Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit 4c472745 authored by Miao Wang's avatar Miao Wang
Browse files

[RenderScript] Fix HEMM and SYRK in IntrinscBLAS to correctly handle 

complex 'Alpha' and 'Beta'

Change-Id: I0aed57d41b0199d5c6dd3c31ca33644a5634dc35
parent b5665c99

api/current.txt

+4 −4
Original line number Diff line number Diff line
@@ -28039,12 +28039,12 @@ package android.renderscript { 
  public final class ScriptIntrinsicBLAS extends android.renderscript.ScriptIntrinsic {

    method public void BNNM(android.renderscript.Allocation, int, android.renderscript.Allocation, int, android.renderscript.Allocation, int, int);

    method public void CGEMM(int, int, android.renderscript.Float2, android.renderscript.Allocation, android.renderscript.Allocation, android.renderscript.Float2, android.renderscript.Allocation);

    method public void CHEMM(int, int, float, android.renderscript.Allocation, android.renderscript.Allocation, float, android.renderscript.Allocation);

    method public void CHEMM(int, int, android.renderscript.Float2, android.renderscript.Allocation, android.renderscript.Allocation, android.renderscript.Float2, android.renderscript.Allocation);

    method public void CHER2K(int, int, android.renderscript.Float2, android.renderscript.Allocation, android.renderscript.Allocation, float, android.renderscript.Allocation);

    method public void CHERK(int, int, float, android.renderscript.Allocation, float, android.renderscript.Allocation);

    method public void CSYMM(int, int, android.renderscript.Float2, android.renderscript.Allocation, android.renderscript.Allocation, android.renderscript.Float2, android.renderscript.Allocation);

    method public void CSYR2K(int, int, android.renderscript.Float2, android.renderscript.Allocation, android.renderscript.Allocation, android.renderscript.Float2, android.renderscript.Allocation);

    method public void CSYRK(int, int, float, float, android.renderscript.Allocation, float, float, android.renderscript.Allocation);

    method public void CSYRK(int, int, android.renderscript.Float2, android.renderscript.Allocation, android.renderscript.Float2, android.renderscript.Allocation);

    method public void CTRMM(int, int, int, int, android.renderscript.Float2, android.renderscript.Allocation, android.renderscript.Allocation);

    method public void CTRSM(int, int, int, int, android.renderscript.Float2, android.renderscript.Allocation, android.renderscript.Allocation);

    method public void DGEMM(int, int, double, android.renderscript.Allocation, android.renderscript.Allocation, double, android.renderscript.Allocation);
@@ -28060,12 +28060,12 @@ package android.renderscript { 
    method public void STRMM(int, int, int, int, float, android.renderscript.Allocation, android.renderscript.Allocation);

    method public void STRSM(int, int, int, int, float, android.renderscript.Allocation, android.renderscript.Allocation);

    method public void ZGEMM(int, int, android.renderscript.Double2, android.renderscript.Allocation, android.renderscript.Allocation, android.renderscript.Double2, android.renderscript.Allocation);

    method public void ZHEMM(int, int, double, android.renderscript.Allocation, android.renderscript.Allocation, double, android.renderscript.Allocation);

    method public void ZHEMM(int, int, android.renderscript.Double2, android.renderscript.Allocation, android.renderscript.Allocation, android.renderscript.Double2, android.renderscript.Allocation);

    method public void ZHER2K(int, int, android.renderscript.Double2, android.renderscript.Allocation, android.renderscript.Allocation, double, android.renderscript.Allocation);

    method public void ZHERK(int, int, double, android.renderscript.Allocation, double, android.renderscript.Allocation);

    method public void ZSYMM(int, int, android.renderscript.Double2, android.renderscript.Allocation, android.renderscript.Allocation, android.renderscript.Double2, android.renderscript.Allocation);

    method public void ZSYR2K(int, int, android.renderscript.Double2, android.renderscript.Allocation, android.renderscript.Allocation, android.renderscript.Double2, android.renderscript.Allocation);

    method public void ZSYRK(int, int, double, double, android.renderscript.Allocation, double, double, android.renderscript.Allocation);

    method public void ZSYRK(int, int, android.renderscript.Double2, android.renderscript.Allocation, android.renderscript.Double2, android.renderscript.Allocation);

    method public void ZTRMM(int, int, int, int, android.renderscript.Double2, android.renderscript.Allocation, android.renderscript.Allocation);

    method public void ZTRSM(int, int, int, int, android.renderscript.Double2, android.renderscript.Allocation, android.renderscript.Allocation);

    method public static android.renderscript.ScriptIntrinsicBLAS create(android.renderscript.RenderScript);

