add Float16 math functions

[simeonschaub]

also fixes #374 by simply removing that device override

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 79.01%. Comparing base (a6f878a) to head (e718ea5).
⚠️ Report is 1 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #368   +/-   ##
=======================================
  Coverage   79.01%   79.01%           
=======================================
  Files          12       12           
  Lines         672      672           
=======================================
  Hits          531      531           
  Misses        141      141           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[simeonschaub]

Hmm, it looks like pocl's Float16 support is incomplete and missing some of these functions. Should we not run the GPUCompiler testsuite on Float16 for now? We could only add those intrinsics pocl supports, but rusticl for example seems to support these.

[github-actions]

Your PR requires formatting changes to meet the project's style guidelines.
Please consider running Runic (git runic master) to apply these changes.

Click here to view the suggested changes.

diff --git a/lib/intrinsics/src/math.jl b/lib/intrinsics/src/math.jl
index 39117bb..a769499 100644
--- a/lib/intrinsics/src/math.jl
+++ b/lib/intrinsics/src/math.jl
@@ -33,7 +33,7 @@ for gentype in generic_types
 
 @device_override Base.cos(x::$gentype) = @builtin_ccall("cos", $gentype, ($gentype,), x)
 @device_override Base.cosh(x::$gentype) = @builtin_ccall("cosh", $gentype, ($gentype,), x)
-@device_override Base.cospi(x::$gentype) = @builtin_ccall("cospi", $gentype, ($gentype,), x)
+        @device_override Base.cospi(x::$gentype) = @builtin_ccall("cospi", $gentype, ($gentype,), x)
 
 @device_override SpecialFunctions.erfc(x::$gentype) = @builtin_ccall("erfc", $gentype, ($gentype,), x)
 @device_override SpecialFunctions.erf(x::$gentype) = @builtin_ccall("erf", $gentype, ($gentype,), x)
@@ -59,10 +59,10 @@ for gentype in generic_types
 #@device_override Base.mod(x::$gentype, y::$gentype) = @builtin_ccall("fmod", $gentype, ($gentype, $gentype), x, y)
 # fract(x::$gentype, $gentype *iptr) = @builtin_ccall("fract", $gentype, ($gentype, $gentype *), x, iptr)
 
-# TODO: remove once https://github.com/pocl/pocl/issues/2034 is addressed
-if $gentype != Float16
-    @device_override Base.hypot(x::$gentype, y::$gentype) = @builtin_ccall("hypot", $gentype, ($gentype, $gentype), x, y)
-end
+        # TODO: remove once https://github.com/pocl/pocl/issues/2034 is addressed
+        if $gentype != Float16
+            @device_override Base.hypot(x::$gentype, y::$gentype) = @builtin_ccall("hypot", $gentype, ($gentype, $gentype), x, y)
+        end
 
 @device_override SpecialFunctions.loggamma(x::$gentype) = @builtin_ccall("lgamma", $gentype, ($gentype,), x)
 
@@ -101,13 +101,13 @@ end
     return sinval, cosval[]
 end
 @device_override Base.sinh(x::$gentype) = @builtin_ccall("sinh", $gentype, ($gentype,), x)
-@device_override Base.sinpi(x::$gentype) = @builtin_ccall("sinpi", $gentype, ($gentype,), x)
+        @device_override Base.sinpi(x::$gentype) = @builtin_ccall("sinpi", $gentype, ($gentype,), x)
 
 @device_override Base.sqrt(x::$gentype) = @builtin_ccall("sqrt", $gentype, ($gentype,), x)
 
 @device_override Base.tan(x::$gentype) = @builtin_ccall("tan", $gentype, ($gentype,), x)
 @device_override Base.tanh(x::$gentype) = @builtin_ccall("tanh", $gentype, ($gentype,), x)
-@device_override Base.tanpi(x::$gentype) = @builtin_ccall("tanpi", $gentype, ($gentype,), x)
+        @device_override Base.tanpi(x::$gentype) = @builtin_ccall("tanpi", $gentype, ($gentype,), x)
 
 @device_override SpecialFunctions.gamma(x::$gentype) = @builtin_ccall("tgamma", $gentype, ($gentype,), x)
 
diff --git a/test/intrinsics.jl b/test/intrinsics.jl
index ad970c0..d2a9c48 100644
--- a/test/intrinsics.jl
+++ b/test/intrinsics.jl
@@ -63,105 +63,105 @@ cl.memory_backend() isa cl.SVMBackend && @on_device atomic_work_item_fence(OpenC
 
 end
 
-@testset "math" begin
-
-@testset "unary - $T" for T in float_types
-    @testset "$f" for f in [
-            acos, acosh,
-            asin, asinh,
-            atan, atanh,
-            cbrt,
-            ceil,
-            cos, cosh, cospi,
-            exp, exp2, exp10, expm1,
-            abs,
-            floor,
-            log, log2, log10, log1p,
-            round,
-            sin, sinh, sinpi,
-            sqrt,
-            tan, tanh, tanpi,
-            trunc,
-        ]
-        x = rand(T)
-        if f == acosh
-            x += 1
+    @testset "math" begin
+
+        @testset "unary - $T" for T in float_types
+            @testset "$f" for f in [
+                    acos, acosh,
+                    asin, asinh,
+                    atan, atanh,
+                    cbrt,
+                    ceil,
+                    cos, cosh, cospi,
+                    exp, exp2, exp10, expm1,
+                    abs,
+                    floor,
+                    log, log2, log10, log1p,
+                    round,
+                    sin, sinh, sinpi,
+                    sqrt,
+                    tan, tanh, tanpi,
+                    trunc,
+                ]
+                x = rand(T)
+                if f == acosh
+                    x += 1
+                end
+                broken = ispocl && T == Float16 && f in [acosh, asinh, atanh, cbrt, cospi, expm1, log1p, sinpi, tanpi]
+                @test call_on_device(f, x) ≈ f(x) broken = broken
+            end
         end
-        broken = ispocl && T == Float16 && f in [acosh, asinh, atanh, cbrt, cospi, expm1, log1p, sinpi, tanpi]
-        @test call_on_device(f, x) ≈ f(x) broken = broken
-    end
-end
 
