Add vscale_range attribute to functions with sve #58433
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
giordano
reviewed
May 16, 2025
| FS.split(Features, ','); | ||
| for (StringRef Feature : Features) { | ||
| if (Feature == "sve") | ||
| F.addFnAttr(llvm::Attribute::getWithVScaleRangeArgs(M.getContext(), 1, 16)); //Hardcode for now |
There was a problem hiding this comment.
This is relevant for risc-v, too. I've got access to a chip with scalable vector registers (RVV extension), now, if only we could compile julia for that architecture: #57569 😇
There was a problem hiding this comment.
Maybe there's a way to check if the target supports vscale? Is https://github.com/llvm/llvm-project/blob/689a960b693446f0d258240fea0c844b6e685d6e/clang/lib/CodeGen/CodeGenFunction.cpp#L1124-L1130 relevant?
There was a problem hiding this comment.
Don't look at how that function is implemented, you will find something similar :)
julia> function axpy!(a, x, y)
@simd for idx in eachindex(x, y)
@inbounds y[idx] = muladd(a, x[idx], y[idx])
end
end
axpy! (generic function with 1 method)
julia> open("axpy.ll", "w") do file code_llvm(file, axpy!, (Float64, Vector{Float64}, Vector{Float64}); raw=true, dump_module=true) end; ModuleID = 'axpy!'
source_filename = "axpy!"
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128-Fn32"
target triple = "aarch64-unknown-linux-gnu"
@jl_nothing = external local_unnamed_addr constant ptr
@"jl_global#3608.jit" = private alias ptr, inttoptr (i64 281472983889952 to ptr)
; Function Signature: axpy!(Float64, Array{Float64, 1}, Array{Float64, 1})
; @ REPL[7]:1 within `axpy!`
define swiftcc void @"julia_axpy!_3603"(ptr nonnull swiftself %pgcstack_arg, double %"a::Float64", ptr noundef nonnull align 8 dereferenceable(24) %"x::Array", ptr noundef nonnull align 8 dereferenceable(24) %"y::Array") local_unnamed_addr #0 !dbg !4 {
top:
%"new::OneTo" = alloca [1 x i64], align 8
%"new::OneTo1" = alloca [1 x i64], align 8
#dbg_value(double %"a::Float64", !19, !DIExpression(), !22)
#dbg_declare(ptr %"x::Array", !20, !DIExpression(), !22)
#dbg_declare(ptr %"y::Array", !21, !DIExpression(), !22)
%ptls_field = getelementptr inbounds nuw i8, ptr %pgcstack_arg, i64 16
%ptls_load = load ptr, ptr %ptls_field, align 8, !tbaa !23
%0 = getelementptr inbounds nuw i8, ptr %ptls_load, i64 16
%safepoint = load ptr, ptr %0, align 8, !tbaa !27, !invariant.load !9
fence syncscope("singlethread") seq_cst
%1 = load volatile i64, ptr %safepoint, align 8, !dbg !22
fence syncscope("singlethread") seq_cst
; @ REPL[7]:2 within `axpy!`
; ┌ @ simdloop.jl:69 within `macro expansion`
; │┌ @ abstractarray.jl:384 within `eachindex` @ abstractarray.jl:394 @ abstractarray.jl:391
; ││┌ @ abstractarray.jl:137 within `axes1`
; │││┌ @ abstractarray.jl:98 within `axes`
; ││││┌ @ array.jl:194 within `size`
%"x::Array.size_ptr" = getelementptr inbounds nuw i8, ptr %"x::Array", i64 16, !dbg !29
%"x::Array.size.0.copyload" = load i64, ptr %"x::Array.size_ptr", align 8, !dbg !29, !tbaa !46, !alias.scope !47, !noalias !51
; ││││└
; ││││┌ @ tuple.jl:358 within `map`
; │││││┌ @ range.jl:486 within `unchecked_oneto`
store i64 %"x::Array.size.0.copyload", ptr %"new::OneTo", align 8, !dbg !55, !tbaa !61, !alias.scope !63, !noalias !64
; ││└└└└
; ││ @ abstractarray.jl:384 within `eachindex` @ abstractarray.jl:395
; ││┌ @ tuple.jl:358 within `map`
; │││┌ @ abstractarray.jl:395 within `#eachindex##0`
; ││││┌ @ abstractarray.jl:391 within `eachindex`
; │││││┌ @ abstractarray.jl:137 within `axes1`
; ││││││┌ @ abstractarray.jl:98 within `axes`
; │││││││┌ @ array.jl:194 within `size`
%"y::Array.size_ptr" = getelementptr inbounds nuw i8, ptr %"y::Array", i64 16, !dbg !65
%"y::Array.size.0.copyload" = load i64, ptr %"y::Array.size_ptr", align 8, !dbg !65, !tbaa !46, !alias.scope !47, !noalias !51
; │││││││└
; │││││││┌ @ tuple.jl:358 within `map`
; ││││││││┌ @ range.jl:486 within `unchecked_oneto`
store i64 %"y::Array.size.0.copyload", ptr %"new::OneTo1", align 8, !dbg !73, !tbaa !61, !alias.scope !63, !noalias !64
