7, + 0x0f, 0x8f, 0x4f, 0xcf, 0x2f, 0xaf, 0x6f, 0xef, 0x1f, 0x9f, 0x5f, 0xdf, + 0x3f, 0xbf, 0x7f, 0xff + }; + const uint128_t mask = zig_bit_mask(uint128_t, zig_type_bit_width); + const uint128_t swapped = zig_byte_swap(value, zig_type_bit_width, signed_type); + const uint128_t reversed = ( + (reverse_lut[(swapped >> 0x78) & 0xff] << 0x78) | + (reverse_lut[(swapped >> 0x70) & 0xff] << 0x70) | + (reverse_lut[(swapped >> 0x68) & 0xff] << 0x68) | + (reverse_lut[(swapped >> 0x60) & 0xff] << 0x60) | + (reverse_lut[(swapped >> 0x58) & 0xff] << 0x58) | + (reverse_lut[(swapped >> 0x50) & 0xff] << 0x50) | + (reverse_lut[(swapped >> 0x48) & 0xff] << 0x48) | + (reverse_lut[(swapped >> 0x40) & 0xff] << 0x40) | + (reverse_lut[(swapped >> 0x38) & 0xff] << 0x38) | + (reverse_lut[(swapped >> 0x30) & 0xff] << 0x30) | + (reverse_lut[(swapped >> 0x28) & 0xff] << 0x28) | + (reverse_lut[(swapped >> 0x20) & 0xff] << 0x20) | + (reverse_lut[(swapped >> 0x18) & 0xff] << 0x18) | + (reverse_lut[(swapped >> 0x10) & 0xff] << 0x10) | + (reverse_lut[(swapped >> 0x08) & 0xff] << 0x08) | + (reverse_lut[(swapped >> 0x00) & 0xff] << 0x00)); + const uint128_t result = + (reversed >> (zig_type_bit_width < 8 ? 8 - zig_type_bit_width : 0)) & mask; + return (signed_type ? zig_sign_extend(result, zig_type_bit_width) : result); +} diff --git a/test/behavior.zig b/test/behavior.zig index 7efe687c75..b353ee78c9 100644 --- a/test/behavior.zig +++ b/test/behavior.zig @@ -139,6 +139,7 @@ test { _ = @import("behavior/eval.zig"); _ = @import("behavior/export_self_referential_type_info.zig"); _ = @import("behavior/int128.zig"); + _ = @import("behavior/popcount.zig"); _ = @import("behavior/translate_c_macros.zig"); _ = @import("behavior/union_with_members.zig"); @@ -147,7 +148,6 @@ test { _ = @import("behavior/atomics.zig"); _ = @import("behavior/export.zig"); _ = @import("behavior/maximum_minimum.zig"); - _ = @import("behavior/popcount.zig"); _ = @import("behavior/saturating_arithmetic.zig"); _ = @import("behavior/widening.zig"); _ = @import("behavior/bugs/2114.zig"); diff --git a/test/behavior/bitreverse.zig b/test/behavior/bitreverse.zig index a355572f84..183e9aac55 100644 --- a/test/behavior/bitreverse.zig +++ b/test/behavior/bitreverse.zig @@ -5,7 +5,6 @@ const minInt = std.math.minInt; test "@bitReverse large exotic integer" { if (builtin.zig_backend == .stage2_wasm) return error.SkipZigTest; - if (builtin.zig_backend == .stage2_c) return error.SkipZigTest; // Currently failing on stage1 for big-endian targets if (builtin.zig_backend == .stage1) return error.SkipZigTest; @@ -14,7 +13,6 @@ test "@bitReverse large exotic integer" { test "@bitReverse" { if (builtin.zig_backend == .stage2_wasm) return error.SkipZigTest; - if (builtin.zig_backend == .stage2_c) return error.SkipZigTest; if (builtin.zig_backend == .stage2_x86_64) return error.SkipZigTest; if (builtin.zig_backend == .stage2_arm) return error.SkipZigTest; if (builtin.zig_backend == .stage2_aarch64) return error.SkipZigTest; diff --git a/test/behavior/byteswap.zig b/test/behavior/byteswap.zig index 265bc40f99..15046cc0e1 100644 --- a/test/behavior/byteswap.zig +++ b/test/behavior/byteswap.zig @@ -4,7 +4,6 @@ const expect = std.testing.expect; test "@byteSwap integers" { if (builtin.zig_backend == .stage2_wasm) return error.SkipZigTest; - if (builtin.zig_backend == .stage2_c) return error.SkipZigTest; if (builtin.zig_backend == .stage2_x86_64) return error.SkipZigTest; if (builtin.zig_backend == .stage2_arm) return error.SkipZigTest; if (builtin.zig_backend == .stage2_aarch64) return error.SkipZigTest; diff --git a/test/behavior/popcount.zig b/test/behavior/popcount.zig index d906c1bcc0..881a784490 100644 --- a/test/behavior/popcount.zig +++ b/test/behavior/popcount.zig @@ -1,3 +1,4 @@ +const builtin = @import("builtin"); const std = @import("std"); const expect = std.testing.expect; const expectEqual = std.testing.expectEqual; @@ -45,6 +46,8 @@ fn testPopCountIntegers() !void { } test "@popCount vectors" { + if (builtin.zig_backend == .stage2_c) return error.SkipZigTest; // TODO + comptime try testPopCountVectors(); try testPopCountVectors(); }