-@testset "binary - $T" for T in float_types
-    @testset "$f" for f in [
-            atan,
-            copysign,
-            max,
-            min,
-            hypot,
-            (^),
-        ]
-        x = rand(T)
-        y = rand(T)
-        broken = ispocl && T == Float16 && f == atan
-        @test call_on_device(f, x, y) ≈ f(x, y) broken = broken
-    end
-end
+        @testset "binary - $T" for T in float_types
+            @testset "$f" for f in [
+                    atan,
+                    copysign,
+                    max,
+                    min,
+                    hypot,
+                    (^),
+                ]
+                x = rand(T)
+                y = rand(T)
+                broken = ispocl && T == Float16 && f == atan
+                @test call_on_device(f, x, y) ≈ f(x, y) broken = broken
+            end
+        end
 
-@testset "ternary - $T" for T in float_types
-    @testset "$f" for f in [
-            fma,
-        ]
-        x = rand(T)
-        y = rand(T)
-        z = rand(T)
-        @test call_on_device(f, x, y, z) ≈ f(x, y, z)
-    end
-end
+        @testset "ternary - $T" for T in float_types
+            @testset "$f" for f in [
+                    fma,
+                ]
+                x = rand(T)
+                y = rand(T)
+                z = rand(T)
+                @test call_on_device(f, x, y, z) ≈ f(x, y, z)
+            end
+        end
 
-@testset "OpenCL-specific unary - $T" for T in float_types
-    @testset "$f" for f in [
-            OpenCL.acospi,
-            OpenCL.asinpi,
-            OpenCL.atanpi,
-            OpenCL.logb,
-            OpenCL.rint,
-            OpenCL.rsqrt,
-        ]
-        x = rand(T)
-        broken = ispocl && T == Float16 && !(f in [OpenCL.rint, OpenCL.rsqrt])
-        @test call_on_device(f, x) isa Real broken = broken  # Just check it doesn't error
-    end
-    broken = ispocl && T == Float16
-    @test call_on_device(OpenCL.ilogb, T(8.0)) isa Int32 broken = broken
-    @test call_on_device(OpenCL.nan, Base.uinttype(T)(0)) isa T
-end
+        @testset "OpenCL-specific unary - $T" for T in float_types
+            @testset "$f" for f in [
+                    OpenCL.acospi,
+                    OpenCL.asinpi,
+                    OpenCL.atanpi,
+                    OpenCL.logb,
+                    OpenCL.rint,
+                    OpenCL.rsqrt,
+                ]
+                x = rand(T)
+                broken = ispocl && T == Float16 && !(f in [OpenCL.rint, OpenCL.rsqrt])
+                @test call_on_device(f, x) isa Real broken = broken  # Just check it doesn't error
+            end
+            broken = ispocl && T == Float16
+            @test call_on_device(OpenCL.ilogb, T(8.0)) isa Int32 broken = broken
+            @test call_on_device(OpenCL.nan, Base.uinttype(T)(0)) isa T
+        end
 
-@testset "OpenCL-specific binary - $T" for T in float_types
-    @testset "$f" for f in [
-            OpenCL.atanpi,
-            OpenCL.dim,
-            OpenCL.maxmag,
-            OpenCL.minmag,
-            OpenCL.nextafter,
-            OpenCL.powr,
-        ]
-        x = rand(T)
-        y = rand(T)
-        broken = ispocl && T == Float16 && !(f in [OpenCL.maxmag, OpenCL.minmag])
-        @test call_on_device(f, x, y) isa Real broken = broken  # Just check it doesn't error
-    end
-    broken = ispocl && T == Float16
-    @test call_on_device(OpenCL.rootn, T(8.0), Int32(3)) ≈ T(2.0) broken = broken
-end
+        @testset "OpenCL-specific binary - $T" for T in float_types
+            @testset "$f" for f in [
+                    OpenCL.atanpi,
+                    OpenCL.dim,
+                    OpenCL.maxmag,
+                    OpenCL.minmag,
+                    OpenCL.nextafter,
+                    OpenCL.powr,
+                ]
+                x = rand(T)
+                y = rand(T)
+                broken = ispocl && T == Float16 && !(f in [OpenCL.maxmag, OpenCL.minmag])
+                @test call_on_device(f, x, y) isa Real broken = broken  # Just check it doesn't error
+            end
+            broken = ispocl && T == Float16
+            @test call_on_device(OpenCL.rootn, T(8.0), Int32(3)) ≈ T(2.0) broken = broken
+        end
 
-@testset "OpenCL-specific ternary - $T" for T in float_types
-    x = rand(T)
-    y = rand(T)
-    z = rand(T)
-    @test call_on_device(OpenCL.mad, x, y, z) ≈ x * y + z
-end
+        @testset "OpenCL-specific ternary - $T" for T in float_types
+            x = rand(T)
+            y = rand(T)
+            z = rand(T)
+            @test call_on_device(OpenCL.mad, x, y, z) ≈ x * y + z
+        end
 
-end
+    end
 
 end

[simeonschaub]

Ok, I think I am happy with this for now. The intrinsics missing in pocl are tracked in pocl/pocl#2034