From 34e7282dd7c58b49702c7ecd9f80a46398efd0c1 Mon Sep 17 00:00:00 2001 From: Cliff Click Date: Fri, 7 Feb 2025 08:58:36 -0800 Subject: Ch19 (#155) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit * Initial push Major CodeGen cleanup; most static globals move into the CodeGen object. * Minor progress bug with mutual recursion * Add test from pr#148 * First cut x86 "port" Only "return 0;", no registers, no encoding, but yah gotta start somewhere * Update README.md * Update README.md * Update ListScheduler.java * command line launcher for simple Parser bug fref in args Some new test cases. * Add basic RegMask for more than 64 bits of register mask. Add calling convention basics to X86; con+ret RegMasks. GCM computes CFG for all users. Drop unused MultiUse. Pick up reg-pressure aware ListScheduler. Re-layout CodeGen file to get code & data chunks nearer each other. minor Ary extension * ASM Printer InstSel handles folding 2+ ideal ops into 1 machine op * handle instSel for some control flow Bool,If,CProj,Region,Phi Handle 2-op expansions Shuffle instSel graph walk back again. * Basic if-block inst selects and asm prints * Add mul-by-coinstant to shift * left shift codegen basic * exclude RSP from *write* mask * 2-arg add (not immediate) mul-by-small constant opt. Some utilities. * Handle first loop Change inst sel walk again to pre-order, to set an early visit bit to stop cycles. CFGNode copies dom/loop info. Cmp not-immediate form. Ret/Fun lazy updates * inst select for new struct allocation * Float, bitwise and arithmetic operands codegen x86-64 (#150) * left shift codegen basic * sar codegen basic * merge * some basic ops * addf * divf, subf, mulf * rsp allowed in bitwise input, not output * formatting * left shift codegen basic * sar codegen basic * merge * some basic ops * addf * divf, subf, mulf * rsp allowed in bitwise input, not output * formatting * Remove non-exsting FP+imm ops * Handle first loop Change inst sel walk again to pre-order, to set an early visit bit to stop cycles. CFGNode copies dom/loop info. Cmp not-immediate form. Ret/Fun lazy updates * float ops without imm values * merge * merge * inst select for new struct allocation * left shift codegen basic * sar codegen basic * merge * some basic ops * addf * divf, subf, mulf * rsp allowed in bitwise input, not output * formatting * Remove non-exsting FP+imm ops * left shift codegen basic * sar codegen basic * merge * some basic ops * addf * divf, subf, mulf * rsp allowed in bitwise input, not output * formatting * float ops without imm values * merge * merge * Rebased on ch 19 * Update tests --------- Co-authored-by: Cliff Click * cleanup after merge alpha-sort helper fcns implicit test vs zero/null * Add LEA op asm print works on ideal nodes for all tests * Minor cleanup lea * x86 addressing modes during inst select Drop New taking inits; just follow with initializing stores. Simplifies inst selelection which otherwise needs to undo this optimization and emit following init stores. Basic load/store for now, op-to-mem comes later add same becomes Shl by 1. Drop DivF-immiedate * extended ch13, 14,15,16 (#153) * merge2 * merge3 * Fix a few minor fuzzer bugs * Force array layout ld/st get size (but not signed/unsigned) Add unsigned LT for later range checks. * Fix a bunch of float issues was deleting required inputs * simpler invariant defensive copy inputs for simpler invariant in complex sharing patterns inc/dec form (just the opcode now). * Call & CallR X86 instructions * Add-from-memory op Remove FP immediate forms. * Remove name from TFP Never shoulda been there (but was convenient for awhile). Moved into FunNode, found by checking the linker table with a TFP. * New (#154) * merge2 * merge3 * GCM for float ops