Merge pull request #22 from topolarity/ct/no-eval-pref

Add Preference to disable runtime invalidation

Author: oscardssmith

.github/workflows/CI.yml

@@ -10,7 +10,7 @@ jobs:
       fail-fast: false
       matrix:
         version:
-          - '1.5'
+          - '1.10'
           - '1'
           - 'nightly'
         os:

Project.toml

@@ -7,13 +7,14 @@ version = "0.1.17"
 BitTwiddlingConvenienceFunctions = "62783981-4cbd-42fc-bca8-16325de8dc4b"
 IfElse = "615f187c-cbe4-4ef1-ba3b-2fcf58d6d173"
 Libdl = "8f399da3-3557-5675-b5ff-fb832c97cbdb"
+Preferences = "21216c6a-2e73-6563-6e65-726566657250"
 Static = "aedffcd0-7271-4cad-89d0-dc628f76c6d3"
 
 [compat]
 BitTwiddlingConvenienceFunctions = "0.1"
 IfElse = "0.1"
 Static = "0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 1"
-julia = "1.5"
+julia = "1.10"
 
 [extras]
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"

docs/src/index.md

@@ -12,3 +12,9 @@ Documentation for [HostCPUFeatures](https://github.com/JuliaSIMD/HostCPUFeatures
 ```@autodocs
 Modules = [HostCPUFeatures]
 ```
+
+## Supported Preferences
+
+  - `cpu_target`: if provided, use this string as your CPU target for feature detection instead of `JULIA_CPU_TARGET`
+  - `freeze_cpu_target`: if `true`, "freeze" the features detected based on your precompile-time CPU target and do not perform runtime feature detection
+  - `allow_runtime_invalidation`: if `false`, warn when performing runtime feature detection (instead of invalidating) when CPU features don't match precompile-time

src/HostCPUFeatures.jl

@@ -7,12 +7,32 @@ end
 using Libdl, Static
 using Static: Zero, One, lt, gt
 using IfElse: ifelse
+using Preferences
 
 using BitTwiddlingConvenienceFunctions: prevpow2, nextpow2, intlog2
 
 export has_feature, fma_fast, pick_vector_width, pick_vector_width_shift, register_count,
   register_size, simd_integer_register_size
 
+_cpu_target = if @has_preference("cpu_target")
+  @load_preference("cpu_target")
+else
+  Base.unsafe_string(Base.JLOptions().cpu_target)
+end
+
+const build_cpu_target = if occursin("native", _cpu_target)
+  "native" # 'native' takes priority if provided
+else
+  split(_cpu_target, ";")[1]
+end
+
+# If true, this will opt-in to "freeze" an under-approximation of the CPU features at precompile-
+# time based on the CPU target.
+#
+# This is only done by default if "native" was excluded from the CPU target (or via a preference).
+const freeze_cpu_target =
+  @load_preference("freeze_cpu_target", false) || build_cpu_target != "native"
+
 function get_cpu_name()::String
   if isdefined(Sys, :CPU_NAME)
     Sys.CPU_NAME
@@ -37,19 +57,43 @@ unwrap(::StaticSymbol{S}) where {S} = S
 
 @noinline function redefine()
   @debug "Defining CPU name."
-  define_cpu_name()
+  redefine_cpu_name()
 
   reset_features!()
   reset_extra_features!()
 end
 const BASELINE_CPU_NAME = get_cpu_name()
+const allow_eval = @load_preference("allow_runtime_invalidation", false)
+
+function make_generic(target)
+  target == "native" && return false
+  if Sys.ARCH === :x86_64 || Sys.ARCH === :i686
+    make_generic_x86(target)
+    return true
+  else
+    return false
+  end
+end
+
+make_generic(build_cpu_target)
+
 function __init__()
   ccall(:jl_generating_output, Cint, ()) == 1 && return
-  if Sys.ARCH === :x86_64 || Sys.ARCH === :i686
-    target = Base.unsafe_string(Base.JLOptions().cpu_target)
-    occursin("native",  target) || return make_generic(target)
+  freeze_cpu_target && return # CPU info fixed at precompile-time
+
+  runtime_target = Base.unsafe_string(Base.JLOptions().cpu_target)
+  if !occursin("native", runtime_target)
+    # The CPU target included "native" at pre-compile time, but at runtime it did not!
+    #
+    # Fixing this discepancy will invalidate the whole world (so it should probably
+    # throw an error), but we do it anyway for backwards-compatibility.
+    if make_generic(runtime_target)
+      return nothing
+    end
+  end
+  if BASELINE_CPU_NAME != Sys.CPU_NAME::String
+    redefine()
   end
-  BASELINE_CPU_NAME == Sys.CPU_NAME::String || redefine()
   return nothing
 end
 

src/cpu_info.jl

@@ -39,17 +39,22 @@ function set_features!()
   end
   Libc.free(features_cstring)
 end
-set_features!()
-
 
