From 7e85d3d10a83b3bd767f30f1bd974c90e086dce6 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Henrik=20Rydg=C3=A5rd?= <hrydgard@gmail.com>
Date: Sat, 9 Dec 2023 14:50:14 +0100
Subject: [PATCH] Disable the new culling on RISC-V for now.

---
 Common/Math/CrossSIMD.h         |  2 +-
 GPU/Common/DrawEngineCommon.cpp | 15 +++++----------
 GPU/GPUCommonHW.cpp             |  9 +++++++++
 ppsspp_config.h                 |  2 ++
 4 files changed, 17 insertions(+), 11 deletions(-)

diff --git a/Common/Math/CrossSIMD.h b/Common/Math/CrossSIMD.h
index c2851b5775b3..3eb8e0e75e2f 100644
--- a/Common/Math/CrossSIMD.h
+++ b/Common/Math/CrossSIMD.h
@@ -10,7 +10,7 @@
 
 #include <cstdint>
 
-#ifdef _M_SSE
+#if PPSSPP_ARCH(SSE2)
 #include <emmintrin.h>
 #endif
 
diff --git a/GPU/Common/DrawEngineCommon.cpp b/GPU/Common/DrawEngineCommon.cpp
index 71ead60a021b..96a81b36b934 100644
--- a/GPU/Common/DrawEngineCommon.cpp
+++ b/GPU/Common/DrawEngineCommon.cpp
@@ -433,12 +433,12 @@ bool DrawEngineCommon::TestBoundingBoxFast(const void *vdata, int vertexCount, u
 		break;
 	case GE_VTYPE_POS_16BIT:
 	{
-#if defined(_M_SSE)
+#if PPSSPP_ARCH(SSE2)
 		__m128 scaleFactor = _mm_set1_ps(1.0f / 32768.0f);
 		for (int i = 0; i < vertexCount; i++) {
 			const s16 *data = ((const s16 *)((const s8 *)vdata + i * stride + offset));
 			__m128i bits = _mm_castpd_si128(_mm_load_sd((const double *)data));
-			// Sign extension. Ugly without SSE4.
+			// Sign extension. Hacky without SSE4.
 			bits = _mm_srai_epi32(_mm_unpacklo_epi16(bits, bits), 16);
 			__m128 pos = _mm_mul_ps(_mm_cvtepi32_ps(bits), scaleFactor);
 			_mm_storeu_ps(verts + i * 3, pos);  // TODO: use stride 4 to avoid clashing writes?
@@ -470,11 +470,7 @@ bool DrawEngineCommon::TestBoundingBoxFast(const void *vdata, int vertexCount, u
 	// We only check the 4 sides. Near/far won't likely make a huge difference.
 	// We test one vertex against 4 planes to get some SIMD. Vertices need to be transformed to world space
 	// for testing, don't want to re-do that, so we have to use that "pivot" of the data.
-<<<<<<< HEAD
-#ifdef _M_SSE
-=======
 #if PPSSPP_ARCH(SSE2)
->>>>>>> c5a94c3799 (Buildfix again)
 	const __m128 worldX = _mm_loadu_ps(gstate.worldMatrix);
 	const __m128 worldY = _mm_loadu_ps(gstate.worldMatrix + 3);
 	const __m128 worldZ = _mm_loadu_ps(gstate.worldMatrix + 6);
@@ -498,9 +494,9 @@ bool DrawEngineCommon::TestBoundingBoxFast(const void *vdata, int vertexCount, u
 		);
 		// OK, now we check it against the four planes.
 		// This is really curiously similar to a matrix multiplication (well, it is one).
-		__m128 posX = _mm_shuffle_ps(worldpos, worldpos, 0);
-		__m128 posY = _mm_shuffle_ps(worldpos, worldpos, 1 | (1 << 2) | (1 << 4) | (1 << 6));
-		__m128 posZ = _mm_shuffle_ps(worldpos, worldpos, 2 | (2 << 2) | (2 << 4) | (2 << 6));
+		__m128 posX = _mm_shuffle_ps(worldpos, worldpos, _MM_SHUFFLE(0, 0, 0, 0));
+		__m128 posY = _mm_shuffle_ps(worldpos, worldpos, _MM_SHUFFLE(1, 1, 1, 1));
+		__m128 posZ = _mm_shuffle_ps(worldpos, worldpos, _MM_SHUFFLE(2, 2, 2, 2));
 		__m128 planeDist = _mm_add_ps(
 			_mm_add_ps(
 				_mm_mul_ps(planeX, posX),
@@ -566,7 +562,6 @@ bool DrawEngineCommon::TestBoundingBoxFast(const void *vdata, int vertexCount, u
 		}
 	}
 #endif
-
 	return true;
 }
 
diff --git a/GPU/GPUCommonHW.cpp b/GPU/GPUCommonHW.cpp
index 8af997b67b64..66906f4b1b47 100644
--- a/GPU/GPUCommonHW.cpp
+++ b/GPU/GPUCommonHW.cpp
@@ -992,8 +992,17 @@ void GPUCommonHW::Execute_Prim(u32 op, u32 diff) {
 
 #define MAX_CULL_CHECK_COUNT 6
 
+// For now, turn off culling on platforms where we don't have SIMD bounding box tests, like RISC-V.
+#if PPSSPP_ARCH(ARM_NEON) || PPSSPP_ARCH(SSE2)
+
 #define PASSES_CULLING ((vertexType & (GE_VTYPE_THROUGH_MASK | GE_VTYPE_MORPHCOUNT_MASK | GE_VTYPE_WEIGHT_MASK | GE_VTYPE_IDX_MASK)) || count > MAX_CULL_CHECK_COUNT)
 
+#else
+
+#define PASSES_CULLING true
+
+#endif
+
 	// If certain conditions are true, do frustum culling.
 	bool passCulling = PASSES_CULLING;
 	if (!passCulling) {
diff --git a/ppsspp_config.h b/ppsspp_config.h
index 2861b621b333..71e7b9c9466d 100644
--- a/ppsspp_config.h
+++ b/ppsspp_config.h
@@ -11,6 +11,7 @@
 #if defined(_M_IX86) || defined(__i386__) || defined (__EMSCRIPTEN__)
     #define PPSSPP_ARCH_X86 1
     #define PPSSPP_ARCH_32BIT 1
+    #define PPSSPP_ARCH_SSE2 1
     //TODO: Remove this compat define
     #ifndef _M_IX86
         #define _M_IX86 600
@@ -19,6 +20,7 @@
 
 #if (defined(_M_X64) || defined(__amd64__) || defined(__x86_64__)) && !defined(__EMSCRIPTEN__)
     #define PPSSPP_ARCH_AMD64 1
+    #define PPSSPP_ARCH_SSE2 1
     #if defined(__ILP32__)
         #define PPSSPP_ARCH_32BIT 1
     #else