From b93ae1a01bbad59ee77b7c84f4743a730138ba87 Mon Sep 17 00:00:00 2001 From: Iain Buclaw Date: Tue, 28 Jun 2022 19:22:34 +0200 Subject: d: Add SIMD intrinsics module and compiler built-ins. Vectors in D are exposed by the use of the `__vector(T[N])' type, and whilst most unary and binary operations work as you'd expect, there are some operations that are not possible without doing the operation unrolled, or calling some target-specific built-in, or with inline asm. This introduces a new `gcc.simd' module that introduces the following. - Prefetching has been exposed by a convenient `prefetch' function in the library. - Loading and storing from an unaligned address have been exposed by `loadUnaligned' and `storeUnaligned' intrinsics. - Vector permutations have been exposed by `shuffle`, and `shufflevector' intrinsics. - Converting between two vectors with a different element type has been exposed by a `convertvector' intrinsic. - The ternary operator has been exposed with a `blendvector' intrinsic. - Comparison operators have been exposed by `equalMask', `notEqualMask', `greaterMask', and `greaterEqualMask' intrinsics. - Logic operators have been exposed by convenient `notMask', `andAndMask', and `orOrMask' functions in the library. To be compatible with the LLVM D compiler's own SIMD intrinsic module, there is also the addition of an `extractelement' and `insertelement' convenience functions, and an alternative interface for calling the `shufflevector' function. The addition of these intrinsics lowers the boundary for users working in SIMD to get the desired codegen they want out of the compiler. Most of what is present here - apart from tests - is the adding of machinery in the intrinsics suite of functions to do validation on templated intrinsics. Whilst these are still matched from the library by their generic (untyped) signature, there is a still an assumption that what has been instantiated and handed down to the code generator is valid, because why would these definitions be found outside of the in-tree D runtime library? The majority of intrinsics are not templates, so the test on the mangled signature string still guarantees all types are as we expect them to be. However there are still a small handful of other templated intrinsics (core.bitop.{rol,ror}, core.math.toPrec, std.math.traits.isNaN, ...) that are currently unchecked, so would benefit from being included into this built-in checking function at some point in the future. gcc/d/ChangeLog: * intrinsics.cc: Include diagnostic.h, langhooks.h, vec-perm-indices.h. (maybe_set_intrinsic): Add cases for new simd intrinsics. (warn_mismatched_return_type): New function. (warn_mismatched_argument): New function. (build_shuffle_mask_type): New function. (maybe_warn_intrinsic_mismatch): New function. (expand_intrinsic_vec_cond): New function. (expand_intrinsic_vec_convert): New function. (expand_intrinsic_vec_blend): New function. (expand_intrinsic_vec_shuffle): New function. (expand_intrinsic_vec_shufflevector): New function. (expand_intrinsic_vec_load_unaligned): New function. (expand_intrinsic_vec_store_unaligned): New function. (maybe_expand_intrinsic): Check signature of intrinsic before handing off to front-end lowering. Add cases for new simd intrinsics. * intrinsics.def (INTRINSIC_LOADUNALIGNED): Define intrinsic. (INTRINSIC_STOREUNALIGNED): Define intrinsic. (INTRINSIC_SHUFFLE): Define