; ││└└└└└└└
; ││ @ abstractarray.jl:384 within `eachindex` @ abstractarray.jl:396
; ││┌ @ anyall.jl:240 within `all`
; │││┌ @ anyall.jl:211 within `_all`
; ││││┌ @ operators.jl:1195 within `Fix`
; │││││┌ @ operators.jl:1195 within `#_#59`
; ││││││┌ @ range.jl:1145 within `==` @ promotion.jl:632
%.not.not = icmp eq i64 %"y::Array.size.0.copyload", %"x::Array.size.0.copyload", !dbg !75
; ││└└└└└
br i1 %.not.not, label %L24, label %L21, !dbg !90
L21: ; preds = %top
call swiftcc void @j_throw_eachindex_mismatch_indices_3607(ptr nonnull swiftself %pgcstack_arg, ptr nonnull @"jl_global#3608.jit", ptr nocapture nonnull readonly %"new::OneTo", ptr nocapture nonnull readonly %"new::OneTo1") #7, !dbg !90
unreachable, !dbg !90
L24: ; preds = %top
; │└
; │ @ simdloop.jl:72 within `macro expansion`
; │┌ @ int.jl:83 within `<`
%2 = icmp slt i64 %"x::Array.size.0.copyload", 1, !dbg !91
; │└
br i1 %2, label %L103, label %iter.check, !dbg !94
iter.check: ; preds = %L24
%memoryref_data = load ptr, ptr %"x::Array", align 8, !tbaa !95, !alias.scope !98, !noalias !99
%memoryref_data12 = load ptr, ptr %"y::Array", align 8, !tbaa !95, !alias.scope !98, !noalias !99
; │ @ simdloop.jl:75 within `macro expansion`
%min.iters.check = icmp eq i64 %"x::Array.size.0.copyload", 1, !dbg !100
br i1 %min.iters.check, label %L31.preheader, label %vector.memcheck, !dbg !100
vector.memcheck: ; preds = %iter.check
%3 = shl i64 %"x::Array.size.0.copyload", 3, !dbg !100
%scevgep = getelementptr i8, ptr %memoryref_data12, i64 %3, !dbg !100
%scevgep44 = getelementptr i8, ptr %memoryref_data, i64 %3, !dbg !100
%bound0 = icmp ult ptr %memoryref_data12, %scevgep44, !dbg !100
%bound1 = icmp ult ptr %memoryref_data, %scevgep, !dbg !100
%found.conflict = and i1 %bound0, %bound1, !dbg !100
br i1 %found.conflict, label %L31.preheader, label %vector.main.loop.iter.check, !dbg !100
vector.main.loop.iter.check: ; preds = %vector.memcheck
%min.iters.check45 = icmp samesign ult i64 %"x::Array.size.0.copyload", 8, !dbg !100
br i1 %min.iters.check45, label %vector.main.loop.iter.check.vec.epilog.ph_crit_edge, label %vector.ph, !dbg !100
vector.main.loop.iter.check.vec.epilog.ph_crit_edge: ; preds = %vector.main.loop.iter.check
%.pre = insertelement <2 x double> poison, double %"a::Float64", i64 0, !dbg !100
%.pre1 = shufflevector <2 x double> %.pre, <2 x double> poison, <2 x i32> zeroinitializer, !dbg !100
br label %vec.epilog.ph, !dbg !100
vector.ph: ; preds = %vector.main.loop.iter.check
%n.mod.vf = and i64 %"x::Array.size.0.copyload", 6, !dbg !100
%n.vec = and i64 %"x::Array.size.0.copyload", 9223372036854775800, !dbg !100
%broadcast.splatinsert = insertelement <2 x double> poison, double %"a::Float64", i64 0, !dbg !100
%broadcast.splat = shufflevector <2 x double> %broadcast.splatinsert, <2 x double> poison, <2 x i32> zeroinitializer, !dbg !100
br label %vector.body, !dbg !100
vector.body: ; preds = %vector.body, %vector.ph
; │ @ simdloop.jl:76 within `macro expansion`
; │┌ @ simdloop.jl:54 within `simd_index`
; ││┌ @ int.jl:87 within `+`
%index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ], !dbg !101
; │└└
; │ @ simdloop.jl:77 within `macro expansion` @ REPL[7]:3
; │┌ @ essentials.jl:953 within `getindex`
%4 = getelementptr inbounds nuw double, ptr %memoryref_data, i64 %index, !dbg !106
%5 = getelementptr inbounds nuw i8, ptr %4, i64 16, !dbg !106
%6 = getelementptr inbounds nuw i8, ptr %4, i64 32, !dbg !106
%7 = getelementptr inbounds nuw i8, ptr %4, i64 48, !dbg !106
%wide.load = load <2 x double>, ptr %4, align 8, !dbg !106, !tbaa !112, !alias.scope !115, !noalias !118
%wide.load46 = load <2 x double>, ptr %5, align 8, !dbg !106, !tbaa !112, !alias.scope !115, !noalias !118
%wide.load47 = load <2 x double>, ptr %6, align 8, !dbg !106, !tbaa !112, !alias.scope !115, !noalias !118
%wide.load48 = load <2 x double>, ptr %7, align 8, !dbg !106, !tbaa !112, !alias.scope !115, !noalias !118
%8 = getelementptr inbounds nuw double, ptr %memoryref_data12, i64 %index, !dbg !106
