diff --git a/test/core/rocarray_broadcast.jl b/test/core/rocarray_broadcast.jl
index cbe859e38..362f12e17 100644
--- a/test/core/rocarray_broadcast.jl
+++ b/test/core/rocarray_broadcast.jl
@@ -45,28 +45,27 @@ end
 end
 
 # https://github.com/JuliaGPU/AMDGPU.jl/issues/1002
-if Base.datatype_alignment(Int128) == 16
-    # note: miscompile only occurs with `julia --check-bounds=yes`
-    @testset "Int128 miscompilation" begin
-        function test_kernel(a::T, b) where T
-            c = a
-            for i=1:5
-                c += T(b)
-                c = a * T(2)
-            end
-            return c
+# note: miscompile only occurs with `julia --check-bounds=yes`
+@testset "Int128 miscompilation" begin
+    function test_kernel(a::T, b) where T
+        c = a
+        for i=1:5
+            c += T(b)
+            c = a * T(2)
         end
-        M = rand(Int128, 10, 10)
-        @test Array(test_kernel.(ROCArray(M), Int128(10))) == test_kernel.(M, Int128(10))
-        AMDGPU.synchronize()
+        return c
     end
+    M = rand(Int128, 10, 10)
+    @test Array(test_kernel.(ROCArray(M), Int128(10))) == test_kernel.(M, Int128(10))
+    AMDGPU.synchronize()
+end
 
-    @testset "Int128 axpby! miscompilation" begin
-        a, b = rand(Int128), rand(Int128)
-        x, y = rand(Int128, 5), rand(Int128, 5)
-        gx, gy = ROCArray(x), ROCArray(y)
-        gy .= gx .* a .+ gy .* b
-        @test Array(gy) == x .* a .+ y .* b
-        AMDGPU.synchronize()
-    end
+# https://github.com/JuliaGPU/AMDGPU.jl/issues/1002
+@testset "Int128 axpby! miscompilation" begin
+    a, b = rand(Int128), rand(Int128)
+    x, y = rand(Int128, 5), rand(Int128, 5)
+    gx, gy = ROCArray(x), ROCArray(y)
+    gy .= gx .* a .+ gy .* b
+    @test Array(gy) == x .* a .+ y .* b
+    AMDGPU.synchronize()
 end
diff --git a/test/device/launch.jl b/test/device/launch.jl
index e1eeabd91..e3725e705 100644
--- a/test/device/launch.jl
+++ b/test/device/launch.jl
@@ -60,6 +60,25 @@ end
     AMDGPU.synchronize()
 end
 
+# https://github.com/JuliaGPU/AMDGPU.jl/issues/1002
+@testset "Int128 layout matches the host" begin
+    # a kernel argument with fields after an Int128
+    function kernel(out, s)
+        out[1] = s[1]
+        out[2] = s[2]
+        out[3] = s[3]
+        return
+    end
+    s = (Int32(-7), typemax(Int128) - 3, Int64(42))
+    out = ROCArray{Int128}(undef, 3)
+    @roc kernel(out, s)
+    @test Array(out) == [-7, typemax(Int128) - 3, 42]
+
+    # device memory holding such values, as laid out by the host
+    xs = [(Int32(i), Int128(i) << 70, Int64(-i)) for i in 1:4]
+    @test Array(map(x -> x[1] + x[2] + x[3], ROCArray(xs))) == map(x -> x[1] + x[2] + x[3], xs)
+end
+
 @testset "Function/Argument Conversion" begin
     @testset "Closure as Argument" begin
         function kernel(closure)
diff --git a/test/runtests.jl b/test/runtests.jl
index 1888f60f0..6899ba7b0 100644
--- a/test/runtests.jl
+++ b/test/runtests.jl
@@ -80,9 +80,7 @@ init_code = quote
     include($gpuarrays_testsuite)
     testf(f, xs...; kwargs...) = TestSuite.compare(f, AMDGPU.ROCArray, xs...; kwargs...)
 
-    # Int128 kernel arguments need the host to align Int128 to 16 bytes like the device (Julia 1.12+), see #1002
-    const eltypes = [Int16, Int32, Int64,
-                     (Base.datatype_alignment(Int128) == 16 ? (Int128,) : ())...,
+    const eltypes = [Int16, Int32, Int64, Int128,
                      Float16, Float32, Float64,
                      ComplexF16, ComplexF32, ComplexF64,
                      Complex{Int16}, Complex{Int32}, Complex{Int64}]