intrinsic. (INTRINSIC_SHUFFLEVECTOR): Define intrinsic. (INTRINSIC_CONVERTVECTOR): Define intrinsic. (INTRINSIC_BLENDVECTOR): Define intrinsic. (INTRINSIC_EQUALMASK): Define intrinsic. (INTRINSIC_NOTEQUALMASK): Define intrinsic. (INTRINSIC_GREATERMASK): Define intrinsic. (INTRINSIC_GREATEREQUALMASK): Define intrinsic. libphobos/ChangeLog: * libdruntime/Makefile.am (DRUNTIME_DSOURCES): Add gcc/simd.d. * libdruntime/Makefile.in: Regenerate. * libdruntime/gcc/simd.d: New file. gcc/testsuite/ChangeLog: * gdc.dg/Wbuiltin_declaration_mismatch.d: Rename to... * gdc.dg/Wbuiltin_declaration_mismatch1.d: ...this. * gdc.dg/Wbuiltin_declaration_mismatch2.d: New test. * gdc.dg/torture/simd_blendvector.d: New test. * gdc.dg/torture/simd_cond.d: New test. * gdc.dg/torture/simd_convertvector.d: New test. * gdc.dg/torture/simd_load.d: New test. * gdc.dg/torture/simd_logical.d: New test. * gdc.dg/torture/simd_shuffle.d: New test. * gdc.dg/torture/simd_shufflevector.d: New test. * gdc.dg/torture/simd_store.d: New test. --- libphobos/libdruntime/Makefile.am | 16 +- libphobos/libdruntime/Makefile.in | 19 +- libphobos/libdruntime/gcc/simd.d | 359 ++++++++++++++++++++++++++++++++++++++ 3 files changed, 377 insertions(+), 17 deletions(-) create mode 100644 libphobos/libdruntime/gcc/simd.d (limited to 'libphobos') diff --git a/libphobos/libdruntime/Makefile.am b/libphobos/libdruntime/Makefile.am index d963aa9..56b332d 100644 --- a/libphobos/libdruntime/Makefile.am +++ b/libphobos/libdruntime/Makefile.am @@ -207,14 +207,14 @@ DRUNTIME_DSOURCES = core/atomic.d core/attribute.d core/bitop.d \ core/vararg.d core/volatile.d gcc/attribute.d gcc/attributes.d \ gcc/backtrace.d gcc/builtins.d gcc/deh.d gcc/emutls.d gcc/gthread.d \ gcc/sections/common.d gcc/sections/elf.d gcc/sections/macho.d \ - gcc/sections/package.d gcc/sections/pecoff.d gcc/unwind/arm.d \ - gcc/unwind/arm_common.d gcc/unwind/c6x.d gcc/unwind/generic.d \ - gcc/unwind/package.d gcc/unwind/pe.d object.d rt/aApply.d rt/aApplyR.d \ - rt/aaA.d rt/adi.d rt/arrayassign.d rt/arraycat.d rt/cast_.d \ - rt/config.d rt/critical_.d rt/deh.d rt/dmain2.d rt/ehalloc.d \ - rt/invariant.d rt/lifetime.d rt/memory.d rt/minfo.d rt/monitor_.d \ - rt/profilegc.d rt/sections.d rt/tlsgc.d rt/util/typeinfo.d \ - rt/util/utility.d + gcc/sections/package.d gcc/sections/pecoff.d gcc/simd.d \ + gcc/unwind/arm.d gcc/unwind/arm_common.d gcc/unwind/c6x.d \ + gcc/unwind/generic.d gcc/unwind/package.d gcc/unwind/pe.d object.d \ + rt/aApply.d rt/aApplyR.d rt/aaA.d rt/adi.d rt/arrayassign.d \ + rt/arraycat.d rt/cast_.d rt/config.d rt/critical_.d rt/deh.d \ + rt/dmain2.d rt/ehalloc.d rt/invariant.d rt/lifetime.d rt/memory.d \ + rt/minfo.d rt/monitor_.d rt/profilegc.d rt/sections.d rt/tlsgc.d \ + rt/util/typeinfo.d rt/util/utility.d DRUNTIME_DSOURCES_STDCXX = core/stdcpp/allocator.d core/stdcpp/array.d \ core/stdcpp/exception.d core/stdcpp/memory.d core/stdcpp/new_.d \ diff --git a/libphobos/libdruntime/Makefile.in b/libphobos/libdruntime/Makefile.in index 45e086a..24865fb 100644 --- a/libphobos/libdruntime/Makefile.in +++ b/libphobos/libdruntime/Makefile.in @@ -236,7 +236,7 @@ am__objects_1 = core/atomic.lo core/attribute.lo core/bitop.lo \ gcc/backtrace.lo gcc/builtins.lo gcc/deh.lo gcc/emutls.lo \ gcc/gthread.lo gcc/sections/common.lo gcc/sections/elf.lo \ gcc/sections/macho.lo gcc/sections/package.lo \ - gcc/sections/pecoff.lo gcc/unwind/arm.lo \ + gcc/sections/pecoff.lo gcc/simd.lo gcc/unwind/arm.lo \ gcc/unwind/arm_common.lo gcc/unwind/c6x.lo \ gcc/unwind/generic.lo gcc/unwind/package.lo gcc/unwind/pe.lo \ object.lo rt/aApply.lo rt/aApplyR.lo rt/aaA.lo rt/adi.lo \ @@ -874,14 +874,14 @@ DRUNTIME_DSOURCES = core/atomic.d core/attribute.d core/bitop.d \ core/vararg.d core/volatile.d gcc/attribute.d gcc/attributes.d \ gcc/backtrace.d gcc/builtins.d gcc/deh.d gcc/emutls.d gcc/gthread.d \ gcc/sections/common.d gcc/sections/elf.d gcc/sections/macho.d \ - gcc/sections/package.d gcc/sections/pecoff.d gcc/unwind/arm.d \ - gcc/unwind/arm_common.d gcc/unwind/c6x.d gcc/unwind/generic.d \ - gcc/unwind/package.d gcc/unwind/pe.d object.d rt/aApply.d rt/aApplyR.d \ - rt/aaA.d rt/adi.d rt/arrayassign.d rt/arraycat.d rt/cast_.d \ - rt/config.d rt/critical_.d rt/deh.d rt/dmain2.d rt/ehalloc.d \ - rt/invariant.d rt/lifetime.d rt/memory.d rt/minfo.d rt/monitor_.d \ - rt/profilegc.d rt/sections.d rt/tlsgc.d rt/util/typeinfo.d \ - rt/util/utility.d + gcc/sections/package.d gcc/sections/pecoff.d gcc/simd.d \ + gcc/unwind/arm.d gcc/unwind/arm_common.d gcc/unwind/c6x.d \ + gcc/unwind/generic.d gcc/unwind/package.d gcc/unwind/pe.d object.d \ + rt/aApply.d rt/aApplyR.d rt/aaA.d rt/adi.d rt/arrayassign.d \ + rt/arraycat.d rt/cast_.d rt/config.d rt/critical_.d rt/deh.d \ + rt/dmain2.d rt/ehalloc.d rt/invariant.d rt/lifetime.d rt/memory.d \ + rt/minfo.d rt/monitor_.d rt/profilegc.d rt/sections.d rt/tlsgc.d \ + rt/util/typeinfo.d rt/util/utility.d DRUNTIME_DSOURCES_STDCXX = core/stdcpp/allocator.d core/stdcpp/array.d \ core/stdcpp/exception.d core/stdcpp/memory.d core/stdcpp/new_.d \ @@ -1340,6 +1340,7 @@ gcc/sections/elf.lo: gcc/sections/$(am__dirstamp) gcc/sections/macho.lo: gcc/sections/$(am__dirstamp) gcc/sections/package.lo: gcc/sections/$(am__dirstamp) gcc/sections/pecoff.lo: gcc/sections/$(am__dirstamp) +gcc/simd.lo: gcc/$(am__dirstamp) gcc/unwind/$(am__dirstamp): @$(MKDIR_P) gcc/unwind @: > gcc/unwind/$(am__dirstamp) diff --git a/libphobos/libdruntime/gcc/simd.d b/libphobos/libdruntime/gcc/simd.d new file mode 100644 index 0000000..ffca50f --- /dev/null +++ b/libphobos/libdruntime/gcc/simd.d @@ -0,0 +1,359 @@ +// GNU D Compiler SIMD support functions and intrinsics. +// Copyright (C) 2022 Free Software Foundation, Inc. + +// GCC is free software; you can redistribute it and/or modify it under +// the terms of the GNU General Public License as published by the Free +// Software Foundation; either version 3, or (at your option) any later +// version. + +// GCC is distributed in the hope that it will be useful, but WITHOUT ANY +// WARRANTY; without even the implied warranty of MERCHANTABILITY or +// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +// for more details. + +// Under Section 7 of GPL version 3, you are granted additional +// permissions described in the GCC Runtime Library Exception, version +// 3.1, as published by the Free Software Foundation. + +// You should have received a copy of the GNU General Public License and +// a copy of the GCC Runtime Library Exception along with this program; +// see the files COPYING3 and COPYING.RUNTIME respectively. If not, see +// . + +module gcc.simd; + +pure: +nothrow: +@safe: +@nogc: +pragma(inline, true): + +/** +* Emit prefetch instruction. +* Params: +* address = address to be prefetched +* writeFetch = true for write fetch, false for read fetch +* locality = 0..3 (0 meaning least local, 3 meaning most local) +*/ +void prefetch(bool writeFetch, ubyte locality)(const(void)* address) +{ + static assert(locality < 4, "0..3 expected for locality"); + import gcc.builtins : __builtin_prefetch; + __builtin_prefetch(address, writeFetch, locality); +} + +/** + * Load unaligned vector from address. + * This is a compiler intrinsic. + * Params: + * p = pointer to vector + * Returns: + * vector + */ +V loadUnaligned(V)(const V* p) if (isVectorType!V); + +/** + * Store vector to unaligned address. + * This is a compiler intrinsic. + * Params: + * p = pointer to vector + * value = value to store + * Returns: + * value + */ +V storeUnaligned(V)(V* p, V value) if (isVectorType!V); + +/** + * Construct a permutation of elements from one or two vectors, returning a + * vector of the same type as the input vector(s). The `mask` is an integral + * vector with the same width and element count as the output vector. + * Params: + * op1 = input vector + * op2 = input vector + * mask = integer vector mask + * Returns: + * vector with the same type as `op1` and `op2` + * Example: + * --- + * int4 a = [1, 2, 3, 4]; + * int4 b = [5, 6, 7, 8]; + * int4 mask1 = [0, 1, 1, 3]; + * int4 mask2 = [0, 4, 2, 5]; + * assert(shuffle(a, mask1).array == [1, 2, 2, 4]); + * assert(shuffle(a, b, mask2).array == [1, 5, 3, 6]); + * --- + */ +template shuffle(V0, V1, M) +{ + static assert(isVectorType!V0, "first argument must be vector"); + static assert(isVectorType!V1, "second argument must be vector"); + static assert(is(BaseType!V0 == BaseType!V1), + "first and second argument vectors must have the same element type"); + static assert(isVectorType!M && is(BaseType!M : long), + "last argument must be an integer vector"); + static assert(numElements!V0 == numElements!M && numElements!V1 == numElements!M, + "argument vectors and mask vector should have the same number of elements"); + static assert(BaseType!V0.sizeof == BaseType!M.sizeof, + "argument vectors and mask vector should have the same element type size"); + + V0 shuffle(V0 op1, V1 op2, M mask); +} + +/// Ditto +template shuffle(V, M) +{ + static assert(isVectorType!V, "first argument must be a vector"); + static assert(isVectorType!M && is(BaseType!M : long), + "last argument must be an integer vector"); + static assert(numElements!V == numElements!M, + "argument vector and mask vector should have the same number of elements"); + static assert(BaseType!V.sizeof == BaseType!M.sizeof, + "argument vector and mask vector should have the same element type size"); + + V shuffle(V op1, M mask) + { + return shuffle(op1, op1, mask); + } +} + +/** + * Construct a permutation of elements from two vectors, returning a vector with + * the same element type as the input vector(s), and same length as the `mask`. + * Params: + * op1 = input vector + * op2 = input vector + * index = elements indices of the vectors that should be extracted and returned + * Returns: + * vector with the same element type as `op1` and `op2`, but has an element count + * equal to the number of indices in `index`. + * Example: + * --- + * int8 a = [1, -2, 3, -4, 5, -6, 7, -8]; + * int4 b = shufflevector(a, a, 0, 2, 4, 6); + * assert(b.array == [1, 3, 5, 7]); + * int4 c = [-2, -4, -6, -8]; + * int d = shufflevector(c, b, 4, 0, 5, 1, 6, 2, 7, 3); + * assert(d.array == a.array); + * --- + */ +template shufflevector(V1, V2, M...) +{ + static assert(isVectorType!V1, "first argument must be vector"); + static assert(isVectorType!V2, "second argument must be vector"); + static assert(is(BaseType!V1 == BaseType!V2), + "first and second argument vectors must have the same element type"); + static assert(isPowerOf2!