diff --git a/gm/shadermaskfilter.cpp b/gm/shadermaskfilter.cpp
index a3f4fd87b7..c1bee02e6d 100644
--- a/gm/shadermaskfilter.cpp
+++ b/gm/shadermaskfilter.cpp
@@ -11,7 +11,6 @@
 #include "SkCanvas.h"
 #include "SkImage.h"
 #include "SkMaskFilter.h"
-#include "SkPictureRecorder.h"
 #include "SkShaderMaskFilter.h"
 
 static void draw_masked_image(SkCanvas* canvas, const SkImage* image, SkScalar x, SkScalar y,
@@ -221,92 +220,3 @@ DEF_SIMPLE_GM(savelayer_maskfilter, canvas, 450, 675) {
     }
 }
 
-static void draw_mask(SkCanvas* canvas) {
-    SkPaint p;
-    p.setAntiAlias(true);
-    canvas->drawOval(SkRect::Make(canvas->imageInfo().bounds()), p);
-}
-
-DEF_SIMPLE_GM(shadermaskfilter_localmatrix, canvas, 1200, 800) {
-    static constexpr SkScalar kSize = 100;
-
-    using ShaderMakerT = sk_sp<SkShader>(*)(SkCanvas*, const SkMatrix& lm);
-    static const ShaderMakerT gShaderMakers[] = {
-        [](SkCanvas* canvas, const SkMatrix& lm) -> sk_sp<SkShader> {
-            auto surface = sk_tool_utils::makeSurface(canvas,
-                                                      SkImageInfo::MakeN32Premul(kSize, kSize));
-            draw_mask(surface->getCanvas());
-            return surface->makeImageSnapshot()->makeShader(SkShader::kClamp_TileMode,
-                                                            SkShader::kClamp_TileMode, &lm);
-        },
-        [](SkCanvas*, const SkMatrix& lm) -> sk_sp<SkShader> {
-            SkPictureRecorder recorder;
-            draw_mask(recorder.beginRecording(kSize, kSize));
-            return SkShader::MakePictureShader(recorder.finishRecordingAsPicture(),
-                                               SkShader::kClamp_TileMode,
-                                               SkShader::kClamp_TileMode,
-                                               &lm, nullptr);
-        },
-    };
-
-    struct Config {
-        SkMatrix fCanvasMatrix,
-                 fMaskMatrix,
-                 fShaderMatrix;
-    } gConfigs[] = {
-        { SkMatrix::I(), SkMatrix::MakeScale(2, 2), SkMatrix::MakeTrans(10, 10) },
-        { SkMatrix::MakeScale(2, 2), SkMatrix::I(), SkMatrix::MakeTrans(10, 10) },
-        { SkMatrix::MakeScale(2, 2), SkMatrix::MakeTrans(10, 10), SkMatrix::I() },
-        { SkMatrix::Concat(SkMatrix::MakeScale(2, 2), SkMatrix::MakeTrans(10, 10)),
-          SkMatrix::I(), SkMatrix::I() },
-        { SkMatrix::I(),
-          SkMatrix::Concat(SkMatrix::MakeScale(2, 2), SkMatrix::MakeTrans(10, 10)),
-          SkMatrix::I() },
-        { SkMatrix::I(), SkMatrix::I(),
-          SkMatrix::Concat(SkMatrix::MakeScale(2, 2), SkMatrix::MakeTrans(10, 10)) },
-    };
-
-    using DrawerT = void(*)(SkCanvas*, const SkRect&, const SkPaint&);
-    static const DrawerT gDrawers[] = {
-        [](SkCanvas* canvas, const SkRect& dest, const SkPaint& mask) {
-            canvas->drawRect(dest, mask);
-        },
-        [](SkCanvas* canvas, const SkRect& dest, const SkPaint& mask) {
-            canvas->saveLayer(&dest, &mask);
-            SkPaint p = mask;
-            p.setMaskFilter(nullptr);
-            canvas->drawPaint(p);
-            canvas->restore();
-        },
-    };
-
-    SkPaint paint, rectPaint;
-    paint.setColor(0xff00ff00);
-    rectPaint.setStyle(SkPaint::kStroke_Style);
-    rectPaint.setColor(0xffff0000);
-
-    for (const auto& sm : gShaderMakers) {
-        for (const auto& drawer : gDrawers) {
-            {
-                SkAutoCanvasRestore acr(canvas, true);
-                for (const auto& cfg : gConfigs) {
-                    paint.setMaskFilter(SkShaderMaskFilter::Make(sm(canvas, cfg.fShaderMatrix))
-                                        ->makeWithMatrix(cfg.fMaskMatrix));
-                    auto dest = SkRect::MakeWH(kSize, kSize);
-                    SkMatrix::Concat(cfg.fMaskMatrix, cfg.fShaderMatrix).mapRect(&dest);
-
-                    {
-                        SkAutoCanvasRestore acr(canvas, true);
-                        canvas->concat(cfg.fCanvasMatrix);
-                        drawer(canvas, dest, paint);
-                        canvas->drawRect(dest, rectPaint);
-                    }
-
-                    canvas->translate(kSize * 2.5f, 0);
-                }
-            }
-            canvas->translate(0, kSize * 2.5f);
-        }
-
-    }
-}
diff --git a/include/core/SkMaskFilter.h b/include/core/SkMaskFilter.h
index bd2a20b825..1ea774a027 100644
--- a/include/core/SkMaskFilter.h
+++ b/include/core/SkMaskFilter.h
@@ -50,14 +50,7 @@ public:
     static sk_sp<SkMaskFilter> MakeCombine(sk_sp<SkMaskFilter> filterA, sk_sp<SkMaskFilter> filterB,
                                            SkCoverageMode mode);
 