api/system-current.txt

+4 −4
Original line number Diff line number Diff line
@@ -30042,12 +30042,12 @@ package android.renderscript { 
  public final class ScriptIntrinsicBLAS extends android.renderscript.ScriptIntrinsic {

    method public void BNNM(android.renderscript.Allocation, int, android.renderscript.Allocation, int, android.renderscript.Allocation, int, int);

    method public void CGEMM(int, int, android.renderscript.Float2, android.renderscript.Allocation, android.renderscript.Allocation, android.renderscript.Float2, android.renderscript.Allocation);

    method public void CHEMM(int, int, float, android.renderscript.Allocation, android.renderscript.Allocation, float, android.renderscript.Allocation);

    method public void CHEMM(int, int, android.renderscript.Float2, android.renderscript.Allocation, android.renderscript.Allocation, android.renderscript.Float2, android.renderscript.Allocation);

    method public void CHER2K(int, int, android.renderscript.Float2, android.renderscript.Allocation, android.renderscript.Allocation, float, android.renderscript.Allocation);

    method public void CHERK(int, int, float, android.renderscript.Allocation, float, android.renderscript.Allocation);

    method public void CSYMM(int, int, android.renderscript.Float2, android.renderscript.Allocation, android.renderscript.Allocation, android.renderscript.Float2, android.renderscript.Allocation);

    method public void CSYR2K(int, int, android.renderscript.Float2, android.renderscript.Allocation, android.renderscript.Allocation, android.renderscript.Float2, android.renderscript.Allocation);

    method public void CSYRK(int, int, float, float, android.renderscript.Allocation, float, float, android.renderscript.Allocation);

    method public void CSYRK(int, int, android.renderscript.Float2, android.renderscript.Allocation, android.renderscript.Float2, android.renderscript.Allocation);

    method public void CTRMM(int, int, int, int, android.renderscript.Float2, android.renderscript.Allocation, android.renderscript.Allocation);

    method public void CTRSM(int, int, int, int, android.renderscript.Float2, android.renderscript.Allocation, android.renderscript.Allocation);

    method public void DGEMM(int, int, double, android.renderscript.Allocation, android.renderscript.Allocation, double, android.renderscript.Allocation);
@@ -30063,12 +30063,12 @@ package android.renderscript { 
    method public void STRMM(int, int, int, int, float, android.renderscript.Allocation, android.renderscript.Allocation);

    method public void STRSM(int, int, int, int, float, android.renderscript.Allocation, android.renderscript.Allocation);

    method public void ZGEMM(int, int, android.renderscript.Double2, android.renderscript.Allocation, android.renderscript.Allocation, android.renderscript.Double2, android.renderscript.Allocation);

    method public void ZHEMM(int, int, double, android.renderscript.Allocation, android.renderscript.Allocation, double, android.renderscript.Allocation);

    method public void ZHEMM(int, int, android.renderscript.Double2, android.renderscript.Allocation, android.renderscript.Allocation, android.renderscript.Double2, android.renderscript.Allocation);

    method public void ZHER2K(int, int, android.renderscript.Double2, android.renderscript.Allocation, android.renderscript.Allocation, double, android.renderscript.Allocation);

    method public void ZHERK(int, int, double, android.renderscript.Allocation, double, android.renderscript.Allocation);

    method public void ZSYMM(int, int, android.renderscript.Double2, android.renderscript.Allocation, android.renderscript.Allocation, android.renderscript.Double2, android.renderscript.Allocation);

    method public void ZSYR2K(int, int, android.renderscript.Double2, android.renderscript.Allocation, android.renderscript.Allocation, android.renderscript.Double2, android.renderscript.Allocation);

    method public void ZSYRK(int, int, double, double, android.renderscript.Allocation, double, double, android.renderscript.Allocation);

    method public void ZSYRK(int, int, android.renderscript.Double2, android.renderscript.Allocation, android.renderscript.Double2, android.renderscript.Allocation);

    method public void ZTRMM(int, int, int, int, android.renderscript.Double2, android.renderscript.Allocation, android.renderscript.Allocation);

    method public void ZTRSM(int, int, int, int, android.renderscript.Double2, android.renderscript.Allocation, android.renderscript.Allocation);

    method public static android.renderscript.ScriptIntrinsicBLAS create(android.renderscript.RenderScript);

rs/java/android/renderscript/ScriptIntrinsicBLAS.java

+8 −8
Original line number Diff line number Diff line
@@ -1145,7 +1145,7 @@ public final class ScriptIntrinsicBLAS extends ScriptIntrinsic { 
        }

        mRS.nScriptIntrinsicBLAS_Double(getID(mRS), RsBlas_dsyrk, Trans, 0, 0, Uplo, 0, 0, C.getType().getX(), K, alpha, A.getID(mRS), 0, beta, C.getID(mRS), 0, 0, 0, 0);

