From 0d3344e17d258106617b0e6d783d073b188a2548 Mon Sep 17 00:00:00 2001
From: Adrian Perez de Castro <aperez@igalia.com>
Date: Thu, 2 Jun 2022 11:19:06 +0300
Subject: [PATCH] [ARM][NEON] FELightningNEON.cpp fails to build, NEON fast
 path seems unused https://bugs.webkit.org/show_bug.cgi?id=241182

Reviewed by NOBODY (OOPS!).

Move the NEON fast path for the SVG lighting filter effects into
FELightingSoftwareApplier, and arrange to actually use them by
forwarding calls to applyPlatformGeneric() into applyPlatformNeon().

Some changes were needed to adapt platformApplyNeon() to the current
state of filters after r286140. This was not detected because the code
bitrotted due to it being guarded with CPU(ARM_TRADITIONAL), which does
not get used much these days: CPU(ARM_THUMB2) is more common. It should
be possible to use the NEON fast paths also in Thumb mode, but that is
left for a follow-up fix.

* Source/WebCore/platform/graphics/cpu/arm/filters/FELightingNEON.cpp:
(WebCore::FELightingSoftwareApplier::platformApplyNeonWorker):
(WebCore::FELightingSoftwareApplier::getPowerCoefficients):
(WebCore::FELighting::platformApplyNeonWorker): Deleted.
(WebCore::FELighting::getPowerCoefficients): Deleted.
* Source/WebCore/platform/graphics/cpu/arm/filters/FELightingNEON.h:
(WebCore::FELightingSoftwareApplier::applyPlatformNeon):
(WebCore::FELighting::platformApplyNeon): Deleted.
* Source/WebCore/platform/graphics/filters/DistantLightSource.h:
* Source/WebCore/platform/graphics/filters/FELighting.h:
* Source/WebCore/platform/graphics/filters/PointLightSource.h:
* Source/WebCore/platform/graphics/filters/SpotLightSource.h:
* Source/WebCore/platform/graphics/filters/software/FELightingSoftwareApplier.h:
---
Upstream-Status: Submitted [https://github.com/WebKit/WebKit/pull/1233]
Signed-off-by: Khem Raj <raj.khem@gmail.com>

 .../cpu/arm/filters/FELightingNEON.cpp        |  4 +-
 .../graphics/cpu/arm/filters/FELightingNEON.h | 54 +++++++++----------
 .../graphics/filters/DistantLightSource.h     |  4 ++
 .../platform/graphics/filters/FELighting.h    |  7 ---
 .../graphics/filters/PointLightSource.h       |  4 ++
 .../graphics/filters/SpotLightSource.h        |  4 ++
 .../software/FELightingSoftwareApplier.h      | 16 ++++++
 7 files changed, 57 insertions(+), 36 deletions(-)

--- a/Source/WebCore/platform/graphics/cpu/arm/filters/FELightingNEON.cpp
+++ b/Source/WebCore/platform/graphics/cpu/arm/filters/FELightingNEON.cpp
@@ -49,7 +49,7 @@ short* feLightingConstantsForNeon()
     return s_FELightingConstantsForNeon;
 }
 
-void FELighting::platformApplyNeonWorker(FELightingPaintingDataForNeon* parameters)
+void FELightingSoftwareApplier::platformApplyNeonWorker(FELightingPaintingDataForNeon* parameters)
 {
     neonDrawLighting(parameters);
 }
@@ -464,7 +464,7 @@ TOSTRING(neonDrawLighting) ":" NL
     "b .lightStrengthCalculated" NL
 ); // NOLINT
 
-int FELighting::getPowerCoefficients(float exponent)
+int FELightingSoftwareApplier::getPowerCoefficients(float exponent)
 {
     // Calling a powf function from the assembly code would require to save
     // and reload a lot of NEON registers. Since the base is in range [0..1]
--- a/Source/WebCore/platform/graphics/cpu/arm/filters/FELightingNEON.h
+++ b/Source/WebCore/platform/graphics/cpu/arm/filters/FELightingNEON.h
@@ -24,14 +24,15 @@
  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
-#ifndef FELightingNEON_h
-#define FELightingNEON_h
+#pragma once
 
 #if CPU(ARM_NEON) && CPU(ARM_TRADITIONAL) && COMPILER(GCC_COMPATIBLE)
 
-#include "FELighting.h"
+#include "FELightingSoftwareApplier.h"
+#include "ImageBuffer.h"
 #include "PointLightSource.h"
 #include "SpotLightSource.h"
+#include <wtf/ObjectIdentifier.h>
 #include <wtf/ParallelJobs.h>
 
 namespace WebCore {
@@ -93,14 +94,14 @@ extern "C" {
 void neonDrawLighting(FELightingPaintingDataForNeon*);
 }
 