(M.length), + "number of index arguments must be a power of 2"); + + __vector(BaseType!V1[M.length]) shufflevector(V1 op1, V2 op2, M index); +} + +/// Ditto +template shufflevector(V, index...) +{ + // Defined for compatibility with LDC. + static assert(isVectorType!V, "first argument must be a vector type"); + static assert(numElements!V == index.length, + "number of index arguments must be the same number of vector elements"); + + private template ctfeConstants(m...) + { + static if (m.length == 0) enum ctfeConstants = 1; + else enum ctfeConstants = m[0] | ctfeConstants!(m[1 .. $]); + } + static assert(__traits(compiles, ctfeConstants!index), + "all index arguments must be compile time constants"); + + private template validIndexes(m...) + { + static if (m.length == 0) enum validIndexes = true; + else enum validIndexes = (cast(long)m[0] > -1) && validIndexes!(m[1 .. $]); + } + static assert(validIndexes!index, + "all index arguments must be greater than or equal to 0"); + + V shufflevector(V op1, V op2) + { + return shufflevector(op1, op2, index); + } +} + +/** + * Extracts a single scalar element from a vector at a specified index. + * Defined for compatibility with LDC. + * Params: + * val = vector to extract element from + * idx = index indicating the position from which to extract the element + * Returns: + * scalar of the same type as the element type of val + * Example: + * --- + * int4 a = [0, 10, 20, 30]; + * int k = extractelement!(int4, 2)(a); + * assert(k == 20); + * --- + */ +BaseType!V extractelement(V, int idx)(V val) + if (isVectorType!V && idx < numElements!V) +{ + return val[idx]; +} + +/** + * Inserts a scalar element into a vector at a specified index. + * Defined for compatibility with LDC. + * Params: + * val = vector to assign element to + * elt = scalar whose type is the element type of val + * idx = index indicating the position from which to extract the element + * Returns: + * vector of the same type as val + * Example: + * --- + * int4 a = [0, 10, 20, 30]; + * int4 b = insertelement!(int4, 2)(a, 50); + * assert(b.array == [0, 10, 50, 30]); + * --- + */ +V insertelement(V, int idx)(V val, BaseType!V elt) + if (isVectorType!V && idx < numElements!V) +{ + val[idx] = elt; + return val; +} + +/** + * Convert a vector from one integral or floating vector type to another. + * The result is an integral or floating vector that has had every element + * cast to the element type of the return type. + * Params: + * from = input vector + * Returns: + * converted vector + * Example: + * --- + * int4 a = [1, -2, 3, -4]; + * float4 b = [1.5, -2.5, 3, 7]; + * assert(convertvector!float4(a).array == [1, -2, 3, -4]); + * assert(convertvector!double4(a).array == [1, -2, 3, -4]); + * assert(convertvector!double4(b).array == [1.5, -2.5, 3, 7]); + * assert(convertvector!int4(b).array == [1, -2, 3, 7]); + * --- + */ + +template convertvector(V, T) +{ + static assert(isVectorType!V && (is(BaseType!V : long) || is(BaseType!V : real)), + "first argument must be an integer or floating vector type"); + static assert(isVectorType!T && (is(BaseType!T : long) || is(BaseType!T : real)), + "second argument must be an integer or floating vector"); + static assert(numElements!V == numElements!T, + "first and second argument vectors should have the same number of elements"); + + V convertvector(T); +} + +/** + * Construct a conditional merge of elements from two vectors, returning a + * vector of the same type as the input vector(s). The `mask` is an integral + * vector with the same width and element count as the output vector. + * Params: + * op1 = input vector + * op2 = input vector + * mask = integer vector mask + * Returns: + * vector with the same type as `op1` and `op2` + * Example: + * --- + * int4 a = [1, 2, 3, 4]; + * int4 b = [5, 6, 7, 8]; + * int4 mask1 = [0, 1, 1, 3]; + * int4 mask2 = [0, 4, 2, 5]; + * assert(shuffle(a, mask1).array == [1, 2, 2, 4]); + * assert(shuffle(a, b, mask2).array == [1, 5, 3, 6]); + * --- + */ +template blendvector(V0, V1, M) +{ + static assert(isVectorType!V0, "first argument must be vector"); + static assert(isVectorType!V1, "second argument must be vector"); + static assert(is(BaseType!V0 == BaseType!V1), + "first and second argument vectors must have the same element type"); + static assert(isVectorType!M && is(BaseType!M : long), + "last argument must be an integer vector"); + static assert(numElements!V0 == numElements!M && numElements!V1 == numElements!M, + "argument vectors and mask vector should have the same number of elements"); + static assert(BaseType!V0.sizeof == BaseType!M.sizeof, + "argument vectors and mask vector should have the same element type size"); + + V0 blendvector(V0 op1, V1 op2, M mask); +} + +/** + * Perform an element-wise comparison between two vectors, producing `0` when + * the comparison is false and `-1` (all bits are set to 1) otherwise. + * Params: + * op1 = input vector + * op2 = input vector + * Returns: + * vector of the same width and number of elements as the comparison + * operands with a signed integral element type + * Example: + * --- + * float4 a = [1, 3, 5, 7]; + * float4 b = [2, 3, 4, 5]; + * int4 c = greaterMask!float4(a, b); + * assert(c.array == [0, 0, -1, -1]); + * --- + */ +V equalMask(V)(V op1, V op2) if (isVectorType!V); +/// Ditto +V notEqualMask(V)(V op1, V op2) if (isVectorType!V); +/// Ditto +V greaterMask(V)(V op1, V op2) if (isVectorType!V); +/// Ditto +V greaterOrEqualMask(V)(V op1, V op2) if (isVectorType!V); + +/** + * Perform an element-wise logical comparison between two vectors, producing + * `0` when the comparison is false and `-1` (all bits are set to 1) otherwise. + * Params: + * op1 = input vector + * op2 = input vector + * Returns: + * vector of the same width and number of elements as the comparison + * operands with a signed integral element type + */ +V notMask(V)(V op1) if (isVectorType!V) +{ + return equalMask(op1, 0); +} + +/// Ditto +V andAndMask(V)(V op1, V op2) if (isVectorType!V) +{ + return notEqualMask(op1, 0) & notEqualMask(op2, 0); +} + +/// Ditto +V orOrMask(V)(V op1, V op2) if (isVectorType!V) +{ + return notEqualMask(op1, 0) | notEqualMask(op2, 0); +} + +// Private helper templates. +private: + +enum bool isVectorType(T) = is(T : __vector(V[N]), V, size_t N); + +template BaseType(V) +{ + alias typeof(V.array[0]) BaseType; +} + +template numElements(V) +{ + enum numElements = V.sizeof / BaseType!(V).sizeof; +} + +enum bool isPowerOf2(int Y) = Y && (Y & -Y) == Y; -- cgit v1.1