-    /**
-     *  Construct a maskfilter with an additional transform.
-     *
-     *  Note: unlike shader local matrices, this transform composes next to the CTM.
-     *
-     *    TotalMatrix = CTM x MaskFilterMatrix x (optional/downstream) ShaderLocalMatrix
-     */
-    sk_sp<SkMaskFilter> makeWithMatrix(const SkMatrix&) const;
+    sk_sp<SkMaskFilter> makeWithLocalMatrix(const SkMatrix&) const;
 
     virtual void toString(SkString* str) const = 0;
     SK_DEFINE_FLATTENABLE_TYPE(SkMaskFilter)
diff --git a/src/core/SkBitmapDevice.cpp b/src/core/SkBitmapDevice.cpp
index 26cdc39313..5f84ba1aec 100644
--- a/src/core/SkBitmapDevice.cpp
+++ b/src/core/SkBitmapDevice.cpp
@@ -487,7 +487,7 @@ void SkBitmapDevice::drawDevice(SkBaseDevice* device, int x, int y, const SkPain
     // todo: can we unify with similar adjustment in SkGpuDevice?
     SkTCopyOnFirstWrite<SkPaint> paint(origPaint);
     if (paint->getMaskFilter()) {
-        paint.writable()->setMaskFilter(paint->getMaskFilter()->makeWithMatrix(this->ctm()));
+        paint.writable()->setMaskFilter(paint->getMaskFilter()->makeWithLocalMatrix(this->ctm()));
     }
     this->drawSprite(static_cast<SkBitmapDevice*>(device)->fBitmap, x, y, *paint);
 }
@@ -546,7 +546,7 @@ void SkBitmapDevice::drawSpecial(SkSpecialImage* src, int x, int y, const SkPain
     }
 
     if (paint->getMaskFilter()) {
-        paint.writable()->setMaskFilter(paint->getMaskFilter()->makeWithMatrix(this->ctm()));
+        paint.writable()->setMaskFilter(paint->getMaskFilter()->makeWithLocalMatrix(this->ctm()));
     }
 
     if (!clipImage) {
diff --git a/src/core/SkMaskFilter.cpp b/src/core/SkMaskFilter.cpp
index 8e555f37ad..e1805637a9 100644
--- a/src/core/SkMaskFilter.cpp
+++ b/src/core/SkMaskFilter.cpp
@@ -635,9 +635,9 @@ void SkCombineMF::toString(SkString* str) const {
 
 ///////////////////////////////////////////////////////////////////////////////////////////////////
 
-class SkMatrixMF : public SkMaskFilterBase {
+class SkLocalMatrixMF : public SkMaskFilterBase {
 public:
-    SkMatrixMF(sk_sp<SkMaskFilter> filter, const SkMatrix& lm)
+    SkLocalMatrixMF(sk_sp<SkMaskFilter> filter, const SkMatrix& lm)
         : fFilter(std::move(filter))
         , fLM(lm)
     {}
@@ -657,7 +657,7 @@ public:
     SkMask::Format getFormat() const override { return as_MFB(fFilter)->getFormat(); }
 
     void toString(SkString* str) const override {
-        str->set("SkMatrixMF:");
+        str->set("SkLocalMatrixMF:");
     }
 
     SK_DECLARE_PUBLIC_FLATTENABLE_DESERIALIZATION_PROCS(SkLocalMatrixMF)
@@ -665,7 +665,16 @@ public:
 protected:
 #if SK_SUPPORT_GPU
     std::unique_ptr<GrFragmentProcessor> onAsFragmentProcessor(const GrFPArgs& args) const override{
-        return as_MFB(fFilter)->asFragmentProcessor(args.makeWithPostLocalMatrix(fLM));
+        GrFPArgs newArgs = args;
+
+        SkMatrix storage;
+        if (args.fLocalMatrix) {
+            storage.setConcat(*args.fLocalMatrix, fLM);
+            newArgs.fLocalMatrix = &storage;
+        } else {
+            newArgs.fLocalMatrix = &fLM;
+        }
+        return as_MFB(fFilter)->asFragmentProcessor(newArgs);
     }
 
     bool onHasFragmentProcessor() const override {
@@ -686,11 +695,11 @@ private:
     typedef SkMaskFilterBase INHERITED;
 };
 
-sk_sp<SkFlattenable> SkMatrixMF::CreateProc(SkReadBuffer& buffer) {
-    SkMatrix m;
-    buffer.readMatrix(&m);
+sk_sp<SkFlattenable> SkLocalMatrixMF::CreateProc(SkReadBuffer& buffer) {
+    SkMatrix lm;
+    buffer.readMatrix(&lm);
     auto filter = buffer.readMaskFilter();
-    return filter ? filter->makeWithMatrix(m) : nullptr;
+    return filter ? filter->makeWithLocalMatrix(lm) : nullptr;
 }
 
