Merge pull request #12217 from unknownbrackets/vfpu-dot

Merge vfpu-dot changes and add compat flag for Tekken
hrydgard · Aug 5, 2019 · 2ba998d · 2ba998d
2 parents dc9e794 + 85b9ad8
commit 2ba998d
Show file tree

Hide file tree

Showing 11 changed files with 478 additions and 63 deletions.
diff --git a/Common/BitScan.h b/Common/BitScan.h
@@ -1,6 +1,7 @@
 #pragma once
 
 #include "ppsspp_config.h"
+#include <cstdint>
 
 #if PPSSPP_PLATFORM(WINDOWS)
 #include "Common/CommonWindows.h"

diff --git a/Core/Compatibility.cpp b/Core/Compatibility.cpp
@@ -65,6 +65,7 @@ void Compatibility::CheckSettings(IniFile &iniFile, const std::string &gameID) {
 	CheckSetting(iniFile, gameID, "ForceMax60FPS", &flags_.ForceMax60FPS);
 	CheckSetting(iniFile, gameID, "JitInvalidationHack", &flags_.JitInvalidationHack);
 	CheckSetting(iniFile, gameID, "HideISOFiles", &flags_.HideISOFiles);
+	CheckSetting(iniFile, gameID, "MoreAccurateVMMUL", &flags_.MoreAccurateVMMUL);
 }
 
 void Compatibility::CheckSetting(IniFile &iniFile, const std::string &gameID, const char *option, bool *flag) {

diff --git a/Core/Compatibility.h b/Core/Compatibility.h
@@ -65,6 +65,7 @@ struct CompatFlags {
 	bool ForceMax60FPS;
 	bool JitInvalidationHack;
 	bool HideISOFiles;
+	bool MoreAccurateVMMUL;
 };
 
 class IniFile;

diff --git a/Core/MIPS/ARM/ArmCompVFPU.cpp b/Core/MIPS/ARM/ArmCompVFPU.cpp
@@ -21,14 +21,16 @@
 #include <cmath>
 #include "math/math_util.h"
 
+#include "Core/Compatibility.h"
+#include "Core/Config.h"
 #include "Core/MemMap.h"
+#include "Core/Reporting.h"
+#include "Core/System.h"
 #include "Core/MIPS/MIPS.h"
 #include "Core/MIPS/MIPSTables.h"
 #include "Core/MIPS/MIPSAnalyst.h"
 #include "Core/MIPS/MIPSCodeUtils.h"
 #include "Common/CPUDetect.h"
-#include "Core/Config.h"
-#include "Core/Reporting.h"
 
 #include "Core/MIPS/ARM/ArmJit.h"
 #include "Core/MIPS/ARM/ArmRegCache.h"
@@ -1468,12 +1470,16 @@ namespace MIPSComp
 
 	void ArmJit::Comp_Vmmul(MIPSOpcode op) {
 		CONDITIONAL_DISABLE(VFPU_MTX_VMMUL);
-		if (js.HasUnknownPrefix()) {
+		if (!js.HasNoPrefix()) {
 			DISABLE;
 		}
 		NEON_IF_AVAILABLE(CompNEON_Vmmul);
 
-		// TODO: This probably ignores prefixes?
+		if (PSP_CoreParameter().compat.flags().MoreAccurateVMMUL) {
+			// Fall back to interpreter, which has the accurate implementation.
+			// Later we might do something more optimized here.
+			DISABLE;
+		}
 
 		MatrixSize sz = GetMtxSize(op);
 		int n = GetMatrixSide(sz);

diff --git a/Core/MIPS/ARM64/Arm64CompVFPU.cpp b/Core/MIPS/ARM64/Arm64CompVFPU.cpp
@@ -21,15 +21,16 @@
 #include <cmath>
 #include "math/math_util.h"
 
+#include "Core/Compatibility.h"
+#include "Core/Config.h"
 #include "Core/MemMap.h"
+#include "Core/Reporting.h"
+#include "Core/System.h"
 #include "Core/MIPS/MIPS.h"
 #include "Core/MIPS/MIPSTables.h"
 #include "Core/MIPS/MIPSAnalyst.h"
 #include "Core/MIPS/MIPSCodeUtils.h"
 #include "Common/CPUDetect.h"
-#include "Core/Config.h"
-#include "Core/Reporting.h"
-
 #include "Common/Arm64Emitter.h"
 #include "Core/MIPS/ARM64/Arm64Jit.h"
 #include "Core/MIPS/ARM64/Arm64RegCache.h"
@@ -1219,6 +1220,12 @@ namespace MIPSComp {
 			DISABLE;
 		}
 
+		if (PSP_CoreParameter().compat.flags().MoreAccurateVMMUL) {
+			// Fall back to interpreter, which has the accurate implementation.
+			// Later we might do something more optimized here.
+			DISABLE;
+		}
+
 		MatrixSize sz = GetMtxSize(op);
 		int n = GetMatrixSide(sz);
 

diff --git a/Core/MIPS/IR/IRCompVFPU.cpp b/Core/MIPS/IR/IRCompVFPU.cpp
@@ -20,6 +20,7 @@
 #include "math/math_util.h"
 
 #include "Common/CPUDetect.h"
+#include "Core/Compatibility.h"
 #include "Core/Config.h"
 #include "Core/MemMap.h"
 #include "Core/MIPS/MIPS.h"
@@ -29,6 +30,7 @@
 #include "Core/MIPS/IR/IRFrontend.h"
 #include "Core/MIPS/IR/IRRegCache.h"
 #include "Core/Reporting.h"
+#include "Core/System.h"
 
 
 // All functions should have CONDITIONAL_DISABLE, so we can narrow things down to a file quickly.
@@ -1242,6 +1244,12 @@ namespace MIPSComp {
 			DISABLE;
 		}
 
+		if (PSP_CoreParameter().compat.flags().MoreAccurateVMMUL) {
+			// Fall back to interpreter, which has the accurate implementation.
+			// Later we might do something more optimized here.
+			DISABLE;
+		}
+
 		// Matrix multiply (weird prefixes)
 		// D[0 .. N, 0 .. M] = S[0 .. N, 0 .. M]' * T[0 .. N, 0 .. M]
 		// Note: Behaves as if it's implemented through a series of vdots.