%9 = getelementptr inbounds nuw i8, ptr %8, i64 16, !dbg !106
%10 = getelementptr inbounds nuw i8, ptr %8, i64 32, !dbg !106
%11 = getelementptr inbounds nuw i8, ptr %8, i64 48, !dbg !106
%wide.load49 = load <2 x double>, ptr %8, align 8, !dbg !106, !tbaa !112, !alias.scope !119, !noalias !121
%wide.load50 = load <2 x double>, ptr %9, align 8, !dbg !106, !tbaa !112, !alias.scope !119, !noalias !121
%wide.load51 = load <2 x double>, ptr %10, align 8, !dbg !106, !tbaa !112, !alias.scope !119, !noalias !121
%wide.load52 = load <2 x double>, ptr %11, align 8, !dbg !106, !tbaa !112, !alias.scope !119, !noalias !121
; │└
; │┌ @ float.jl:497 within `muladd`
%12 = fmul contract <2 x double> %broadcast.splat, %wide.load, !dbg !122
%13 = fmul contract <2 x double> %broadcast.splat, %wide.load46, !dbg !122
%14 = fmul contract <2 x double> %broadcast.splat, %wide.load47, !dbg !122
%15 = fmul contract <2 x double> %broadcast.splat, %wide.load48, !dbg !122
%16 = fadd contract <2 x double> %12, %wide.load49, !dbg !122
%17 = fadd contract <2 x double> %13, %wide.load50, !dbg !122
%18 = fadd contract <2 x double> %14, %wide.load51, !dbg !122
%19 = fadd contract <2 x double> %15, %wide.load52, !dbg !122
; │└
; │┌ @ array.jl:986 within `setindex!`
; ││┌ @ array.jl:991 within `_setindex!`
store <2 x double> %16, ptr %8, align 8, !dbg !125, !tbaa !112, !alias.scope !119, !noalias !121
store <2 x double> %17, ptr %9, align 8, !dbg !125, !tbaa !112, !alias.scope !119, !noalias !121
store <2 x double> %18, ptr %10, align 8, !dbg !125, !tbaa !112, !alias.scope !119, !noalias !121
store <2 x double> %19, ptr %11, align 8, !dbg !125, !tbaa !112, !alias.scope !119, !noalias !121
; │└└
; │ @ simdloop.jl:76 within `macro expansion`
; │┌ @ simdloop.jl:54 within `simd_index`
; ││┌ @ int.jl:87 within `+`
%index.next = add nuw i64 %index, 8, !dbg !101
%20 = icmp eq i64 %index.next, %n.vec, !dbg !101
br i1 %20, label %middle.block, label %vector.body, !dbg !101, !llvm.loop !129
middle.block: ; preds = %vector.body
; │└└
; │ @ simdloop.jl:75 within `macro expansion`
%cmp.n = icmp eq i64 %"x::Array.size.0.copyload", %n.vec, !dbg !100
br i1 %cmp.n, label %L103, label %vec.epilog.iter.check, !dbg !100
vec.epilog.iter.check: ; preds = %middle.block
%min.epilog.iters.check = icmp eq i64 %n.mod.vf, 0, !dbg !100
br i1 %min.epilog.iters.check, label %L31.preheader, label %vec.epilog.ph, !dbg !100
vec.epilog.ph: ; preds = %vector.main.loop.iter.check.vec.epilog.ph_crit_edge, %vec.epilog.iter.check
%broadcast.splat59.pre-phi = phi <2 x double> [ %.pre1, %vector.main.loop.iter.check.vec.epilog.ph_crit_edge ], [ %broadcast.splat, %vec.epilog.iter.check ], !dbg !100
%vec.epilog.resume.val = phi i64 [ 0, %vector.main.loop.iter.check.vec.epilog.ph_crit_edge ], [ %n.vec, %vec.epilog.iter.check ]
%n.vec54 = and i64 %"x::Array.size.0.copyload", 9223372036854775806, !dbg !100
br label %vec.epilog.vector.body, !dbg !100
vec.epilog.vector.body: ; preds = %vec.epilog.vector.body, %vec.epilog.ph
; │ @ simdloop.jl:76 within `macro expansion`
; │┌ @ simdloop.jl:54 within `simd_index`
; ││┌ @ int.jl:87 within `+`
%index55 = phi i64 [ %vec.epilog.resume.val, %vec.epilog.ph ], [ %index.next60, %vec.epilog.vector.body ], !dbg !101
; │└└
; │ @ simdloop.jl:77 within `macro expansion` @ REPL[7]:3
; │┌ @ essentials.jl:953 within `getindex`
%gep = getelementptr double, ptr %memoryref_data, i64 %index55, !dbg !106
%wide.load56 = load <2 x double>, ptr %gep, align 8, !dbg !106, !tbaa !112, !alias.scope !132, !noalias !118
%gep63 = getelementptr double, ptr %memoryref_data12, i64 %index55, !dbg !106
%wide.load57 = load <2 x double>, ptr %gep63, align 8, !dbg !106, !tbaa !112, !alias.scope !135, !noalias !137
; │└
; │┌ @ float.jl:497 within `muladd`
%21 = fmul contract <2 x double> %broadcast.splat59.pre-phi, %wide.load56, !dbg !122
%22 = fadd contract <2 x double> %21, %wide.load57, !dbg !122
; │└
; │┌ @ array.jl:986 within `setindex!`
; ││┌ @ array.jl:991 within `_setindex!`