 ///////////////////////////////////////////////////////////////////////////////////////////////////
@@ -726,16 +735,16 @@ sk_sp<SkMaskFilter> SkMaskFilter::MakeCombine(sk_sp<SkMaskFilter> dst, sk_sp<SkM
     return sk_sp<SkMaskFilter>(new SkCombineMF(std::move(dst), std::move(src), mode));
 }
 
-sk_sp<SkMaskFilter> SkMaskFilter::makeWithMatrix(const SkMatrix& lm) const {
+sk_sp<SkMaskFilter> SkMaskFilter::makeWithLocalMatrix(const SkMatrix& lm) const {
     sk_sp<SkMaskFilter> me = sk_ref_sp(const_cast<SkMaskFilter*>(this));
     if (lm.isIdentity()) {
         return me;
     }
-    return sk_sp<SkMaskFilter>(new SkMatrixMF(std::move(me), lm));
+    return sk_sp<SkMaskFilter>(new SkLocalMatrixMF(std::move(me), lm));
 }
 
 void SkMaskFilter::InitializeFlattenables() {
-    SK_DEFINE_FLATTENABLE_REGISTRAR_ENTRY(SkMatrixMF)
+    SK_DEFINE_FLATTENABLE_REGISTRAR_ENTRY(SkLocalMatrixMF)
     SK_DEFINE_FLATTENABLE_REGISTRAR_ENTRY(SkComposeMF)
     SK_DEFINE_FLATTENABLE_REGISTRAR_ENTRY(SkCombineMF)
     sk_register_blur_maskfilter_createproc();
diff --git a/src/gpu/GrFPArgs.h b/src/gpu/GrFPArgs.h
index 3c1b8e37b9..cdb2e5de50 100644
--- a/src/gpu/GrFPArgs.h
+++ b/src/gpu/GrFPArgs.h
@@ -9,84 +9,39 @@
 #define GrFPArgs_DEFINED
 
 #include "SkFilterQuality.h"
-#include "SkMatrix.h"
 
+class SkMatrix;
 class GrContext;
 class GrColorSpaceInfo;
 
 struct GrFPArgs {
+    GrFPArgs(GrContext* context,
+             const SkMatrix* viewMatrix,
+             const SkMatrix* localMatrix,
+             SkFilterQuality filterQuality,
+             const GrColorSpaceInfo* dstColorSpaceInfo)
+        : fContext(context)
+        , fViewMatrix(viewMatrix)
+        , fLocalMatrix(localMatrix)
+        , fFilterQuality(filterQuality)
+        , fDstColorSpaceInfo(dstColorSpaceInfo) {}
+
     GrFPArgs(GrContext* context,
              const SkMatrix* viewMatrix,
              SkFilterQuality filterQuality,
              const GrColorSpaceInfo* dstColorSpaceInfo)
     : fContext(context)
     , fViewMatrix(viewMatrix)
+    , fLocalMatrix(nullptr)
     , fFilterQuality(filterQuality)
-    , fDstColorSpaceInfo(dstColorSpaceInfo) {
-        SkASSERT(fContext);
-        SkASSERT(fViewMatrix);
-    }
-
-    class WithMatrixStorage;
-
-    WithMatrixStorage makeWithPreLocalMatrix(const SkMatrix&) const;
-    WithMatrixStorage makeWithPostLocalMatrix(const SkMatrix&) const;
+    , fDstColorSpaceInfo(dstColorSpaceInfo) {}
 
     GrContext* fContext;
     const SkMatrix* fViewMatrix;
-
-    // We track both pre and post local matrix adjustments.  For a given FP:
-    //
-    //   total_local_matrix = postLocalMatrix x FP_localMatrix x preLocalMatrix
-    //
-    // Use the helpers above to create pre/post GrFPArgs wrappers.
-    //
-    const SkMatrix* fPreLocalMatrix  = nullptr;
-    const SkMatrix* fPostLocalMatrix = nullptr;
-
+    const SkMatrix* fLocalMatrix;
     SkFilterQuality fFilterQuality;
     const GrColorSpaceInfo* fDstColorSpaceInfo;
 };
 
-class GrFPArgs::WithMatrixStorage final : public GrFPArgs {
-private:
-    explicit WithMatrixStorage(const GrFPArgs& args) : INHERITED(args) {}
-
-    friend struct GrFPArgs;
-    SkMatrix fStorage;
-
-    using INHERITED = GrFPArgs;
-};
-
-inline GrFPArgs::WithMatrixStorage GrFPArgs::makeWithPreLocalMatrix(const SkMatrix& lm) const {
-    WithMatrixStorage newArgs(*this);
-
-    if (!lm.isIdentity()) {
-        if (fPreLocalMatrix) {
-            newArgs.fStorage.setConcat(lm, *fPreLocalMatrix);
-            newArgs.fPreLocalMatrix = newArgs.fStorage.isIdentity() ? nullptr : &newArgs.fStorage;
-        } else {
-            newArgs.fPreLocalMatrix = &lm;
-        }
-    }
-
-    return newArgs;
-}
-
-inline GrFPArgs::WithMatrixStorage GrFPArgs::makeWithPostLocalMatrix(const SkMatrix& lm) const {
-    WithMatrixStorage newArgs(*this);
-
-    if (!lm.isIdentity()) {
-        if (fPostLocalMatrix) {
-            newArgs.fStorage.setConcat(*fPostLocalMatrix, lm);
-            newArgs.fPostLocalMatrix = newArgs.fStorage.isIdentity() ? nullptr : &newArgs.fStorage;
-        } else {
-            newArgs.fPostLocalMatrix = &lm;
-        }
-    }
-
-    return newArgs;
-}
-
 #endif
 
