From d904b7e0f03542398b97c106d5c70d099c216fe4 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Henrik=20Rydg=C3=A5rd?= <hrydgard@gmail.com>
Date: Sun, 31 May 2026 19:43:30 +0200
Subject: [PATCH] Loongarch type casting fixes, remove redundant code, buildfix

---
 Common/Math/SIMDHeaders.h               | 41 +++++++------------------
 Windows/.github/copilot-instructions.md | 12 +++++++-
 2 files changed, 22 insertions(+), 31 deletions(-)

diff --git a/Common/Math/SIMDHeaders.h b/Common/Math/SIMDHeaders.h
index 8c740a328c..c7d11e2886 100644
--- a/Common/Math/SIMDHeaders.h
+++ b/Common/Math/SIMDHeaders.h
@@ -31,8 +31,8 @@
 #endif
 #endif
 
-#if PPSSPP_ARCH(LOONGARCH64)
-#if PPSSPP_ARCH(LOONGARCH64_LSX)
+#if PPSSPP_ARCH(LOONGARCH64) && PPSSPP_ARCH(LOONGARCH64_LSX)
+
 #include <lsxintrin.h>
 
 static inline __m128 __lsx_vreplfr2vr_s(float val) {
@@ -45,16 +45,15 @@ static inline __m128 zero_nans_lsx(__m128 val) {
 	// vfcmp.ceq.s compares each float lane.
 	// NaN == NaN is false, so NaNs result in 0x00000000.
 	// Numbers result in 0xFFFFFFFF.
-	__m128 mask = __lsx_vfcmp_ceq_s(val, val);
+	__m128i mask = (__m128i)__lsx_vfcmp_ceq_s(val, val);
 	// Bitwise AND to keep non-NaNs and turn NaNs into 0.0f
-	return __lsx_vand_v(val, mask);
+	return (__m128)__lsx_vand_v((__m128i)val, mask);
 }
 
 static inline __m128 clean_nan_inf_lsx(__m128 val) {
-#if 0
 	// Strip sign bit
 	__m128i sign_mask = __lsx_vreplgr2vr_w(0x7FFFFFFF);
-	__m128i abs_x = __lsx_vand_v(x, sign_mask);
+	__m128i abs_x = __lsx_vand_v((__m128i)val, sign_mask);
 
 	// Infinity threshold
 	__m128i inf_val = __lsx_vreplgr2vr_w(0x7F800000);
@@ -62,28 +61,9 @@ static inline __m128 clean_nan_inf_lsx(__m128 val) {
 	// Compare: Is |x| < Inf? (using signed comparison since 0x7F800000 fits)
 	__m128i finite_mask = __lsx_vslt_w(abs_x, inf_val);
 
-	return __lsx_vand_v(x, finite_mask);
-#else
-	// 1. Broadcast the floating-point limits
-	// (Compilers will typically optimize these to a single vector load)
-	__m128 max_finite = (__m128)__lsx_vreplgr2vr_w(0x7F7FFFFF); // FLT_MAX bits
-	__m128 min_finite = (__m128)__lsx_vreplgr2vr_w(0xFF7FFFFF); // -FLT_MAX bits
-
-	// 2. Clear NaNs and clamp positive infinity
-	__m128 upper_clamped = __lsx_vfmin_s((__m128)x, max_finite);
-
-	// 3. Clamp negative infinity
-	return (__m128i)__lsx_vfmax_s(upper_clamped, min_finite);
-#endif
+	return (__m128)__lsx_vand_v((__m128i)val, finite_mask);
 }
 
-#endif
-#endif
-
-// Basic types
-
-#if PPSSPP_ARCH(LOONGARCH64) && PPSSPP_ARCH(LOONGARCH64_LSX)
-
 static inline __m128i zero_nans_lsx(__m128i x) {
 	// vfcmp.ceq.s compares each float lane.
 	// NaN == NaN is false, so NaNs result in 0x00000000.
@@ -93,10 +73,11 @@ static inline __m128i zero_nans_lsx(__m128i x) {
 	return __lsx_vand_v(x, mask);
 }
 
-__m128 zero_nan_inf_lsx(__m128 x) {
-	// 1. Take absolute value of the single-precision float vector
-	// Emits: vfabs.s vr, vr
-	__m128 x_abs = __lsx_vfabs_s(x);
+static inline __m128 zero_nan_inf_lsx(__m128 x) {
+	// 1. Take absolute value by masking off the sign bit
+	// LSX doesn't have __lsx_vfabs_s, so we compute |x| manually
+	__m128i sign_mask = __lsx_vreplgr2vr_w(0x7FFFFFFF);
+	__m128 x_abs = (__m128)__lsx_vand_v((__m128i)x, sign_mask);
 
 	// 2. Load the Positive Infinity bit pattern (0x7F800000)
 	// Emits: vreplgr2vr.w to duplicate the GPR value into all SIMD lanes
diff --git a/Windows/.github/copilot-instructions.md b/Windows/.github/copilot-instructions.md
index 5ec29a9930..1a38d1cd41 100644
--- a/Windows/.github/copilot-instructions.md
+++ b/Windows/.github/copilot-instructions.md
@@ -1,4 +1,14 @@
 # Copilot Instructions
 
 ## Project Guidelines
-- When working in this PPSSPP workspace, the workspace root is C:\dev\ppsspp\Windows\, so files outside the Windows directory (like GPU/Common/, Core/, etc.) must be accessed with the ".." prefix. For example, use "..\GPU\Common\DrawEngineCommon.cpp" instead of "GPU\Common\DrawEngineCommon.cpp" when calling file editing tools.
+- When working in this PPSSPP workspace, the workspace root is C:\dev\ppsspp\Windows\, so files outside the Windows directory (like GPU/Common/, Core/, etc.)
+  must be accessed with the ".." prefix. For example, use "..\GPU\Common\DrawEngineCommon.cpp" instead of "GPU\Common\DrawEngineCommon.cpp" when calling file editing tools.
+
+## LSX documentation
+
+- When writing code for LSX, the SIMD instruction set for Loongarch64, you can refer to:
+	- http://ppsspp.org/unofficial/lsx/LSX_LASX_REFERENCE.md
+	- http://ppsspp.org/unofficial/lsx/LSX_LASX_DETAILED.md
+	- They might also be available at D:\Data\LSX.
+	- These AI-friendly text files were generated from https://jia.je/unofficial-loongarch-intrinsics-guide/.
+	- Note, verifying by compiling doesn't work, MSVC doesn't support LoongArch so the code is all ifdeff'ed out.