feat[fastlanes]: add AVX-512 VBMI transpose with 7.5x speedup

Add AVX-512 VBMI optimized transpose implementation using vpermi2b/vpermb
for vectorized gather and scatter operations.

Performance improvements:
- VBMI: 13.6 cycles/call (7.5x faster than avx512_gfni at 102.6 cycles)
- VBMI: 240x faster than baseline (3276 cycles)

Key optimizations:
- Use vpermi2b to gather 8 bytes at stride-16 in parallel
- Use vpermb for 8x8 byte transpose during scatter phase
- Static permutation tables to avoid stack allocation

Also adds:
- Dual-block transpose_1024x2_avx512 for batch processing
- VBMI detection via has_vbmi() function
- Updated dispatch to prefer VBMI when available

Signed-off-by: Claude <noreply@anthropic.com>

Author: claude

encodings/fastlanes/examples/perf_transpose.rs

@@ -63,6 +63,8 @@ fn run_all_benchmarks(input: &[u8; 128], output: &mut [u8; 128]) {
         "avx2",
         "avx2_gfni",
         "avx512_gfni",
+        "avx512_vbmi",
+        "avx512_dual",
     ];
 
     for mode in &modes {
@@ -119,6 +121,31 @@ fn run_benchmark(mode: &str, input: &[u8; 128], output: &mut [u8; 128]) {
                     }
                 }
             }
+            #[cfg(target_arch = "x86_64")]
+            "avx512_vbmi" => {
+                use vortex_fastlanes::transpose::x86;
+                if x86::has_vbmi() {
+                    unsafe {
+                        x86::transpose_1024_vbmi(black_box(input), black_box(output));
+                    }
+                }
+            }
+            #[cfg(target_arch = "x86_64")]
+            "avx512_dual" => {
+                use vortex_fastlanes::transpose::x86;
+                if x86::has_avx512() {
+                    let input2 = *input;
+                    let mut output2 = [0u8; 128];
+                    unsafe {
+                        x86::transpose_1024x2_avx512(
+                            black_box(input),
+                            black_box(&input2),
+                            black_box(output),
+                            black_box(&mut output2),
+                        );
+                    }
+                }
+            }
             _ => {}
         }
     }
@@ -198,6 +225,42 @@ fn run_benchmark(mode: &str, input: &[u8; 128], output: &mut [u8; 128]) {
                 return;
             }
         }
+        #[cfg(target_arch = "x86_64")]
+        "avx512_vbmi" => {
+            use vortex_fastlanes::transpose::x86;
+            if x86::has_vbmi() {
+                for _ in 0..MEASURE_ITERATIONS {
+                    unsafe {
+                        x86::transpose_1024_vbmi(black_box(input), black_box(output));
+                    }
+                }
+            } else {
+                println!("{:15} AVX-512 VBMI not available", mode);
+                return;
+            }
+        }
+        #[cfg(target_arch = "x86_64")]
+        "avx512_dual" => {
+            use vortex_fastlanes::transpose::x86;
+            if x86::has_avx512() {
+                let input2 = *input;
+                let mut output2 = [0u8; 128];
+                // Note: we do MEASURE_ITERATIONS/2 since each call processes 2 blocks
+                for _ in 0..MEASURE_ITERATIONS / 2 {
+                    unsafe {
+                        x86::transpose_1024x2_avx512(
+                            black_box(input),
+                            black_box(&input2),
+                            black_box(output),
+                            black_box(&mut output2),
+                        );
+                    }
+                }
+            } else {
+                println!("{:15} AVX-512 not available", mode);
+                return;
+            }
+        }
         _ => {
             println!("Unknown mode: {}", mode);
             return;