diff --git a/src/gpu/GrTestUtils.cpp b/src/gpu/GrTestUtils.cpp
index 2aff35dad2..8958f84704 100644
--- a/src/gpu/GrTestUtils.cpp
+++ b/src/gpu/GrTestUtils.cpp
@@ -337,7 +337,8 @@ TestAsFPArgs::TestAsFPArgs(GrProcessorTestData* d)
     : fViewMatrixStorage(TestMatrix(d->fRandom))
     , fColorSpaceInfoStorage(skstd::make_unique<GrColorSpaceInfo>(TestColorSpace(d->fRandom),
                                                                   kRGBA_8888_GrPixelConfig))
-    , fArgs(d->context(), &fViewMatrixStorage, kNone_SkFilterQuality, fColorSpaceInfoStorage.get())
+    , fArgs(d->context(), &fViewMatrixStorage, nullptr, kNone_SkFilterQuality,
+            fColorSpaceInfoStorage.get())
 {}
 
 TestAsFPArgs::~TestAsFPArgs() {}
diff --git a/src/gpu/SkGpuDevice.cpp b/src/gpu/SkGpuDevice.cpp
index fca372eabd..eb26ba2f9b 100644
--- a/src/gpu/SkGpuDevice.cpp
+++ b/src/gpu/SkGpuDevice.cpp
@@ -1092,7 +1092,7 @@ void SkGpuDevice::drawSpecial(SkSpecialImage* special1, int left, int top, const
     if (tmpUnfiltered.getMaskFilter()) {
         SkMatrix ctm = this->ctm();
         ctm.postTranslate(-SkIntToScalar(left + offset.fX), -SkIntToScalar(top + offset.fY));
-        tmpUnfiltered.setMaskFilter(tmpUnfiltered.getMaskFilter()->makeWithMatrix(ctm));
+        tmpUnfiltered.setMaskFilter(tmpUnfiltered.getMaskFilter()->makeWithLocalMatrix(ctm));
     }
 
     tmpUnfiltered.setImageFilter(nullptr);
diff --git a/src/gpu/SkGr.cpp b/src/gpu/SkGr.cpp
index 99465f620b..b4cb22bd10 100644
--- a/src/gpu/SkGr.cpp
+++ b/src/gpu/SkGr.cpp
@@ -514,7 +514,7 @@ bool SkPaintToGrPaintWithTexture(GrContext* context,
     if (textureIsAlphaOnly) {
         if (const auto* shader = as_SB(paint.getShader())) {
             shaderFP = shader->asFragmentProcessor(GrFPArgs(
-                    context, &viewM, paint.getFilterQuality(), &colorSpaceInfo));
+                    context, &viewM, nullptr, paint.getFilterQuality(), &colorSpaceInfo));
             if (!shaderFP) {
                 return false;
             }
diff --git a/src/shaders/SkImageShader.cpp b/src/shaders/SkImageShader.cpp
index 1838e9e498..265f7e6013 100644
--- a/src/shaders/SkImageShader.cpp
+++ b/src/shaders/SkImageShader.cpp
@@ -213,11 +213,19 @@ static GrSamplerState::WrapMode tile_mode_to_wrap_mode(const SkShader::TileMode
 
 std::unique_ptr<GrFragmentProcessor> SkImageShader::asFragmentProcessor(
         const GrFPArgs& args) const {
-    const auto lm = this->totalLocalMatrix(args.fPreLocalMatrix, args.fPostLocalMatrix);
+    SkMatrix lm = this->getLocalMatrix();
     SkMatrix lmInverse;
-    if (!lm->invert(&lmInverse)) {
+    if (!lm.invert(&lmInverse)) {
         return nullptr;
     }
+    if (args.fLocalMatrix) {
+        SkMatrix inv;
+        if (!args.fLocalMatrix->invert(&inv)) {
+            return nullptr;
+        }
+        lmInverse.postConcat(inv);
+        lm.preConcat(*args.fLocalMatrix);
+    }
 