store <2 x double> %22, ptr %gep63, align 8, !dbg !125, !tbaa !112, !alias.scope !135, !noalias !137
; │└└
; │ @ simdloop.jl:76 within `macro expansion`
; │┌ @ simdloop.jl:54 within `simd_index`
; ││┌ @ int.jl:87 within `+`
%index.next60 = add nuw i64 %index55, 2, !dbg !101
%23 = icmp eq i64 %index.next60, %n.vec54, !dbg !101
br i1 %23, label %vec.epilog.middle.block, label %vec.epilog.vector.body, !dbg !101, !llvm.loop !138
vec.epilog.middle.block: ; preds = %vec.epilog.vector.body
; │└└
; │ @ simdloop.jl:75 within `macro expansion`
%cmp.n61 = icmp eq i64 %"x::Array.size.0.copyload", %n.vec54, !dbg !100
br i1 %cmp.n61, label %L103, label %L31.preheader, !dbg !100
L31.preheader: ; preds = %vec.epilog.middle.block, %vec.epilog.iter.check, %vector.memcheck, %iter.check
%value_phi343.ph = phi i64 [ %n.vec, %vec.epilog.iter.check ], [ 0, %iter.check ], [ 0, %vector.memcheck ], [ %n.vec54, %vec.epilog.middle.block ]
%.neg = or disjoint i64 %value_phi343.ph, 1, !dbg !100
%xtraiter = and i64 %"x::Array.size.0.copyload", 1, !dbg !100
%lcmp.mod.not = icmp eq i64 %xtraiter, 0, !dbg !100
br i1 %lcmp.mod.not, label %L31.prol.loopexit, label %L31.prol.preheader, !dbg !100
L31.prol.preheader: ; preds = %L31.preheader
; │ @ simdloop.jl:77 within `macro expansion` @ REPL[7]:3
; │┌ @ essentials.jl:953 within `getindex`
%memoryref_data8.prol = getelementptr inbounds nuw double, ptr %memoryref_data, i64 %value_phi343.ph, !dbg !106
%24 = load double, ptr %memoryref_data8.prol, align 8, !dbg !106, !tbaa !112, !alias.scope !139, !noalias !118
%memoryref_data19.prol = getelementptr inbounds nuw double, ptr %memoryref_data12, i64 %value_phi343.ph, !dbg !106
%25 = load double, ptr %memoryref_data19.prol, align 8, !dbg !106, !tbaa !112, !alias.scope !139, !noalias !118
; │└
; │┌ @ float.jl:497 within `muladd`
%26 = fmul contract double %"a::Float64", %24, !dbg !122
%27 = fadd contract double %26, %25, !dbg !122
; │└
; │┌ @ array.jl:986 within `setindex!`
; ││┌ @ array.jl:991 within `_setindex!`
store double %27, ptr %memoryref_data19.prol, align 8, !dbg !125, !tbaa !112, !alias.scope !139, !noalias !118
; │└└
; │ @ simdloop.jl:75 within `macro expansion`
br label %L31.prol.loopexit, !dbg !100
L31.prol.loopexit: ; preds = %L31.prol.preheader, %L31.preheader
%value_phi343.unr = phi i64 [ %value_phi343.ph, %L31.preheader ], [ %.neg, %L31.prol.preheader ]
%28 = icmp eq i64 %"x::Array.size.0.copyload", %.neg, !dbg !100
br i1 %28, label %L103, label %L31, !dbg !100
L31: ; preds = %L31.prol.loopexit, %L31
%value_phi343 = phi i64 [ %34, %L31 ], [ %value_phi343.unr, %L31.prol.loopexit ]
; │ @ simdloop.jl:76 within `macro expansion`
; │┌ @ simdloop.jl:54 within `simd_index`
; ││┌ @ int.jl:87 within `+`
%29 = add nuw nsw i64 %value_phi343, 1, !dbg !101
; │└└
; │ @ simdloop.jl:77 within `macro expansion` @ REPL[7]:3
; │┌ @ essentials.jl:953 within `getindex`
%memoryref_data8 = getelementptr inbounds nuw double, ptr %memoryref_data, i64 %value_phi343, !dbg !106
%30 = load double, ptr %memoryref_data8, align 8, !dbg !106, !tbaa !112, !alias.scope !139, !noalias !118
%memoryref_data19 = getelementptr inbounds nuw double, ptr %memoryref_data12, i64 %value_phi343, !dbg !106
%31 = load double, ptr %memoryref_data19, align 8, !dbg !106, !tbaa !112, !alias.scope !139, !noalias !118
; │└
; │┌ @ float.jl:497 within `muladd`
%32 = fmul contract double %"a::Float64", %30, !dbg !122
%33 = fadd contract double %32, %31, !dbg !122
; │└
; │┌ @ array.jl:986 within `setindex!`
; ││┌ @ array.jl:991 within `_setindex!`
store double %33, ptr %memoryref_data19, align 8, !dbg !125, !tbaa !112, !alias.scope !139, !noalias !118
; │└└
; │ @ simdloop.jl:76 within `macro expansion`
; │┌ @ simdloop.jl:54 within `simd_index`
; ││┌ @ int.jl:87 within `+`
%34 = add nuw nsw i64 %value_phi343, 2, !dbg !101
; │└└
; │ @ simdloop.jl:77 within `macro expansion` @ REPL[7]:3
; │┌ @ essentials.jl:953 within `getindex`
%memoryref_data8.1 = getelementptr inbounds nuw double, ptr %memoryref_data, i64 %29, !dbg !106
