Merge pull request numpy#379 from howjmay/vcgt

feat: Add vcgt* and vcgtz* intrinsics

Author: howjmay

neon2rvv.h

@@ -2638,55 +2638,109 @@ FORCE_INLINE uint32x4_t vcgtq_f32(float32x4_t a, float32x4_t b) {
   return __riscv_vmerge_vvm_u32m1(vdupq_n_u32(0x0), vdupq_n_u32(UINT32_MAX), cmp_res, 4);
 }
 
-// FORCE_INLINE uint64x1_t vcgt_s64(int64x1_t a, int64x1_t b);
+FORCE_INLINE uint64x1_t vcgt_s64(int64x1_t a, int64x1_t b) {
+  vbool64_t cmp_res = __riscv_vmsgt_vv_i64m1_b64(a, b, 1);
+  return __riscv_vmerge_vvm_u64m1(vdupq_n_u64(0x0), vdupq_n_u64(UINT64_MAX), cmp_res, 1);
+}
 
-// FORCE_INLINE uint64x2_t vcgtq_s64(int64x2_t a, int64x2_t b);
+FORCE_INLINE uint64x2_t vcgtq_s64(int64x2_t a, int64x2_t b) {
+  vbool64_t cmp_res = __riscv_vmsgt_vv_i64m1_b64(a, b, 2);
+  return __riscv_vmerge_vvm_u64m1(vdupq_n_u64(0x0), vdupq_n_u64(UINT64_MAX), cmp_res, 2);
+}
 
-// FORCE_INLINE uint64x1_t vcgt_u64(uint64x1_t a, uint64x1_t b);
+FORCE_INLINE uint64x1_t vcgt_u64(uint64x1_t a, uint64x1_t b) {
+  vbool64_t cmp_res = __riscv_vmsgeu_vv_u64m1_b64(a, b, 1);
+  return __riscv_vmerge_vvm_u64m1(vdupq_n_u64(0x0), vdupq_n_u64(UINT64_MAX), cmp_res, 1);
+}
 
-// FORCE_INLINE uint64x2_t vcgtq_u64(uint64x2_t a, uint64x2_t b);
+FORCE_INLINE uint64x2_t vcgtq_u64(uint64x2_t a, uint64x2_t b) {
+  vbool64_t cmp_res = __riscv_vmsgeu_vv_u64m1_b64(a, b, 2);
+  return __riscv_vmerge_vvm_u64m1(vdupq_n_u64(0x0), vdupq_n_u64(UINT64_MAX), cmp_res, 2);
+}
 
-// FORCE_INLINE uint64x1_t vcgt_f64(float64x1_t a, float64x1_t b);
+FORCE_INLINE uint64x1_t vcgt_f64(float64x1_t a, float64x1_t b) {
+  vbool64_t cmp_res = __riscv_vmfgt_vv_f64m1_b64(a, b, 1);
+  return __riscv_vmerge_vvm_u64m1(vdupq_n_u64(0x0), vdupq_n_u64(UINT64_MAX), cmp_res, 1);
+}
 
-// FORCE_INLINE uint64x2_t vcgtq_f64(float64x2_t a, float64x2_t b);
+FORCE_INLINE uint64x2_t vcgtq_f64(float64x2_t a, float64x2_t b) {
+  vbool64_t cmp_res = __riscv_vmfgt_vv_f64m1_b64(a, b, 2);
+  return __riscv_vmerge_vvm_u64m1(vdupq_n_u64(0x0), vdupq_n_u64(UINT64_MAX), cmp_res, 2);
+}
 
-// FORCE_INLINE uint64_t vcgtd_s64(int64_t a, int64_t b);
+FORCE_INLINE uint64_t vcgtd_s64(int64_t a, int64_t b) { return a > b ? UINT64_MAX : 0x0; }
 
-// FORCE_INLINE uint64_t vcgtd_u64(uint64_t a, uint64_t b);
+FORCE_INLINE uint64_t vcgtd_u64(uint64_t a, uint64_t b) { return a > b ? UINT64_MAX : 0x0; }
 
