feat(asm): byte-identical go-flac AVX2 assembly with scalar families and jump relaxation

Assisted-by: Qwen 3.8 Max Preview
2026-07-08 12:51:35 +02:00
parent 39870f91f6
commit a82f575aee
12 changed files with 662 additions and 73 deletions
@@ -244,3 +244,76 @@ TEXT ·hsum(SB), NOSPLIT, $0
 		t.Errorf("VEX kernel mismatch:\n got:  %s\n want: %s", hexBytes(code), hexBytes(want))
 	}
 }
+
+// TestAssembleShortJumps checks that a tight loop settles on the short (rel8)
+// jump forms, byte for byte with the Go assembler.
+func TestAssembleShortJumps(t *testing.T) {
+	fn := firstText(t, `
+#include "textflag.h"
+TEXT ·loop(SB), NOSPLIT, $0
+	XORQ AX, AX
+l1:
+	ADDQ $1, AX
+	CMPQ AX, $10
+	JLT l1
+	RET
+`)
+	code, _, err := Assemble(fn)
+	if err != nil {
+		t.Fatalf("Assemble: %v", err)
+	}
+	// From the Go-assembled function:
+	//   XORQ AX, AX       4831c0
+	//   ADDQ $1, AX       4883c001
+	//   CMPQ AX, $10      4883f80a
+	//   JLT l1            7cf6      (short, rel8)
+	//   RET               c3
+	want := []byte{
+		0x48, 0x31, 0xc0,
+		0x48, 0x83, 0xc0, 0x01,
+		0x48, 0x83, 0xf8, 0x0a,
+		0x7c, 0xf6,
+		0xc3,
+	}
+	if hexBytes(code) != hexBytes(want) {
+		t.Errorf("short-jump mismatch:\n got:  %s\n want: %s", hexBytes(code), hexBytes(want))
+	}
+}
+
+// TestAssembleJumpFolding checks jump-to-jump folding: a conditional jump to a
+// label that only holds an unconditional jump is redirected to the ultimate
+// target, exactly as the Go toolchain does before it encodes branches.
+func TestAssembleJumpFolding(t *testing.T) {
+	fn := firstText(t, `
+#include "textflag.h"
+TEXT ·fold(SB), NOSPLIT, $0
+	XORQ AX, AX
+	JGE done
+	INCQ AX
+done:
+	JMP end
+end:
+	RET
+`)
+	code, _, err := Assemble(fn)
+	if err != nil {
+		t.Fatalf("Assemble: %v", err)
+	}
+	// From the Go-assembled function: the JGE skips past the done: trampoline
+	// straight to end:
+	//   XORQ AX, AX       4831c0
+	//   JGE end           7d05      (folded past done)
+	//   INCQ AX           48ffc0
+	//   JMP end           eb00
+	//   RET               c3
+	want := []byte{
+		0x48, 0x31, 0xc0,
+		0x7d, 0x05,
+		0x48, 0xff, 0xc0,
+		0xeb, 0x00,
+		0xc3,
+	}
+	if hexBytes(code) != hexBytes(want) {
+		t.Errorf("jump-folding mismatch:\n got:  %s\n want: %s", hexBytes(code), hexBytes(want))
+	}
+}