Merge "Add BNNM intrinsic."

2 files changed, 32 insertions, 0 deletions

diff --git a/rs/java/android/renderscript/RenderScript.java b/rs/java/android/renderscript/RenderScript.java
index 9f3348c..6b1939c 100644
--- a/rs/java/android/renderscript/RenderScript.java
+++ b/rs/java/android/renderscript/RenderScript.java
@@ -951,6 +951,17 @@ public class RenderScript {
         rsnScriptIntrinsicBLAS_Z(mContext, id, func, TransA, TransB, Side, Uplo, Diag, M, N, K, alphaX, alphaY, A, B, betaX, betaY, C, incX, incY, KL, KU);
     }
 
+    native void rsnScriptIntrinsicBLAS_BNNM(long con, long id, int M, int N, int K,
+                                             long A, int a_offset, long B, int b_offset, long C, int c_offset,
+                                             int c_mult_int);
+    synchronized void nScriptIntrinsicBLAS_BNNM(long id, int M, int N, int K,
+                                             long A, int a_offset, long B, int b_offset, long C, int c_offset,
+                                             int c_mult_int) {
+        validate();
+        rsnScriptIntrinsicBLAS_BNNM(mContext, id, M, N, K, A, a_offset, B, b_offset, C, c_offset, c_mult_int);
+    }
+
+
 
     long     mDev;
     long     mContext;
diff --git a/rs/java/android/renderscript/ScriptIntrinsicBLAS.java b/rs/java/android/renderscript/ScriptIntrinsicBLAS.java
index 90d2300..16b7033 100644
--- a/rs/java/android/renderscript/ScriptIntrinsicBLAS.java
+++ b/rs/java/android/renderscript/ScriptIntrinsicBLAS.java
@@ -176,6 +176,9 @@ public final class ScriptIntrinsicBLAS extends ScriptIntrinsic {
     private static final int RsBlas_zherk = 141;
     private static final int RsBlas_zher2k = 142;
 
+    // BLAS extensions start here
+    private static final int RsBlas_bnnm = 1000;
+
     /**
      */
     public static ScriptIntrinsicBLAS create(RenderScript rs) {
@@ -1485,5 +1488,23 @@ public final class ScriptIntrinsicBLAS extends ScriptIntrinsic {
     }
 
 
+    /**
+     *
+     * 8-bit GEMM-like operation for neural networks
+     *
+     * @hide
+     **/
+    public void BNNM(Allocation A, int a_offset, Allocation B, int b_offset, Allocation C, int c_offset, int c_mult) {
+        validateL3(Element.U8(mRS), NO_TRANSPOSE, TRANSPOSE, 0, A, B, C);
+
+        int M = -1, N = -1, K = -1;
+        M = A.getType().getY();
+        N = B.getType().getY();
+        K = A.getType().getX();
+
+
+        mRS.nScriptIntrinsicBLAS_BNNM(getID(mRS), M, N, K, A.getID(mRS), a_offset, B.getID(mRS), b_offset, C.getID(mRS), c_offset, c_mult);
+
+    }
 
 }