     GrSamplerState::WrapMode wrapModes[] = {tile_mode_to_wrap_mode(fTileModeX),
                                             tile_mode_to_wrap_mode(fTileModeY)};
@@ -228,7 +236,7 @@ std::unique_ptr<GrFragmentProcessor> SkImageShader::asFragmentProcessor(
     // are provided by the caller.
     bool doBicubic;
     GrSamplerState::Filter textureFilterMode = GrSkFilterQualityToGrFilterMode(
-            args.fFilterQuality, *args.fViewMatrix, *lm,
+            args.fFilterQuality, *args.fViewMatrix, lm,
             args.fContext->contextPriv().sharpenMipmappedTextures(), &doBicubic);
     GrSamplerState samplerState(wrapModes, textureFilterMode);
     sk_sp<SkColorSpace> texColorSpace;
diff --git a/src/shaders/SkLocalMatrixShader.cpp b/src/shaders/SkLocalMatrixShader.cpp
index 78110f5821..ca99af1bef 100644
--- a/src/shaders/SkLocalMatrixShader.cpp
+++ b/src/shaders/SkLocalMatrixShader.cpp
@@ -15,8 +15,13 @@
 #if SK_SUPPORT_GPU
 std::unique_ptr<GrFragmentProcessor> SkLocalMatrixShader::asFragmentProcessor(
         const GrFPArgs& args) const {
+    SkMatrix tmp = this->getLocalMatrix();
+    if (args.fLocalMatrix) {
+        tmp.preConcat(*args.fLocalMatrix);
+    }
     return as_SB(fProxyShader)
-            ->asFragmentProcessor(args.makeWithPreLocalMatrix(this->getLocalMatrix()));
+            ->asFragmentProcessor(GrFPArgs(args.fContext, args.fViewMatrix, &tmp,
+                                           args.fFilterQuality, args.fDstColorSpaceInfo));
 }
 #endif
 
diff --git a/src/shaders/SkPerlinNoiseShader.cpp b/src/shaders/SkPerlinNoiseShader.cpp
index 8f6c85a494..6179da3d32 100644
--- a/src/shaders/SkPerlinNoiseShader.cpp
+++ b/src/shaders/SkPerlinNoiseShader.cpp
@@ -1394,8 +1394,13 @@ std::unique_ptr<GrFragmentProcessor> SkPerlinNoiseShaderImpl::asFragmentProcesso
         const GrFPArgs& args) const {
     SkASSERT(args.fContext);
 
-    const auto localMatrix = this->totalLocalMatrix(args.fPreLocalMatrix, args.fPostLocalMatrix);
-    const auto matrix = SkMatrix::Concat(*args.fViewMatrix, *localMatrix);
+    SkMatrix localMatrix = this->getLocalMatrix();
+    if (args.fLocalMatrix) {
+        localMatrix.preConcat(*args.fLocalMatrix);
+    }
+
+    SkMatrix matrix = *args.fViewMatrix;
+    matrix.preConcat(localMatrix);
 
     // Either we don't stitch tiles, either we have a valid tile size
     SkASSERT(!fStitchTiles || !fTileSize.isEmpty());
@@ -1408,8 +1413,8 @@ std::unique_ptr<GrFragmentProcessor> SkPerlinNoiseShaderImpl::asFragmentProcesso
                                                                   matrix);
 
     SkMatrix m = *args.fViewMatrix;
-    m.setTranslateX(-localMatrix->getTranslateX() + SK_Scalar1);
-    m.setTranslateY(-localMatrix->getTranslateY() + SK_Scalar1);
+    m.setTranslateX(-localMatrix.getTranslateX() + SK_Scalar1);
+    m.setTranslateY(-localMatrix.getTranslateY() + SK_Scalar1);
 
     auto proxyProvider = args.fContext->contextPriv().proxyProvider();
     if (fType == kImprovedNoise_Type) {
diff --git a/src/shaders/SkPictureShader.cpp b/src/shaders/SkPictureShader.cpp
index de642a7dda..1d19b04721 100644
--- a/src/shaders/SkPictureShader.cpp
+++ b/src/shaders/SkPictureShader.cpp
@@ -175,20 +175,24 @@ void SkPictureShader::flatten(SkWriteBuffer& buffer) const {
 //
 // 1) a cached image shader, which wraps a single picture tile at the given CTM/local matrix
 //
-// 2) a tile scale adjustment, to be applied downstream when dispatching createContext(),
+// 2) a "composite" local matrix, to be passed down when dispatching createContext(),
 //    appendStages() and asFragmentProcessor() in callers
 //
 // The composite local matrix includes the actual local matrix, any inherited/outer local matrix
 // and a scale component (to mape the actual tile bitmap size -> fTile size).
 //
 sk_sp<SkShader> SkPictureShader::refBitmapShader(const SkMatrix& viewMatrix,
-                                                 const SkMatrix& localMatrix,
+                                                 const SkMatrix* outerLocalMatrix,
                                                  SkColorSpace* dstColorSpace,
-                                                 SkVector* scaleAdjust,
+                                                 SkMatrix* compositeLocalMatrix,
                                                  const int maxTextureSize) const {
     SkASSERT(fPicture && !fPicture->cullRect().isEmpty());
 
-    const SkMatrix m = SkMatrix::Concat(viewMatrix, localMatrix);
+    *compositeLocalMatrix = this->getLocalMatrix();
+    if (outerLocalMatrix) {
+        compositeLocalMatrix->preConcat(*outerLocalMatrix);
+    }
+    const SkMatrix m = SkMatrix::Concat(viewMatrix, *compositeLocalMatrix);
 
     // Use a rotation-invariant scale
     SkPoint scale;
@@ -270,52 +274,37 @@ sk_sp<SkShader> SkPictureShader::refBitmapShader(const SkMatrix& viewMatrix,
         fAddedToCache.store(true);
     }
 
