1 files changed, 44 insertions, 0 deletions
diff --git a/riscv/insns/vaesdm_vs.h b/riscv/insns/vaesdm_vs.h
new file mode 100644
index 0000000..3c23e69
--- /dev/null
+++ b/riscv/insns/vaesdm_vs.h
@@ -0,0 +1,44 @@
+// vaesdm.vs vd, vs2
+
+#include "zvkned_ext_macros.h"
+#include "zvk_ext_macros.h"
+
+require_vaes_vs_constraints;
+
+VI_ZVK_VD_VS2_NOOPERANDS_PRELOOP_EGU32x4_NOVM_LOOP(
+  {},
+  // This statement will be executed before the first execution
+  // of the loop, and only if the loop is going to be entered.
+  // We cannot use a block ( { ... } ) since we want the variables declared
+  // here to be visible in the loop block.
+  // We capture the "scalar", vs2's first element, by copy, even though
+  // the "no overlap" constraint means that vs2 should remain constant
+  // during the loop.
+  const EGU8x16_t scalar_key = P.VU.elt_group<EGU8x16_t>(vs2_num, 0);,
+  {
+    // For AES128, AES192, or AES256, state and key are 128b/16B values:
+    //  - vd in contains the input state,
+    //  - vs2 contains the input round key,
+    //  - vd out receives the output state.
+    //
+    // While the spec calls for handling the vector as made of EGU32x4
+    // element groups (i.e., 4 uint32_t), it is convenient to treat
+    // AES state and key as EGU8x16 (i.e., 16 uint8_t). This is why
+    // we extract the operands here instead of using the existing LOOP
+    // macro that defines/extracts the operand variables as EGU32x4.
+    EGU8x16_t aes_state = P.VU.elt_group<EGU8x16_t>(vd_num, idx_eg);
+
+    // InvShiftRows - Rotate each row bytes by 0, 1, 2, 3 positions.
+    VAES_INV_SHIFT_ROWS(aes_state);
+    // InvSubBytes - Apply S-box to every byte in the state
+    VAES_INV_SUB_BYTES(aes_state);
+    // AddRoundKey (which is also InvAddRoundKey as it's xor)
+    EGU8x16_XOREQ(aes_state, scalar_key);
+    // InvMixColumns
+    VAES_INV_MIX_COLUMNS(aes_state);
+
+    // Update the destination register.
+    EGU8x16_t &vd = P.VU.elt_group<EGU8x16_t>(vd_num, idx_eg, true);
+    EGU8x16_COPY(vd, aes_state);
+  }
+);