-inline void FELighting::platformApplyNeon(const LightingData& data, const LightSource::PaintingData& paintingData)
+inline void FELightingSoftwareApplier::applyPlatformNeon(const FELightingSoftwareApplier::LightingData& data, const LightSource::PaintingData& paintingData)
 {
-    alignas(16) FELightingFloatArgumentsForNeon floatArguments;
-    FELightingPaintingDataForNeon neonData = {
+    WebCore::FELightingFloatArgumentsForNeon alignas(16) floatArguments;
+    WebCore::FELightingPaintingDataForNeon neonData = {
         data.pixels->data(),
         1,
-        data.widthDecreasedByOne - 1,
-        data.heightDecreasedByOne - 1,
+        data.width - 2,
+        data.height - 2,
         0,
         0,
         0,
@@ -111,23 +112,23 @@ inline void FELighting::platformApplyNeo
     // Set light source arguments.
     floatArguments.constOne = 1;
 
-    auto color = m_lightingColor.toColorTypeLossy<SRGBA<uint8_t>>().resolved();
+    auto color = data.lightingColor.toColorTypeLossy<SRGBA<uint8_t>>().resolved();
 
     floatArguments.colorRed = color.red;
     floatArguments.colorGreen = color.green;
     floatArguments.colorBlue = color.blue;
     floatArguments.padding4 = 0;
 
-    if (m_lightSource->type() == LS_POINT) {
+    if (data.lightSource->type() == LS_POINT) {
         neonData.flags |= FLAG_POINT_LIGHT;
-        PointLightSource& pointLightSource = static_cast<PointLightSource&>(m_lightSource.get());
+        const auto& pointLightSource = *static_cast<const PointLightSource*>(data.lightSource);
         floatArguments.lightX = pointLightSource.position().x();
         floatArguments.lightY = pointLightSource.position().y();
         floatArguments.lightZ = pointLightSource.position().z();
         floatArguments.padding2 = 0;
-    } else if (m_lightSource->type() == LS_SPOT) {
+    } else if (data.lightSource->type() == LS_SPOT) {
         neonData.flags |= FLAG_SPOT_LIGHT;
-        SpotLightSource& spotLightSource = static_cast<SpotLightSource&>(m_lightSource.get());
+        const auto& spotLightSource = *static_cast<const SpotLightSource*>(data.lightSource);
         floatArguments.lightX = spotLightSource.position().x();
         floatArguments.lightY = spotLightSource.position().y();
         floatArguments.lightZ = spotLightSource.position().z();
@@ -145,7 +146,7 @@ inline void FELighting::platformApplyNeo
         if (spotLightSource.specularExponent() == 1)
             neonData.flags |= FLAG_CONE_EXPONENT_IS_1;
     } else {
-        ASSERT(m_lightSource->type() == LS_DISTANT);
+        ASSERT(data.lightSource->type() == LS_DISTANT);
         floatArguments.lightX = paintingData.initialLightingData.lightVector.x();
         floatArguments.lightY = paintingData.initialLightingData.lightVector.y();
         floatArguments.lightZ = paintingData.initialLightingData.lightVector.z();
@@ -155,38 +156,39 @@ inline void FELighting::platformApplyNeo
     // Set lighting arguments.
     floatArguments.surfaceScale = data.surfaceScale;
     floatArguments.minusSurfaceScaleDividedByFour = -data.surfaceScale / 4;
-    if (m_lightingType == FELighting::DiffuseLighting)
-        floatArguments.diffuseConstant = m_diffuseConstant;
+    if (data.filterType == FilterEffect::Type::FEDiffuseLighting)
+        floatArguments.diffuseConstant = data.diffuseConstant;
     else {
         neonData.flags |= FLAG_SPECULAR_LIGHT;
-        floatArguments.diffuseConstant = m_specularConstant;
-        neonData.specularExponent = getPowerCoefficients(m_specularExponent);
-        if (m_specularExponent == 1)
+        floatArguments.diffuseConstant = data.specularConstant;
+        neonData.specularExponent = getPowerCoefficients(data.specularExponent);
+        if (data.specularExponent == 1)
             neonData.flags |= FLAG_SPECULAR_EXPONENT_IS_1;
     }
     if (floatArguments.diffuseConstant == 1)
         neonData.flags |= FLAG_DIFFUSE_CONST_IS_1;
 
-    int optimalThreadNumber = ((data.widthDecreasedByOne - 1) * (data.heightDecreasedByOne - 1)) / s_minimalRectDimension;
+    static constexpr int minimalRectDimension = 100 * 100; // Empirical data limit for parallel jobs
+    int optimalThreadNumber = ((data.width - 2) * (data.height - 2)) / minimalRectDimension;
     if (optimalThreadNumber > 1) {
         // Initialize parallel jobs
-        ParallelJobs<FELightingPaintingDataForNeon> parallelJobs(&WebCore::FELighting::platformApplyNeonWorker, optimalThreadNumber);
+        ParallelJobs<FELightingPaintingDataForNeon> parallelJobs(&FELightingSoftwareApplier::platformApplyNeonWorker, optimalThreadNumber);
 