-    scaleAdjust->set(1 / tileScale.width(), 1 / tileScale.height());
+    compositeLocalMatrix->preScale(1 / tileScale.width(), 1 / tileScale.height());
 
     return tileShader;
 }
 
 bool SkPictureShader::onAppendStages(const StageRec& rec) const {
-    auto lm = this->totalLocalMatrix(rec.fLocalM);
-    SkVector scaleAdjust;
-
     // Keep bitmapShader alive by using alloc instead of stack memory
     auto& bitmapShader = *rec.fAlloc->make<sk_sp<SkShader>>();
-    bitmapShader = this->refBitmapShader(rec.fCTM, *lm, rec.fDstCS, &scaleAdjust);
-
-    if (!bitmapShader) {
-        return false;
-    }
-
-    if (scaleAdjust != SkVector::Make(1, 1)) {
-        lm.writable()->preScale(scaleAdjust.fX, scaleAdjust.fY);
-    }
+    SkMatrix compositeLocalMatrix;
+    bitmapShader = this->refBitmapShader(rec.fCTM, rec.fLocalM, rec.fDstCS, &compositeLocalMatrix);
 
     StageRec localRec = rec;
-    localRec.fLocalM = lm->isIdentity() ? nullptr : &(*lm);
+    localRec.fLocalM = compositeLocalMatrix.isIdentity() ? nullptr : &compositeLocalMatrix;
 
-    return as_SB(bitmapShader)->appendStages(localRec);
+    return bitmapShader && as_SB(bitmapShader)->appendStages(localRec);
 }
 
 /////////////////////////////////////////////////////////////////////////////////////////
 SkShaderBase::Context* SkPictureShader::onMakeContext(const ContextRec& rec, SkArenaAlloc* alloc)
 const {
-    auto lm = this->totalLocalMatrix(rec.fLocalMatrix);
-    SkVector scaleAdjust;
+    SkMatrix compositeLocalMatrix;
     sk_sp<SkShader> bitmapShader = this->refBitmapShader(*rec.fMatrix,
-                                                         *lm,
+                                                         rec.fLocalMatrix,
                                                          rec.fDstColorSpace,
-                                                         &scaleAdjust);
+                                                         &compositeLocalMatrix);
     if (!bitmapShader) {
         return nullptr;
     }
 
-    if (scaleAdjust != SkVector::Make(1, 1)) {
-        lm.writable()->preScale(scaleAdjust.fX, scaleAdjust.fY);
-    }
-
     ContextRec localRec = rec;
-    localRec.fLocalMatrix = lm->isIdentity() ? nullptr : &(*lm);
+    localRec.fLocalMatrix = compositeLocalMatrix.isIdentity() ? nullptr : &compositeLocalMatrix;
 
     PictureShaderContext* ctx =
         alloc->make<PictureShaderContext>(*this, localRec, std::move(bitmapShader), alloc);
@@ -380,22 +369,20 @@ std::unique_ptr<GrFragmentProcessor> SkPictureShader::asFragmentProcessor(
     if (args.fContext) {
         maxTextureSize = args.fContext->caps()->maxTextureSize();
     }
-
-    auto lm = this->totalLocalMatrix(args.fPreLocalMatrix, args.fPostLocalMatrix);
-    SkVector scaleAdjust;
-    sk_sp<SkShader> bitmapShader(this->refBitmapShader(*args.fViewMatrix,*lm,
+    SkMatrix compositeLocalMatrix;
+    sk_sp<SkShader> bitmapShader(this->refBitmapShader(*args.fViewMatrix, args.fLocalMatrix,
                                                        args.fDstColorSpaceInfo->colorSpace(),
-                                                       &scaleAdjust, maxTextureSize));
+                                                       &compositeLocalMatrix,
+                                                       maxTextureSize));
     if (!bitmapShader) {
         return nullptr;
     }
 
-    if (scaleAdjust != SkVector::Make(1, 1)) {
-        lm.writable()->preScale(scaleAdjust.fX, scaleAdjust.fY);
-    }
-
-    GrFPArgs newArgs(args.fContext, args.fViewMatrix, args.fFilterQuality, args.fDstColorSpaceInfo);
-
-    return as_SB(bitmapShader)->asFragmentProcessor(newArgs.makeWithPreLocalMatrix(*lm));
+    return as_SB(bitmapShader)->asFragmentProcessor(
+        GrFPArgs(args.fContext,
+                 args.fViewMatrix,
+                 compositeLocalMatrix.isIdentity() ? nullptr : &compositeLocalMatrix,
+                 args.fFilterQuality,
+                 args.fDstColorSpaceInfo));
 }
 #endif
diff --git a/src/shaders/SkPictureShader.h b/src/shaders/SkPictureShader.h
index 5f28b6074b..4970dac81b 100644
--- a/src/shaders/SkPictureShader.h
+++ b/src/shaders/SkPictureShader.h
@@ -46,9 +46,9 @@ private:
     SkPictureShader(sk_sp<SkPicture>, TileMode, TileMode, const SkMatrix*, const SkRect*,
                     sk_sp<SkColorSpace>);
 