%35 = load double, ptr %memoryref_data8.1, align 8, !dbg !106, !tbaa !112, !alias.scope !139, !noalias !118
%memoryref_data19.1 = getelementptr inbounds nuw double, ptr %memoryref_data12, i64 %29, !dbg !106
%36 = load double, ptr %memoryref_data19.1, align 8, !dbg !106, !tbaa !112, !alias.scope !139, !noalias !118
; │└
; │┌ @ float.jl:497 within `muladd`
%37 = fmul contract double %"a::Float64", %35, !dbg !122
%38 = fadd contract double %37, %36, !dbg !122
; │└
; │┌ @ array.jl:986 within `setindex!`
; ││┌ @ array.jl:991 within `_setindex!`
store double %38, ptr %memoryref_data19.1, align 8, !dbg !125, !tbaa !112, !alias.scope !139, !noalias !118
; │└└
; │ @ simdloop.jl:75 within `macro expansion`
; │┌ @ int.jl:83 within `<`
%exitcond.not.1 = icmp eq i64 %34, %"x::Array.size.0.copyload", !dbg !140
; │└
br i1 %exitcond.not.1, label %L103, label %L31, !dbg !100, !llvm.loop !141
L103: ; preds = %L31.prol.loopexit, %L31, %vec.epilog.middle.block, %middle.block, %L24
; │ @ simdloop.jl:76 within `macro expansion`
; │┌ @ simdloop.jl:54 within `simd_index`
; ││┌ @ array.jl:3149 within `getindex`
; │││┌ @ range.jl:942 within `_getindex`
; ││││┌ @ abstractarray.jl:697 within `checkbounds`
ret void, !dbg !142
; └└└└└
}
; Function Attrs: noinline optnone
define nonnull ptr @"jfptr_axpy!_3604"(ptr %"function::Core.Function", ptr noalias nocapture noundef readonly %"args::Any[]", i32 %"nargs::UInt32") local_unnamed_addr #1 {
top:
%thread_ptr = call ptr asm "mrs $0, tpidr_el0", "=r"()
%tls_ppgcstack = getelementptr inbounds i8, ptr %thread_ptr, i64 16
%tls_pgcstack = load ptr, ptr %tls_ppgcstack, align 8
%0 = getelementptr inbounds i8, ptr %"args::Any[]", i32 0
%1 = load ptr, ptr %0, align 8, !tbaa !27, !invariant.load !9, !alias.scope !148, !noalias !149, !nonnull !9, !dereferenceable !150, !align !150
%2 = getelementptr inbounds i8, ptr %"args::Any[]", i32 8
%3 = load ptr, ptr %2, align 8, !tbaa !27, !invariant.load !9, !alias.scope !148, !noalias !149, !nonnull !9, !dereferenceable !151, !align !150
%4 = getelementptr inbounds i8, ptr %"args::Any[]", i32 16
%5 = load ptr, ptr %4, align 8, !tbaa !27, !invariant.load !9, !alias.scope !148, !noalias !149, !nonnull !9, !dereferenceable !151, !align !150
%.unbox = load double, ptr %1, align 8, !tbaa !152, !alias.scope !139, !noalias !118
call swiftcc void @"julia_axpy!_3603"(ptr nonnull swiftself %tls_pgcstack, double %.unbox, ptr %3, ptr %5)
%jl_nothing = load ptr, ptr @jl_nothing, align 8, !tbaa !27, !invariant.load !9, !alias.scope !148, !noalias !149, !nonnull !9
ret ptr %jl_nothing
}
; Function Attrs: memory(argmem: readwrite, inaccessiblemem: readwrite)
declare void @julia.safepoint(ptr) local_unnamed_addr #2
; Function Attrs: mustprogress nofree norecurse nosync nounwind speculatable willreturn memory(none)
declare noundef nonnull ptr @julia.gc_loaded(ptr nocapture noundef nonnull readnone, ptr noundef nonnull readnone) local_unnamed_addr #3
; Function Signature: throw_eachindex_mismatch_indices(String, Base.OneTo{Int64}, Base.OneTo{Int64})
; Function Attrs: noreturn
declare swiftcc void @j_throw_eachindex_mismatch_indices_3607(ptr nonnull swiftself, ptr, ptr nocapture readonly, ptr nocapture readonly) local_unnamed_addr #4
; Function Attrs: nounwind willreturn allockind("alloc") allocsize(2) memory(argmem: read, inaccessiblemem: readwrite)
declare noalias nonnull ptr @ijl_gc_small_alloc(ptr, i32, i32, i64) #5
; Function Attrs: memory(argmem: readwrite, inaccessiblemem: readwrite)
declare void @ijl_gc_queue_root(ptr) #2
; Function Attrs: nounwind willreturn allockind("alloc") allocsize(1) memory(argmem: read, inaccessiblemem: readwrite)
declare noalias nonnull ptr @ijl_gc_big_alloc(ptr, i64, i64) #6
; Function Attrs: nounwind willreturn allockind("alloc") allocsize(1) memory(argmem: read, inaccessiblemem: readwrite)
declare noalias nonnull ptr @ijl_gc_alloc_typed(ptr, i64, i64) #6
attributes #0 = { "frame-pointer"="all" "julia.fsig"="axpy!(Float64, Array{Float64, 1}, Array{Float64, 1})" "probe-stack"="inline-asm" }