    }

    public void CSYRK(@Uplo int Uplo, @Transpose int Trans, float alphaX, float alphaY, Allocation A, float betaX, float betaY, Allocation C) {

    public void CSYRK(@Uplo int Uplo, @Transpose int Trans, Float2 alpha, Allocation A, Float2 beta, Allocation C) {

        validateTranspose(Trans);

        validateUplo(Uplo);

        validateL3(Element.F32_2(mRS), Trans, 0, 0, A, null, C);
@@ -1155,10 +1155,10 @@ public final class ScriptIntrinsicBLAS extends ScriptIntrinsic { 
        } else {

            K = A.getType().getX();

        }

        mRS.nScriptIntrinsicBLAS_Complex(getID(mRS), RsBlas_csyrk, Trans, 0, 0, Uplo, 0, 0, C.getType().getX(), K, alphaX, alphaY, A.getID(mRS), 0, betaX, betaY,

        mRS.nScriptIntrinsicBLAS_Complex(getID(mRS), RsBlas_csyrk, Trans, 0, 0, Uplo, 0, 0, C.getType().getX(), K, alpha.x, alpha.y, A.getID(mRS), 0, beta.x, beta.y,

                                         C.getID(mRS), 0, 0, 0, 0);

    }

    public void ZSYRK(@Uplo int Uplo, @Transpose int Trans, double alphaX, double alphaY, Allocation A, double betaX, double betaY, Allocation C) {

    public void ZSYRK(@Uplo int Uplo, @Transpose int Trans, Double2 alpha, Allocation A, Double2 beta, Allocation C) {

        validateTranspose(Trans);

        validateUplo(Uplo);

        validateL3(Element.F64_2(mRS), Trans, 0, 0, A, null, C);
@@ -1168,7 +1168,7 @@ public final class ScriptIntrinsicBLAS extends ScriptIntrinsic { 
        } else {

            K = A.getType().getX();

        }

        mRS.nScriptIntrinsicBLAS_Z(getID(mRS), RsBlas_zsyrk, Trans, 0, 0, Uplo, 0, 0, C.getType().getX(), K, alphaX, alphaY, A.getID(mRS), 0, betaX, betaY,

        mRS.nScriptIntrinsicBLAS_Z(getID(mRS), RsBlas_zsyrk, Trans, 0, 0, Uplo, 0, 0, C.getType().getX(), K, alpha.x, alpha.y, A.getID(mRS), 0, beta.x, beta.y,

                                   C.getID(mRS), 0, 0, 0, 0);

    }
@@ -1379,17 +1379,17 @@ public final class ScriptIntrinsicBLAS extends ScriptIntrinsic { 
            throw new RSRuntimeException("Called HEMM with mismatched B and C");

        }

    }

    public void CHEMM(@Side int Side, @Uplo int Uplo, float alpha, Allocation A, Allocation B, float beta, Allocation C) {

    public void CHEMM(@Side int Side, @Uplo int Uplo, Float2 alpha, Allocation A, Allocation B, Float2 beta, Allocation C) {

        validateUplo(Uplo);

        validateHEMM(Element.F32_2(mRS), Side, A, B, C);

        mRS.nScriptIntrinsicBLAS_Complex(getID(mRS), RsBlas_chemm, 0, 0, Side, Uplo, 0, C.getType().getY(), C.getType().getX(), 0,

                                         alpha, 0, A.getID(mRS), B.getID(mRS), beta, 0, C.getID(mRS), 0, 0, 0, 0);

                                         alpha.x, alpha.y, A.getID(mRS), B.getID(mRS), beta.x, beta.y, C.getID(mRS), 0, 0, 0, 0);

    }

    public void ZHEMM(@Side int Side, @Uplo int Uplo, double alpha, Allocation A, Allocation B, double beta, Allocation C) {

    public void ZHEMM(@Side int Side, @Uplo int Uplo, Double2 alpha, Allocation A, Allocation B, Double2 beta, Allocation C) {

        validateUplo(Uplo);

        validateHEMM(Element.F32_2(mRS), Side, A, B, C);

        mRS.nScriptIntrinsicBLAS_Z(getID(mRS), RsBlas_zhemm, 0, 0, Side, Uplo, 0, C.getType().getY(), C.getType().getX(), 0,

                                   alpha, 0, A.getID(mRS), B.getID(mRS), beta, 0, C.getID(mRS), 0, 0, 0, 0);

                                   alpha.x, alpha.y, A.getID(mRS), B.getID(mRS), beta.x, beta.y, C.getID(mRS), 0, 0, 0, 0);

    }



    static void validateHERK(Element e, @Transpose int Trans, Allocation A, Allocation C) {