-    sk_sp<SkShader> refBitmapShader(const SkMatrix&, const SkMatrix& localMatrix,
+    sk_sp<SkShader> refBitmapShader(const SkMatrix&, const SkMatrix* localMatrix,
                                     SkColorSpace* dstColorSpace,
-                                    SkVector* scaleAdjust,
+                                    SkMatrix* compositeLocalMatrix,
                                     const int maxTextureSize = 0) const;
 
     class PictureShaderContext : public Context {
diff --git a/src/shaders/SkShader.cpp b/src/shaders/SkShader.cpp
index f3ffd26ef4..265f22beaa 100644
--- a/src/shaders/SkShader.cpp
+++ b/src/shaders/SkShader.cpp
@@ -67,26 +67,15 @@ void SkShaderBase::flatten(SkWriteBuffer& buffer) const {
     }
 }
 
-SkTCopyOnFirstWrite<SkMatrix>
-SkShaderBase::totalLocalMatrix(const SkMatrix* preLocalMatrix,
-                               const SkMatrix* postLocalMatrix) const {
-    SkTCopyOnFirstWrite<SkMatrix> m(fLocalMatrix);
-
-    if (preLocalMatrix) {
-        m.writable()->preConcat(*preLocalMatrix);
-    }
-
-    if (postLocalMatrix) {
-        m.writable()->postConcat(*postLocalMatrix);
-    }
-
-    return m;
-}
-
 bool SkShaderBase::computeTotalInverse(const SkMatrix& ctm,
                                        const SkMatrix* outerLocalMatrix,
                                        SkMatrix* totalInverse) const {
-    return SkMatrix::Concat(ctm, *this->totalLocalMatrix(outerLocalMatrix)).invert(totalInverse);
+    SkMatrix total = SkMatrix::Concat(ctm, fLocalMatrix);
+    if (outerLocalMatrix) {
+        total.preConcat(*outerLocalMatrix);
+    }
+
+    return total.invert(totalInverse);
 }
 
 bool SkShaderBase::asLuminanceColor(SkColor* colorPtr) const {
diff --git a/src/shaders/SkShaderBase.h b/src/shaders/SkShaderBase.h
index cddc5540ee..89674a6ebb 100644
--- a/src/shaders/SkShaderBase.h
+++ b/src/shaders/SkShaderBase.h
@@ -12,7 +12,6 @@
 #include "SkMask.h"
 #include "SkMatrix.h"
 #include "SkShader.h"
-#include "SkTLazy.h"
 
 #if SK_SUPPORT_GPU
 #include "GrFPArgs.h"
@@ -186,16 +185,9 @@ public:
     // If this returns false, then we draw nothing (do not fall back to shader context)
     bool appendStages(const StageRec&) const;
 
-    bool SK_WARN_UNUSED_RESULT computeTotalInverse(const SkMatrix& ctm,
-                                                   const SkMatrix* outerLocalMatrix,
-                                                   SkMatrix* totalInverse) const;
-
-    // Returns the total local matrix for this shader:
-    //
-    //   M = postLocalMatrix x shaderLocalMatrix x preLocalMatrix
-    //
-    SkTCopyOnFirstWrite<SkMatrix> totalLocalMatrix(const SkMatrix* preLocalMatrix,
-                                                   const SkMatrix* postLocalMatrix = nullptr) const;
+    bool computeTotalInverse(const SkMatrix& ctm,
+                             const SkMatrix* outerLocalMatrix,
+                             SkMatrix* totalInverse) const;
 
 #ifdef SK_SUPPORT_LEGACY_SHADER_ISABITMAP
     virtual bool onIsABitmap(SkBitmap*, SkMatrix*, TileMode[2]) const {
diff --git a/src/shaders/gradients/SkLinearGradient.cpp b/src/shaders/gradients/SkLinearGradient.cpp
index a377442dc4..b6436ca801 100644
--- a/src/shaders/gradients/SkLinearGradient.cpp
+++ b/src/shaders/gradients/SkLinearGradient.cpp
@@ -195,9 +195,16 @@ std::unique_ptr<GrFragmentProcessor> SkLinearGradient::asFragmentProcessor(
     SkASSERT(args.fContext);
 
     SkMatrix matrix;
-    if (!this->totalLocalMatrix(args.fPreLocalMatrix, args.fPostLocalMatrix)->invert(&matrix)) {
+    if (!this->getLocalMatrix().invert(&matrix)) {
         return nullptr;
     }
+    if (args.fLocalMatrix) {
+        SkMatrix inv;
+        if (!args.fLocalMatrix->invert(&inv)) {
+            return nullptr;
+        }
+        matrix.postConcat(inv);
+    }
     matrix.postConcat(fPtsToUnit);
 
     return GrLinearGradient::Make(GrGradientEffect::CreateArgs(
diff --git a/src/shaders/gradients/SkRadialGradient.cpp b/src/shaders/gradients/SkRadialGradient.cpp
index 6d193615a4..3f71f38c1f 100644
--- a/src/shaders/gradients/SkRadialGradient.cpp
+++ b/src/shaders/gradients/SkRadialGradient.cpp
@@ -162,10 +162,19 @@ void GrRadialGradient::GLSLRadialProcessor::emitCode(EmitArgs& args) {
 
 std::unique_ptr<GrFragmentProcessor> SkRadialGradient::asFragmentProcessor(
         const GrFPArgs& args) const {
+    SkASSERT(args.fContext);
+
     SkMatrix matrix;
-    if (!this->totalLocalMatrix(args.fPreLocalMatrix, args.fPostLocalMatrix)->invert(&matrix)) {
+    if (!this->getLocalMatrix().invert(&matrix)) {
         return nullptr;
     }
+    if (args.fLocalMatrix) {
+        SkMatrix inv;
+        if (!args.fLocalMatrix->invert(&inv)) {
+            return nullptr;
+        }
+        matrix.postConcat(inv);
+    }
     matrix.postConcat(fPtsToUnit);
 