+if build_cpu_target == "native"
+    set_features!()
+end
 
 function reset_features!()
     features, features_cstring = feature_string()
     for ext ∈ features
         feature, has = process_feature(ext)
         if _has_feature(feature) ≠ has
-            @debug "Defining $(has ? "presence" : "absense") of feature $feature."
-            set_feature(feature, has)
+            if allow_eval
+                @debug "Defining $(has ? "presence" : "absense") of feature $feature."
+                set_feature(feature, has)
+            else
+                @warn "Runtime invalidation was disabled, but the CPU info is out-of-date.\nWill continue with incorrect CPU feature flag: $ext."
+            end
         end
     end
     Libc.free(features_cstring)
@@ -58,8 +63,16 @@ end
 register_size(::Type{T}) where {T} = register_size()
 register_size(::Type{T}) where {T<:Union{Signed,Unsigned}} = simd_integer_register_size()
 
-function define_cpu_name()
+function redefine_cpu_name()
     cpu = QuoteNode(Symbol(get_cpu_name()))
-    @eval cpu_name() = Val{$cpu}()
+    if allow_eval
+      @eval cpu_name() = Val{$cpu}()
+    else
+      @warn "Runtime invalidation was disabled, but the CPU info is out-of-date.\nWill continue with incorrect CPU name (from build time)."
+    end
+end
+
+let _cpu = QuoteNode(Symbol(build_cpu_target == "native" ?
+			    get_cpu_name() : build_cpu_target))
+   @eval cpu_name() = Val{$_cpu}()
 end
-define_cpu_name()

src/cpu_info_aarch64.jl

@@ -28,7 +28,7 @@ function _set_sve_vector_width!(bytes = _dynamic_register_size())
 end
 
 
-if _has_aarch64_sve()# && !(Bool(has_feature(Val(:aarch64_sve))))
+if build_cpu_target == "native" && _has_aarch64_sve()# && !(Bool(has_feature(Val(:aarch64_sve))))
     has_feature(::Val{:aarch64_sve_cpuid}) = True()
     _set_sve_vector_width!()
 else
@@ -39,10 +39,20 @@ end
 
 function reset_extra_features!()
     drs = _dynamic_register_size()
-    register_size() ≠ drs && _set_sve_vector_width!(drs)
+    if register_size() ≠ drs
+        if allow_eval
+            _set_sve_vector_width!(drs)
+        else
+            @warn "Runtime invalidation was disabled, but the CPU info is out-of-date.\nWill continue with incorrect CPU register size."
+        end
+    end
     hassve = _has_aarch64_sve()
     if hassve ≠ has_feature(Val(:aarch64_sve_cpuid))
-        @eval has_feature(::Val{:aarch64_sve_cpuid}) = $(Expr(:call, hassve ? :True : :False))
+        if allow_eval
+            @eval has_feature(::Val{:aarch64_sve_cpuid}) = $(Expr(:call, hassve ? :True : :False))
+        else
+            @warn "Runtime invalidation was disabled, but the CPU info is out-of-date.\nWill continue with incorrect CPU feature flag: :aarch64_sve_cpuid."
+        end
     end
 end
 

src/cpu_info_x86.jl

@@ -32,18 +32,26 @@ fast_int64_to_double() = has_feature(Val(:x86_64_avx512dq))
 
 fast_half() = False()
 
-@noinline function setfeaturefalse(s)
+@inline function setfeaturefalse(s)
   if has_feature(Val(s)) === True()
-    @eval has_feature(::Val{$(QuoteNode(s))}) = False()
+    if allow_eval
+      @eval has_feature(::Val{$(QuoteNode(s))}) = False()
+    else
+      @warn "Runtime invalidation was disabled, but the CPU info is out-of-date.\nWill continue with incorrect CPU feature flag: $s."
+    end
   end
 end
-@noinline function setfeaturetrue(s)
+@inline function setfeaturetrue(s)
   if has_feature(Val(s)) === False()
-    @eval has_feature(::Val{$(QuoteNode(s))}) = True()
+    if allow_eval
+      @eval has_feature(::Val{$(QuoteNode(s))}) = True()
+    else
+      @warn "Runtime invalidation was disabled, but the CPU info is out-of-date.\nWill continue with incorrect CPU feature flag: $s."
+    end
   end
 end
 
-function make_generic(target)
+function make_generic_x86(target)
   if occursin("tigerlake", target) || occursin("znver4", target) || occursin("sapphirerapids", target)
     # most feature-complete architectures we use
     setfeaturetrue(:x86_64_avx512ifma)