attributes #1 = { noinline optnone "frame-pointer"="all" "probe-stack"="inline-asm" }
attributes #2 = { memory(argmem: readwrite, inaccessiblemem: readwrite) }
attributes #3 = { mustprogress nofree norecurse nosync nounwind speculatable willreturn memory(none) }
attributes #4 = { noreturn "frame-pointer"="all" "julia.fsig"="throw_eachindex_mismatch_indices(String, Base.OneTo{Int64}, Base.OneTo{Int64})" "probe-stack"="inline-asm" }
attributes #5 = { nounwind willreturn allockind("alloc") allocsize(2) memory(argmem: read, inaccessiblemem: readwrite) }
attributes #6 = { nounwind willreturn allockind("alloc") allocsize(1) memory(argmem: read, inaccessiblemem: readwrite) }
attributes #7 = { noreturn }
!llvm.module.flags = !{!0, !1}
!llvm.dbg.cu = !{!2}
!0 = !{i32 2, !"Dwarf Version", i32 4}
!1 = !{i32 2, !"Debug Info Version", i32 3}
!2 = distinct !DICompileUnit(language: DW_LANG_Julia, file: !3, producer: "julia", isOptimized: true, runtimeVersion: 0, emissionKind: NoDebug, nameTableKind: GNU)
!3 = !DIFile(filename: "julia", directory: ".")
!4 = distinct !DISubprogram(name: "axpy!", linkageName: "julia_axpy!_3603", scope: null, file: !5, line: 1, type: !6, scopeLine: 1, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2, retainedNodes: !17)
!5 = !DIFile(filename: "REPL[7]", directory: ".")
!6 = !DISubroutineType(types: !7)
!7 = !{!8, !10, !11, !12, !12}
!8 = !DICompositeType(tag: DW_TAG_structure_type, name: "Nothing", align: 8, elements: !9, runtimeLang: DW_LANG_Julia, identifier: "281472950670688")
!9 = !{}
!10 = !DICompositeType(tag: DW_TAG_structure_type, name: "#axpy!", align: 8, elements: !9, runtimeLang: DW_LANG_Julia, identifier: "281472803406096")
!11 = !DIBasicType(name: "Float64", size: 64, encoding: DW_ATE_unsigned)
!12 = !DIDerivedType(tag: DW_TAG_typedef, name: "Array", baseType: !13)
!13 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !14, size: 64, align: 64)
!14 = !DICompositeType(tag: DW_TAG_structure_type, name: "jl_value_t", file: !15, line: 71, align: 64, elements: !16)
!15 = !DIFile(filename: "julia.h", directory: "")
!16 = !{!13}
!17 = !{!18, !19, !20, !21}
!18 = !DILocalVariable(name: "#self#", arg: 1, scope: !4, file: !5, line: 1, type: !10)
!19 = !DILocalVariable(name: "a", arg: 2, scope: !4, file: !5, line: 1, type: !11)
!20 = !DILocalVariable(name: "x", arg: 3, scope: !4, file: !5, line: 1, type: !12)
!21 = !DILocalVariable(name: "y", arg: 4, scope: !4, file: !5, line: 1, type: !12)
!22 = !DILocation(line: 1, scope: !4)
!23 = !{!24, !24, i64 0}
!24 = !{!"jtbaa_gcframe", !25, i64 0}
!25 = !{!"jtbaa", !26, i64 0}
!26 = !{!"jtbaa"}
!27 = !{!28, !28, i64 0}
!28 = !{!"jtbaa_const", !25, i64 0}
!29 = !DILocation(line: 194, scope: !30, inlinedAt: !33)
!30 = distinct !DISubprogram(name: "size;", linkageName: "size", scope: !31, file: !31, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!31 = !DIFile(filename: "array.jl", directory: ".")
!32 = !DISubroutineType(types: !9)
!33 = !DILocation(line: 98, scope: !34, inlinedAt: !36)
!34 = distinct !DISubprogram(name: "axes;", linkageName: "axes", scope: !35, file: !35, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!35 = !DIFile(filename: "abstractarray.jl", directory: ".")
!36 = !DILocation(line: 137, scope: !37, inlinedAt: !38)
!37 = distinct !DISubprogram(name: "axes1;", linkageName: "axes1", scope: !35, file: !35, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!38 = !DILocation(line: 391, scope: !39, inlinedAt: !40)
!39 = distinct !DISubprogram(name: "eachindex;", linkageName: "eachindex", scope: !35, file: !35, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!40 = !DILocation(line: 394, scope: !39, inlinedAt: !41)
!41 = !DILocation(line: 384, scope: !39, inlinedAt: !42)
!42 = !DILocation(line: 69, scope: !43, inlinedAt: !45)
!43 = distinct !DISubprogram(name: "macro expansion;", linkageName: "macro expansion", scope: !44, file: !44, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!44 = !DIFile(filename: "simdloop.jl", directory: ".")
