ADDED: Some code sample for RISC-V RVV vector instructions -WIP-

Author: raysan5

src/external/rlsw.h

@@ -616,69 +616,74 @@ SWAPI void swBindTexture(uint32_t id);
 #include <math.h>           // Required for: floorf(), fabsf()
 
 #if defined(__FMA__) && defined(__AVX2__)
-#   define SW_HAS_FMA_AVX2
-#   include <immintrin.h>
+    #define SW_HAS_FMA_AVX2
+    #include <immintrin.h>
 #endif
 
 #if defined(__FMA__) && defined(__AVX__)
-#    define SW_HAS_FMA_AVX
-#    include <immintrin.h>
+    #define SW_HAS_FMA_AVX
+    #include <immintrin.h>
 #endif
 
 #if defined(__AVX2__)
-#    define SW_HAS_AVX2
-#    include <immintrin.h>
+    #define SW_HAS_AVX2
+    #include <immintrin.h>
 #endif
 
 #if defined(__AVX__)
-#    define SW_HAS_AVX
-#    include <immintrin.h>
+    #define SW_HAS_AVX
+    #include <immintrin.h>
 #endif
 
 #if defined(__SSE4_2__)
-#    define SW_HAS_SSE42
-#    include <nmmintrin.h>
+    #define SW_HAS_SSE42
+    #include <nmmintrin.h>
 #endif
 
 #if defined(__SSE4_1__)
-#    define SW_HAS_SSE41
-#    include <smmintrin.h>
+    #define SW_HAS_SSE41
+    #include <smmintrin.h>
 #endif
 
 #if defined(__SSSE3__)
-#    define SW_HAS_SSSE3
-#    include <tmmintrin.h>
+    #define SW_HAS_SSSE3
+    #include <tmmintrin.h>
 #endif
 
 #if defined(__SSE3__)
-#    define SW_HAS_SSE3
-#    include <pmmintrin.h>
+    #define SW_HAS_SSE3
+    #include <pmmintrin.h>
 #endif
 
 #if defined(__SSE2__)
-#    define SW_HAS_SSE2
-#    include <emmintrin.h>
+    #define SW_HAS_SSE2
+    #include <emmintrin.h>
 #endif
 
 #if defined(__SSE__)
-#    define SW_HAS_SSE
-#    include <xmmintrin.h>
+    #define SW_HAS_SSE
+    #include <xmmintrin.h>
 #endif
 
 #if defined(__ARM_NEON) || defined(__aarch64__)
-#    if defined(__ARM_FEATURE_FMA)
-#        define SW_HAS_NEON_FMA
-#    else
-#        define SW_HAS_NEON
-#    endif
-#    include <arm_neon.h>
+    #if defined(__ARM_FEATURE_FMA)
+        #define SW_HAS_NEON_FMA
+    #else
+        #define SW_HAS_NEON
+    #endif
+    #include <arm_neon.h>
+#endif
+
+#ifdef __riscv_vector
+    #define SW_HAS_RVV
+    #include <riscv_vector.h>
 #endif
 
 //----------------------------------------------------------------------------------
 // Defines and Macros
 //----------------------------------------------------------------------------------
 #define SW_PI       3.14159265358979323846f
-#define SW_INV_255  0.00392156862745098f
+#define SW_INV_255  0.00392156862745098f            // 1.0f/255.0f
 #define SW_DEG2RAD  (SW_PI/180.0f)
 #define SW_RAD2DEG  (180.0f/SW_PI)
 