         // Fill the parameter array
         int job = parallelJobs.numberOfJobs();
         if (job > 1) {
             int yStart = 1;
-            int yStep = (data.heightDecreasedByOne - 1) / job;
+            int yStep = (data.height - 2) / job;
             for (--job; job >= 0; --job) {
                 FELightingPaintingDataForNeon& params = parallelJobs.parameter(job);
                 params = neonData;
                 params.yStart = yStart;
-                params.pixels += (yStart - 1) * (data.widthDecreasedByOne + 1) * 4;
+                params.pixels += (yStart - 1) * data.width * 4;
                 if (job > 0) {
                     params.absoluteHeight = yStep;
                     yStart += yStep;
                 } else
-                    params.absoluteHeight = data.heightDecreasedByOne - yStart;
+                    params.absoluteHeight = (data.height - 1) - yStart;
             }
             parallelJobs.execute();
             return;
@@ -199,5 +201,3 @@ inline void FELighting::platformApplyNeo
 } // namespace WebCore
 
 #endif // CPU(ARM_NEON) && COMPILER(GCC_COMPATIBLE)
-
-#endif // FELightingNEON_h
--- a/Source/WebCore/platform/graphics/filters/DistantLightSource.h
+++ b/Source/WebCore/platform/graphics/filters/DistantLightSource.h
@@ -25,6 +25,10 @@
 #include "LightSource.h"
 #include <wtf/Ref.h>
 
+namespace WTF {
+class TextStream;
+} // namespace WTF
+
 namespace WebCore {
 
 class DistantLightSource : public LightSource {
--- a/Source/WebCore/platform/graphics/filters/FELighting.h
+++ b/Source/WebCore/platform/graphics/filters/FELighting.h
@@ -35,8 +35,6 @@
 
 namespace WebCore {
 
-struct FELightingPaintingDataForNeon;
-
 class FELighting : public FilterEffect {
 public:
     const Color& lightingColor() const { return m_lightingColor; }
@@ -67,11 +65,6 @@ protected:
 
     std::unique_ptr<FilterEffectApplier> createSoftwareApplier() const override;
 
-#if CPU(ARM_NEON) && CPU(ARM_TRADITIONAL) && COMPILER(GCC_COMPATIBLE)
-    static int getPowerCoefficients(float exponent);
-    inline void platformApplyNeon(const LightingData&, const LightSource::PaintingData&);
-#endif
-
     Color m_lightingColor;
     float m_surfaceScale;
     float m_diffuseConstant;
--- a/Source/WebCore/platform/graphics/filters/PointLightSource.h
+++ b/Source/WebCore/platform/graphics/filters/PointLightSource.h
@@ -26,6 +26,10 @@
 #include "LightSource.h"
 #include <wtf/Ref.h>
 
+namespace WTF {
+class TextStream;
+} // namespace WTF
+
 namespace WebCore {
 
 class PointLightSource : public LightSource {
--- a/Source/WebCore/platform/graphics/filters/SpotLightSource.h
+++ b/Source/WebCore/platform/graphics/filters/SpotLightSource.h
@@ -26,6 +26,10 @@
 #include "LightSource.h"
 #include <wtf/Ref.h>
 
+namespace WTF {
+class TextStream;
+} // namespace WTF
+
 namespace WebCore {
 
 class SpotLightSource : public LightSource {
--- a/Source/WebCore/platform/graphics/filters/software/FELightingSoftwareApplier.h
+++ b/Source/WebCore/platform/graphics/filters/software/FELightingSoftwareApplier.h
@@ -36,6 +36,7 @@
 namespace WebCore {
 
 class FELighting;
+struct FELightingPaintingDataForNeon;
 
 class FELightingSoftwareApplier final : public FilterEffectConcreteApplier<FELighting> {
     WTF_MAKE_FAST_ALLOCATED;
@@ -132,8 +133,23 @@ private:
 
     static void applyPlatformGenericPaint(const LightingData&, const LightSource::PaintingData&, int startY, int endY);
     static void applyPlatformGenericWorker(ApplyParameters*);
+
+#if CPU(ARM_NEON) && CPU(ARM_TRADITIONAL) && COMPILER(GCC_COMPATIBLE)
+    static int getPowerCoefficients(float exponent);
+    static void platformApplyNeonWorker(FELightingPaintingDataForNeon*);
+    inline static void applyPlatformNeon(const LightingData&, const LightSource::PaintingData&);
+
+    inline static void applyPlatformGeneric(const LightingData& data, const LightSource::PaintingData& paintingData)
+    {
+        applyPlatformNeon(data, paintingData);
+    }
+#else
     static void applyPlatformGeneric(const LightingData&, const LightSource::PaintingData&);
+#endif
+
     static void applyPlatform(const LightingData&);
 };
 
 } // namespace WebCore
+
+#include "FELightingNEON.h"