     return GrRadialGradient::Make(GrGradientEffect::CreateArgs(
diff --git a/src/shaders/gradients/SkSweepGradient.cpp b/src/shaders/gradients/SkSweepGradient.cpp
index 3f605d7761..e9c4450b00 100644
--- a/src/shaders/gradients/SkSweepGradient.cpp
+++ b/src/shaders/gradients/SkSweepGradient.cpp
@@ -216,9 +216,16 @@ void GrSweepGradient::GLSLSweepProcessor::emitCode(EmitArgs& args) {
 std::unique_ptr<GrFragmentProcessor> SkSweepGradient::asFragmentProcessor(
         const GrFPArgs& args) const {
     SkMatrix matrix;
-    if (!this->totalLocalMatrix(args.fPreLocalMatrix, args.fPostLocalMatrix)->invert(&matrix)) {
+    if (!this->getLocalMatrix().invert(&matrix)) {
         return nullptr;
     }
+    if (args.fLocalMatrix) {
+        SkMatrix inv;
+        if (!args.fLocalMatrix->invert(&inv)) {
+            return nullptr;
+        }
+        matrix.postConcat(inv);
+    }
     matrix.postConcat(fPtsToUnit);
 
     return GrSweepGradient::Make(
diff --git a/src/shaders/gradients/SkTwoPointConicalGradient.cpp b/src/shaders/gradients/SkTwoPointConicalGradient.cpp
index 88f0f82245..cce25d00b7 100644
--- a/src/shaders/gradients/SkTwoPointConicalGradient.cpp
+++ b/src/shaders/gradients/SkTwoPointConicalGradient.cpp
@@ -175,13 +175,9 @@ void SkTwoPointConicalGradient::flatten(SkWriteBuffer& buffer) const {
 
 std::unique_ptr<GrFragmentProcessor> SkTwoPointConicalGradient::asFragmentProcessor(
         const GrFPArgs& args) const {
-    SkMatrix matrix;
-    if (!this->totalLocalMatrix(args.fPreLocalMatrix, args.fPostLocalMatrix)->invert(&matrix)) {
-        return nullptr;
-    }
-
+    SkASSERT(args.fContext);
     return Gr2PtConicalGradientEffect::Make(
-            GrGradientEffect::CreateArgs(args.fContext, this, &matrix, fTileMode,
+            GrGradientEffect::CreateArgs(args.fContext, this, args.fLocalMatrix, fTileMode,
                                          args.fDstColorSpaceInfo->colorSpace()));
 }
 
diff --git a/src/shaders/gradients/SkTwoPointConicalGradient_gpu.cpp b/src/shaders/gradients/SkTwoPointConicalGradient_gpu.cpp
index 9d447785ed..c242895f1c 100644
--- a/src/shaders/gradients/SkTwoPointConicalGradient_gpu.cpp
+++ b/src/shaders/gradients/SkTwoPointConicalGradient_gpu.cpp
@@ -388,7 +388,18 @@ std::unique_ptr<GrFragmentProcessor> Gr2PtConicalGradientEffect::Make(
     const SkTwoPointConicalGradient& shader =
         *static_cast<const SkTwoPointConicalGradient*>(args.fShader);
 
-    SkMatrix matrix = *args.fMatrix;
+    SkMatrix matrix;
+    if (!shader.getLocalMatrix().invert(&matrix)) {
+        return nullptr;
+    }
+    if (args.fMatrix) {
+        SkMatrix inv;
+        if (!args.fMatrix->invert(&inv)) {
+            return nullptr;
+        }
+        matrix.postConcat(inv);
+    }
+
     GrGradientEffect::CreateArgs newArgs(args.fContext, args.fShader, &matrix, args.fWrapMode,
         args.fDstColorSpace);
     // Data and matrix has to be prepared before constructing TwoPointConicalEffect so its parent
diff --git a/tests/TessellatingPathRendererTests.cpp b/tests/TessellatingPathRendererTests.cpp
index 0ff0af71e0..8fefc9b56e 100644
--- a/tests/TessellatingPathRendererTests.cpp
+++ b/tests/TessellatingPathRendererTests.cpp
@@ -479,7 +479,8 @@ static std::unique_ptr<GrFragmentProcessor> create_linear_gradient_processor(GrC
     sk_sp<SkShader> shader = SkGradientShader::MakeLinear(
         pts, colors, nullptr, SK_ARRAY_COUNT(colors), SkShader::kClamp_TileMode);
     GrColorSpaceInfo colorSpaceInfo(nullptr, kRGBA_8888_GrPixelConfig);
-    GrFPArgs args(ctx, &SkMatrix::I(), SkFilterQuality::kLow_SkFilterQuality, &colorSpaceInfo);
+    GrFPArgs args(ctx, &SkMatrix::I(), &SkMatrix::I(), SkFilterQuality::kLow_SkFilterQuality,
+                  &colorSpaceInfo);
     return as_SB(shader)->asFragmentProcessor(args);
 }