@@ -1102,6 +1107,27 @@ static inline void sw_float_to_unorm8_simd(uint8_t dst[4], const float src[4])
     clamped = _mm_packs_epi32(clamped, clamped);    // s32 -> s16 (saturated)
     clamped = _mm_packus_epi16(clamped, clamped);   // s16 -> u8 (saturated < 0 to 0)
     *(uint32_t*)dst = _mm_cvtsi128_si32(clamped);
+#elif defined(SW_HAS_RVV)
+    // TODO: Sample code generated by AI, needs testing and review
+    size_t vl = vsetvl_e32m1(4); // Load up to 4 floats into a vector register
+    vfloat32m1_t vsrc = vle32_v_f32m1(src, vl); // Load float32 values
+
+    // Clamp to [0.0f, 1.0f]
+    vfloat32m1_t vzero = vfmv_v_f_f32m1(0.0f, vl);
+    vfloat32m1_t vone  = vfmv_v_f_f32m1(1.0f, vl);
+    vsrc = vfmin_vv_f32m1(vsrc, vone, vl);
+    vsrc = vfmax_vv_f32m1(vsrc, vzero, vl);
+
+    // Multiply by 255.0f and add 0.5f for rounding
+    vfloat32m1_t vscaled = vfmul_vf_f32m1(vsrc, 255.0f, vl);
+    vscaled = vfadd_vf_f32m1(vscaled, 0.5f, vl);
+
+    // Convert to unsigned integer (truncate toward zero)
+    vuint32m1_t vu32 = vfcvt_xu_f_v_u32m1(vscaled, vl);
+
+    // Narrow from u32 -> u8
+    vuint8m1_t vu8 = vnclipu_wx_u8m1(vu32, 0, vl); // Round toward zero
+    vse8_v_u8m1(dst, vu8, vl); // Store result
 #else
     for (int i = 0; i < 4; i++)
     {
@@ -1123,18 +1149,26 @@ static inline void sw_float_from_unorm8_simd(float dst[4], const uint8_t src[4])
     floats = vmulq_n_f32(floats, SW_INV_255);
     vst1q_f32(dst, floats);
 #elif defined(SW_HAS_SSE41)
-    __m128i bytes = _mm_cvtsi32_si128(*(const uint32_t*)src);
+    __m128i bytes = _mm_cvtsi32_si128(*(const uint32_t *)src);
     __m128i ints = _mm_cvtepu8_epi32(bytes);
     __m128 floats = _mm_cvtepi32_ps(ints);
     floats = _mm_mul_ps(floats, _mm_set1_ps(SW_INV_255));
     _mm_storeu_ps(dst, floats);
 #elif defined(SW_HAS_SSE2)
-    __m128i bytes = _mm_cvtsi32_si128(*(const uint32_t*)src);
+    __m128i bytes = _mm_cvtsi32_si128(*(const uint32_t *)src);
     bytes = _mm_unpacklo_epi8(bytes, _mm_setzero_si128());
     __m128i ints = _mm_unpacklo_epi16(bytes, _mm_setzero_si128());
     __m128 floats = _mm_cvtepi32_ps(ints);
     floats = _mm_mul_ps(floats, _mm_set1_ps(SW_INV_255));
     _mm_storeu_ps(dst, floats);
+#elif defined(SW_HAS_RVV)
+    // TODO: Sample code generated by AI, needs testing and review
+    size_t vl = vsetvl_e8m1(4); // Set vector length for 8-bit input elements
+    vuint8m1_t vsrc_u8 = vle8_v_u8m1(src, vl); // Load 4 unsigned 8-bit integers
+    vuint32m1_t vsrc_u32 = vwcvt_xu_u_v_u32m1(vsrc_u8, vl); // Widen to 32-bit unsigned integers
+    vfloat32m1_t vsrc_f32 = vfcvt_f_xu_v_f32m1(vsrc_u32, vl); // Convert to float32
+    vfloat32m1_t vnorm = vfmul_vf_f32m1(vsrc_f32, SW_INV_255, vl); // Multiply by 1/255.0 to normalize
+    vse32_v_f32m1(dst, vnorm, vl); // Store result
 #else
     dst[0] = (float)src[0]*SW_INV_255;
     dst[1] = (float)src[1]*SW_INV_255;
@@ -2672,8 +2706,8 @@ static inline void FUNC_NAME(void)
     float ySubstep = 1.0f - sw_fract(v0->screen[1]);                            \
                                                                                 \
     /* Calculation of vertex gradients in X and Y */                            \
-    float dUdx, dVdx;                                                           \
-    float dUdy, dVdy;                                                           \
+    float dUdx = 0.0f, dVdx = 0.0f;                                             \
+    float dUdy = 0.0f, dVdy = 0.0f;                                             \
     if (ENABLE_TEXTURE) {                                                       \
         dUdx = (v1->texcoord[0] - v0->texcoord[0])*wRcp;                        \
         dVdx = (v1->texcoord[1] - v0->texcoord[1])*wRcp;                        \