fixed, other cleanup - small extensions * call setup * merge2 * remove debug prints add check for bad call convention fix 3 tests (bad merge?) * cleanup call convention abi Needs more love at some point... * one more cleanup ABI --------- Co-authored-by: Cliff Click * Add CmpMem form Common addressing mode print LEA can skip a base * Add a AddFMemX86 A bunch of mem op patterns are missing, hopefully they are just cut-n-paste from the existing patterns. * Docs * XMM reg mask fix Minor README updates * Allow inverted cmp/mem narrowing stores can bypass an AndMask Array length loads do not need control Some missing print info * Support "*ptr op= val" at least for MemAdd * missed golden rule update * Array len is u32 load-after-store zero/sign-extends if the store is truncating * One more missed case --------- Co-authored-by: hadha Co-authored-by: Tamás Hadházy <85063808+Hels15@users.noreply.github.com> --- chapter14/README.md | 270 ++++++++++++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 260 insertions(+), 10 deletions(-) (limited to 'chapter14') diff --git a/chapter14/README.md b/chapter14/README.md index a08499d1..496cbb8c 100644 --- a/chapter14/README.md +++ b/chapter14/README.md @@ -1,11 +1,25 @@ # Chapter 14: Narrow types +# Table of Contents + +1. [Narrow Word Types](#narrow-or-sub-word-types) +2. [OverFlow handling](#overflow-handling) +3. [Integer ranges](#integer-ranges) +4. [Bitwise operations](#examples) +5. [Precedence](#precedence) +6. [Type lattice](#type-implementation) +7. [Nodes](#nodes) + + In this chapter, we add sub-word integer types. You can also read [this chapter](https://github.com/SeaOfNodes/Simple/tree/linear-chapter14) in a linear Git revision history on the [linear](https://github.com/SeaOfNodes/Simple/tree/linear) branch and [compare](https://github.com/SeaOfNodes/Simple/compare/linear-chapter13...linear-chapter14) it to the previous chapter. ## Narrow or sub-word types +Narrow or sub-worded integer types refer to integer data types that occupy less memory (fewer bits) than the standard or "full-width"(Typically 32, or 64 bits) integer types provided by a system or programming language. + + As a prelude to arrays, especially arrays of bytes common in all networking codes, Simple needs some way to manipulate sub-word integer types. Computations are always done in 64 bits for both integer and float ops, but @@ -49,19 +63,242 @@ Similarly, floats can be limited to IEEE754 32-bit standard format and will be rounded when assigned, and inflated to 64-bits when loaded. The complete list of primitive types is now: + +| Data Type | Size (Bits) | Description | +|-----------|-------------|--------------------------------------------------------------| +| `int`, `i64` | 64 | 64-bit signed integer, matching modern hardware. | +| `u64` | 64 | **Not implemented**, hints at behavior beyond 64-bit. | +| `i32` | 32 | 32-bit signed integer, sign-extended on load. | +| `u32` | 32 | 32-bit unsigned integer, zero-extended on load. | +| `i16` | 16 | 16-bit signed integer, sign-extended on load. | +| `u16` | 16 | 16-bit unsigned integer, zero-extended on load. | +| `i8` | 8 | 8-bit signed integer, sign-extended on load. | +| `u8` | 8 | 8-bit unsigned integer, zero-extended on load. | +| `i1` | 1 | 1-bit signed integer, sign-extended on load. | +| `u1` | 1 | 1-bit unsigned integer, zero-extended on load. | +| `bool` | 1 | Alias for `u1`. | +| `flt`, `f64`| 64 | IEEE 754 64-bit floating-point number, often called a double. | +| `f32` | 32 | IEEE 754 32-bit floating-point number. | + +## Overflow handling + +An integer overflow occurs when you attempt to store inside an integer variable a value that is larger than the maximum value the variable can hold. +There are several