// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
// SPDX-License-Identifier: MIT

package linalg

import (
	"sourcedock.dev/petrbalvin/tensor/internal/core"
	"testing"

	"sourcedock.dev/petrbalvin/tensor/internal/engine"
)

// TestMatMul2DFloat32ScratchCleaned pins the regression where the
// pooled accumulation buffer was handed back dirty: leftover values
// from an earlier borrower used to leak into every float32 product sum,
// making results depend on allocation history instead of inputs alone.
func TestMatMul2DFloat32ScratchCleaned(t *testing.T) {
	const size = 35
	warm := engine.GetFloat64Buf(size)
	for i := range warm {
		warm[i] = 1.5 // leave obvious residue in the pooled window
	}
	engine.PutFloat64Buf(warm)

	a := mustFromFloat32s(t, []float32{1, 2, 3, 4}, 2, 2)
	b := mustFromFloat32s(t, []float32{0.5, 0, 2, 1}, 2, 2)
	want := []float32{4.5, 2, 9.5, 4}

	for run := range 2 {
		got, err := core.MatMul2D(a, b)
		if err != nil {
			t.Fatal(err)
		}
		for i := range want {
			if g := got.RawFloat32s()[i]; g != want[i] {
				t.Fatalf("run %d: c[%d] = %v, want %v", run, i, g, want[i])
			}
		}
	}
}