!45 = !DILocation(line: 2, scope: !4)
!46 = !{!25, !25, i64 0}
!47 = !{!48, !50}
!48 = !{!"jnoalias_typemd", !49}
!49 = !{!"jnoalias"}
!50 = !{!"jnoalias_stack", !49}
!51 = !{!52, !53, !54}
!52 = !{!"jnoalias_gcframe", !49}
!53 = !{!"jnoalias_data", !49}
!54 = !{!"jnoalias_const", !49}
!55 = !DILocation(line: 486, scope: !56, inlinedAt: !58)
!56 = distinct !DISubprogram(name: "unchecked_oneto;", linkageName: "unchecked_oneto", scope: !57, file: !57, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!57 = !DIFile(filename: "range.jl", directory: ".")
!58 = !DILocation(line: 358, scope: !59, inlinedAt: !33)
!59 = distinct !DISubprogram(name: "map;", linkageName: "map", scope: !60, file: !60, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!60 = !DIFile(filename: "tuple.jl", directory: ".")
!61 = !{!62, !62, i64 0}
!62 = !{!"jtbaa_stack", !25, i64 0}
!63 = !{!50}
!64 = !{!52, !53, !48, !54}
!65 = !DILocation(line: 194, scope: !30, inlinedAt: !66)
!66 = !DILocation(line: 98, scope: !34, inlinedAt: !67)
!67 = !DILocation(line: 137, scope: !37, inlinedAt: !68)
!68 = !DILocation(line: 391, scope: !39, inlinedAt: !69)
!69 = !DILocation(line: 395, scope: !70, inlinedAt: !71)
!70 = distinct !DISubprogram(name: "#eachindex##0;", linkageName: "#eachindex##0", scope: !35, file: !35, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!71 = !DILocation(line: 358, scope: !59, inlinedAt: !72)
!72 = !DILocation(line: 395, scope: !39, inlinedAt: !41)
!73 = !DILocation(line: 486, scope: !56, inlinedAt: !74)
!74 = !DILocation(line: 358, scope: !59, inlinedAt: !66)
!75 = !DILocation(line: 632, scope: !76, inlinedAt: !78)
!76 = distinct !DISubprogram(name: "==;", linkageName: "==", scope: !77, file: !77, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!77 = !DIFile(filename: "promotion.jl", directory: ".")
!78 = !DILocation(line: 1145, scope: !79, inlinedAt: !80)
!79 = distinct !DISubprogram(name: "==;", linkageName: "==", scope: !57, file: !57, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!80 = !DILocation(line: 1195, scope: !81, inlinedAt: !83)
!81 = distinct !DISubprogram(name: "#_#59;", linkageName: "#_#59", scope: !82, file: !82, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!82 = !DIFile(filename: "operators.jl", directory: ".")
!83 = !DILocation(line: 1195, scope: !84, inlinedAt: !85)
!84 = distinct !DISubprogram(name: "Fix;", linkageName: "Fix", scope: !82, file: !82, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!85 = !DILocation(line: 211, scope: !86, inlinedAt: !88)
!86 = distinct !DISubprogram(name: "_all;", linkageName: "_all", scope: !87, file: !87, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!87 = !DIFile(filename: "anyall.jl", directory: ".")
!88 = !DILocation(line: 240, scope: !89, inlinedAt: !90)
!89 = distinct !DISubprogram(name: "all;", linkageName: "all", scope: !87, file: !87, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!90 = !DILocation(line: 396, scope: !39, inlinedAt: !41)
!91 = !DILocation(line: 83, scope: !92, inlinedAt: !94)
!92 = distinct !DISubprogram(name: "<;", linkageName: "<", scope: !93, file: !93, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!93 = !DIFile(filename: "int.jl", directory: ".")
!94 = !DILocation(line: 72, scope: !43, inlinedAt: !45)
!95 = !{!96, !96, i64 0}
!96 = !{!"jtbaa_arrayptr", !97, i64 0}
!97 = !{!"jtbaa_array", !25, i64 0}
!98 = !{!48}
!99 = !{!52, !50, !53, !54}
!100 = !DILocation(line: 75, scope: !43, inlinedAt: !45)
!101 = !DILocation(line: 87, scope: !102, inlinedAt: !103)
!102 = distinct !DISubprogram(name: "+;", linkageName: "+", scope: !93, file: !93, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!103 = !DILocation(line: 54, scope: !104, inlinedAt: !105)
!104 = distinct !DISubprogram(name: "simd_index;", linkageName: "simd_index", scope: !44, file: !44, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!105 = !DILocation(line: 76, scope: !43, inlinedAt: !45)
!106 = !DILocation(line: 953, scope: !107, inlinedAt: !109)
!107 = distinct !DISubprogram(name: "getindex;", linkageName: "getindex", scope: !108, file: !108, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!108 = !DIFile(filename: "essentials.jl", directory: ".")