cases of overflow: + +#### Unsigned(fixed-width) integer overflow: + +```java +u8 a = 255; +u8 b = a + 1; // match u8 type against expr.type +// a = 256 +// u8 = [0...255] +return b; +``` +The expression type does not match the type that was provided. expr: ConstantNode(256) t: u8(0, 255) +```java +private Node parseExpressionStatement() { + Type t = type(); // u8 = [0...255] + ... + // expr.type = TypeInteger(256); + // Auto-narrow wide ints to narrow ints + expr = zsMask(expr,t); +} +``` +Since it is unsigned we just do a bitmask on the new value that overflows the specified type: +The operands to the "bitwise AND" are the values(256) and the mask(255, max boundary). +This is called truncating. + +`ZsMask:` +```java + // zero/sign extend. "i" is limited to either classic unsigned (min==0) or + // classic signed (min=minus-power-of-2); max=power-of-2-minus-1. +private Node zsMask(Node val, Type t ) { + if( !(val._type instanceof TypeInteger tval && t instanceof TypeInteger t0 && !tval.isa(t0)) ) { + if( !(val._type instanceof TypeFloat tval && t instanceof TypeFloat t0 && !tval.isa(t0)) ) + return val; + // Float rounding + return new RoundF32Node(val).peephole(); + } + if( t0._min==0 ) // Unsigned + return new AndNode(val,new ConstantNode(TypeInteger.constant(t0._max)).peephole()).peephole(); + // Signed extension + int shift = Long.numberOfLeadingZeros(t0._max)-1; + Node shf = new ConstantNode(TypeInteger.constant(shift)).peephole(); + if( shf._type==TypeInteger.ZERO ) + return val; + return new SarNode(new ShlNode(val,shf.keep()).peephole(),shf.unkeep()).peephole(); +} +``` + +The truncation is done by the `AndNode`: +```java +... +if( t0._min==0 ) // Unsigned + return new AndNode(val,con(t0._max)).peephole(); +``` +Since they are constants, after calling peephole it truns into`256 & 255` = 0; +Output: +``` +return 0; +``` + +#### Maximum integer overflow +```java +i64 a = 9223372036854775807; +i64 b = a + 1; +return b; +``` +When we are computing the addition, we make sure they don't overflow: + +`AddNode.overflow:` +```java +private static boolean overflow( long x, long y ) { + if( (x ^ y ) < 0 ) return false; // unequal signs, never overflow + return (x ^ (x + y)) < 0; // sum has unequal signs, so overflow +} +``` +`AddNode.compute` +```java +... +// Fold ranges like {0-1} + {2-3} into {2-4}. +if( !overflow(i1._min,i2._min) && + !overflow(i1._max,i2._max) ) + return TypeInteger.make(i1._min+i2._min,i1._max+i2._max); +``` +Here **x + y** overflows and becomes **negative**, x ^ negative = negative. Any negative number is less than 0, so it returns true. +If it **overflows** we just return the bottom type of TypeInteger. +``` + return TypeInteger.BOT; // [] +``` +However, in our case we are dealing with constants, so we just allow the overflow naturally here: ```java -int, i64 // 64-bits, matching modern hardware -// u64 // NOT implemented, as it hints at some kind of >64 bit behavior -i32, u32 // 32 bits, sign or zero extended on load -i16, u16 // 16 bits, sign or zero extended on load - i8, u8 // 8 bits, sign or zero extended on load - i1, u1 // 1 bit , sign or zero extended on load -bool // Same as u1 -flt, f64 // IEEE 754 64-bit float, sometimes called a "double" -f32 // IEEE 754 32-bit float +// i1: 9223372036854775807 +// i2: 1 +// i1 + i2 will overflow. +return TypeInteger.constant(i1.value()+i2.value()); // -9223372036854775808; ``` +#### Adding constant ranges +```java +u8 a = 12; +u8 b = 13; +u8 c = 124; +u8 d = a + b + c; +return d; +``` +if `is_con` is true, then generally we can state: `min = max`. +Such that: +```java +public static TypeInteger make(boolean is_con, long con) { + return make(is_con ? con : (con==0 ? Long.MAX_VALUE : Long.MIN_VALUE), + is_con ? con : (con==0 ? Long.MIN_VALUE : Long.MAX_VALUE)); +``` +- This results in: +```java +return make(con, con); +``` +`value()` is only called in constants, e.g `_min == _max;` invariants holds, then since +both `min_` and `max_` are equal, we can return either one. + +This is the same as regular additions: +```java +public long value() { assert isConstant(); return _min; } +``` + +#### Truncate(2) + +```java +u8 a = arg; // arg & 255 +u8 b = arg; // arg & 255(same as a) +u8 c = a + b; // (arg&255)*2, type = TypeInteger.BOT +return c; // (((arg&255)*2)&255), type = TypeInteger.U8 +``` + +Same as before, **arg** turns into **(arg&255)**, this node is then inserted into the **GVN** table and **b** retrieves this value. +The AddNode peephole will change them into `((arg&255)*2`, and put on extra mask +on the expression to keep it in the valid boundaries. + + +## Integer ranges +Link hacker's delight here. + +For range math, visit:[^1] +## Bitwise operations +We support the following bitwise operators: +And `&`, Or `|`, Xor `^`, Shift left `<<`, Shift right `>>`, Shift right zero `>>>` + +| Operator | Symbol | Description | +|--------------------------|---------|---------------------------------------------------------------------------------------------------------| +| **And** | `&` | Performs a bitwise AND operation. | +| **Or** | `\|` | Performs a bitwise OR (inclusive OR) operation. | +| **Xor** | `^` | Performs a bitwise XOR (exclusive OR) operation. | +| **Shift Left** | `<<` | Shifts bits to the left, filling with zeros. | +| **Shift Right** | `>>` | Shifts bits to the right, any vacated bit postions are filled by replicating the value of the sign bit. | +| **Unsigned Right Shift** | `>>>` | Shifts bits to the right, any vacated bit postions are always filled with zero | + +Note `>>>` is a logical shift and not an arithmetic shift. + +#### AndNode: + +##### lhs & -1 = lhs; +```java +// And of -1. We do not check for (-1&x) because this will already +// canonicalize to (x&-1) +if( t2.isConstant() && t2 instanceof TypeInteger i && i.value()==-1 ) +return lhs; + +``` + +#### OrNode: +##### lhs | 0 = lhs; +```java +// Or of 0. We do not check for (0|x) because this will already +// canonicalize to (x|0) +if( t2.isConstant() && t2 instanceof TypeInteger i && i.value()==0 ) +return lhs; + +``` + +#### Shift right +lhs >> 0 = lhs; +```java +// Sar of 0. +if( t2.isConstant() && t2 instanceof TypeInteger i && (i.value()&63)==0 ) + return lhs; +``` + +#### Shift left +lhs << 0 = lhs; +```java +// Shl of 0. +if( t2.isConstant() && t2 instanceof TypeInteger i && (i.value()&63)==0 ) + return lhs; + +``` +#### Unsigned right shift +lhs >>> 0 = lhs; +```java +// Shr of 0. +if( t2.isConstant() && t2 instanceof TypeInteger i && (i.value()&63)==0 ) +return lhs; +``` + +#### XorNode: +##### lhs ^ 0 = lhs; +```java +// Xor of 0. We do not check for (0^x) because this will already +// canonicalize to (x^0) +if( t2.isConstant() && t2 instanceof TypeInteger i && i.value()==0 ) + return lhs; + +``` +### precedence + +Bitwise operators have the lowest precedence, e.g +``` +arg + 123 & 3 = (arg + 123) & 3 +``` ## Type Implementation @@ -85,8 +322,18 @@ The TypeFloat class is also reworked ![Graph1](./docs/lattice.svg) +### meet: - +Meet of two ranges: +```java +@Override +public Type xmeet(Type other) { +// Invariant from caller: 'this' != 'other' and same class (TypeInteger) +TypeInteger i = (TypeInteger)other; // Contract +return make(Math.min(_min,i._min), Math.max(_max,i._max)); +} +``` +Combine all values that could possibly occur within either range. ## Nodes @@ -98,3 +345,6 @@ slightly different `idealize()` calls. +[^1]: Hacker's delight. + 4-2 Propagating Bounds through Add's and Subtract's + -- cgit v1.2.3