!109 = !DILocation(line: 3, scope: !110, inlinedAt: !111)
!110 = distinct !DISubprogram(name: "macro expansion;", linkageName: "macro expansion", scope: !5, file: !5, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!111 = !DILocation(line: 77, scope: !43, inlinedAt: !45)
!112 = !{!113, !113, i64 0}
!113 = !{!"jtbaa_arraybuf", !114, i64 0}
!114 = !{!"jtbaa_data", !25, i64 0}
!115 = !{!53, !116}
!116 = distinct !{!116, !117}
!117 = distinct !{!117, !"LVerDomain"}
!118 = !{!52, !50, !48, !54}
!119 = !{!53, !120}
!120 = distinct !{!120, !117}
!121 = !{!52, !50, !48, !54, !116}
!122 = !DILocation(line: 497, scope: !123, inlinedAt: !109)
!123 = distinct !DISubprogram(name: "muladd;", linkageName: "muladd", scope: !124, file: !124, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!124 = !DIFile(filename: "float.jl", directory: ".")
!125 = !DILocation(line: 991, scope: !126, inlinedAt: !127)
!126 = distinct !DISubprogram(name: "_setindex!;", linkageName: "_setindex!", scope: !31, file: !31, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!127 = !DILocation(line: 986, scope: !128, inlinedAt: !109)
!128 = distinct !DISubprogram(name: "setindex!;", linkageName: "setindex!", scope: !31, file: !31, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!129 = distinct !{!129, !130, !131}
!130 = !{!"llvm.loop.isvectorized", i32 1}
!131 = !{!"llvm.loop.unroll.runtime.disable"}
!132 = !{!53, !133}
!133 = distinct !{!133, !134}
!134 = distinct !{!134, !"LVerDomain"}
!135 = !{!53, !136}
!136 = distinct !{!136, !134}
!137 = !{!52, !50, !48, !54, !133}
!138 = distinct !{!138, !130, !131}
!139 = !{!53}
!140 = !DILocation(line: 83, scope: !92, inlinedAt: !100)
!141 = distinct !{!141, !130}
!142 = !DILocation(line: 697, scope: !143, inlinedAt: !144)
!143 = distinct !DISubprogram(name: "checkbounds;", linkageName: "checkbounds", scope: !35, file: !35, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!144 = !DILocation(line: 942, scope: !145, inlinedAt: !146)
!145 = distinct !DISubprogram(name: "_getindex;", linkageName: "_getindex", scope: !57, file: !57, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!146 = !DILocation(line: 3149, scope: !147, inlinedAt: !103)
!147 = distinct !DISubprogram(name: "getindex;", linkageName: "getindex", scope: !31, file: !31, type: !32, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2)
!148 = !{!54}
!149 = !{!52, !50, !53, !48}
!150 = !{i64 8}
!151 = !{i64 24}
!152 = !{!153, !153, i64 0}
!153 = !{!"jtbaa_immut", !154, i64 0}
!154 = !{!"jtbaa_value", !114, i64 0}Still no vscale. |
|
The attributte didn't get added to the function :\ |
giordano
reviewed
May 16, 2025
| SmallVector<StringRef, 128> Features; | ||
| FS.split(Features, ','); | ||
| for (StringRef Feature : Features) { | ||
| if (Feature == "sve") |
There was a problem hiding this comment.
Suggested change
| if (Feature == "sve") | |
| if (Feature == "+sve") |
With this the function gets the attribute vscale_range
; Function Attrs: vscale_range(1,16)
define swiftcc void @"julia_axpy!_763"(ptr nonnull swiftself %pgcstack_arg, double %"a::Float64", ptr noundef nonnull align 8 dereferenceable(24) %"x::Array", ptr noundef nonnull align 8 dereferenceable(24) %"y::Array") local_unnamed_addr #0 !dbg !4 {
; [...]
attributes #0 = { vscale_range(1,16) "frame-pointer"="all" "julia.fsig"="axpy!(Float64, Array{Float64, 1}, Array{Float64, 1})" "probe-stack"="inline-asm" }
attributes #1 = { noinline optnone vscale_range(1,16) "frame-pointer"="all" "probe-stack"="inline-asm" }
attributes #2 = { memory(argmem: readwrite, inaccessiblemem: readwrite) vscale_range(1,16) }
attributes #3 = { mustprogress nofree norecurse nosync nounwind speculatable willreturn memory(none) vscale_range(1,16) }
attributes #4 = { noreturn vscale_range(1,16) "frame-pointer"="all" "julia.fsig"="throw_eachindex_mismatch_indices(String, Base.OneTo{Int64}, Base.OneTo{Int64})" "probe-stack"="inline-asm" }
attributes #5 = { nounwind willreturn allockind("alloc") allocsize(2) memory(argmem: read, inaccessiblemem: readwrite) }
attributes #6 = { memory(argmem: readwrite, inaccessiblemem: readwrite) }
attributes #7 = { nounwind willreturn allockind("alloc") allocsize(1) memory(argmem: read, inaccessiblemem: readwrite) }
attributes #8 = { noreturn }but there's still not vscale in the vector body.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
For some reason LLVM expects frontends to do this intead of querying the TTI. So just add a default when it makes sense. ( I wonder if this should be done elsewhere but this pass seems like it can handle this fine and everything is already there)