-// FORCE_INLINE uint32_t vcgts_f32(float32_t a, float32_t b);
+FORCE_INLINE uint32_t vcgts_f32(float32_t a, float32_t b) { return a > b ? UINT32_MAX : 0x0; }
 
-// FORCE_INLINE uint64_t vcgtd_f64(float64_t a, float64_t b);
+FORCE_INLINE uint64_t vcgtd_f64(float64_t a, float64_t b) { return a > b ? UINT64_MAX : 0x0; }
 
-// FORCE_INLINE uint8x8_t vcgtz_s8(int8x8_t a);
+FORCE_INLINE uint8x8_t vcgtz_s8(int8x8_t a) {
+  vbool8_t cmp_res = __riscv_vmsgt_vx_i8m1_b8(a, 0, 8);
+  return __riscv_vmerge_vvm_u8m1(vdup_n_u8(0x0), vdup_n_u8(UINT8_MAX), cmp_res, 8);
+}
 
-// FORCE_INLINE uint8x16_t vcgtzq_s8(int8x16_t a);
+FORCE_INLINE uint8x16_t vcgtzq_s8(int8x16_t a) {
+  vbool8_t cmp_res = __riscv_vmsgt_vx_i8m1_b8(a, 0, 16);
+  return __riscv_vmerge_vvm_u8m1(vdupq_n_u8(0x0), vdupq_n_u8(UINT8_MAX), cmp_res, 16);
+}
 
-// FORCE_INLINE uint16x4_t vcgtz_s16(int16x4_t a);
+FORCE_INLINE uint16x4_t vcgtz_s16(int16x4_t a) {
+  vbool16_t cmp_res = __riscv_vmsgt_vx_i16m1_b16(a, 0, 4);
+  return __riscv_vmerge_vvm_u16m1(vdup_n_u16(0x0), vdup_n_u16(UINT16_MAX), cmp_res, 4);
+}
 
-// FORCE_INLINE uint16x8_t vcgtzq_s16(int16x8_t a);
+FORCE_INLINE uint16x8_t vcgtzq_s16(int16x8_t a) {
+  vbool16_t cmp_res = __riscv_vmsgt_vx_i16m1_b16(a, 0, 8);
+  return __riscv_vmerge_vvm_u16m1(vdupq_n_u16(0x0), vdupq_n_u16(UINT16_MAX), cmp_res, 8);
+}
 
-// FORCE_INLINE uint32x2_t vcgtz_s32(int32x2_t a);
+FORCE_INLINE uint32x2_t vcgtz_s32(int32x2_t a) {
+  vbool32_t cmp_res = __riscv_vmsgt_vx_i32m1_b32(a, 0, 2);
+  return __riscv_vmerge_vvm_u32m1(vdup_n_u32(0x0), vdup_n_u32(UINT32_MAX), cmp_res, 2);
+}
 
-// FORCE_INLINE uint32x4_t vcgtzq_s32(int32x4_t a);
+FORCE_INLINE uint32x4_t vcgtzq_s32(int32x4_t a) {
+  vbool32_t cmp_res = __riscv_vmsgt_vx_i32m1_b32(a, 0, 4);
+  return __riscv_vmerge_vvm_u32m1(vdupq_n_u32(0x0), vdupq_n_u32(UINT32_MAX), cmp_res, 4);
+}
 
-// FORCE_INLINE uint64x1_t vcgtz_s64(int64x1_t a);
+FORCE_INLINE uint64x1_t vcgtz_s64(int64x1_t a) {
+  vbool64_t cmp_res = __riscv_vmsgt_vx_i64m1_b64(a, 0, 1);
+  return __riscv_vmerge_vvm_u64m1(vdupq_n_u64(0x0), vdupq_n_u64(UINT64_MAX), cmp_res, 1);
+}
 
-// FORCE_INLINE uint64x2_t vcgtzq_s64(int64x2_t a);
+FORCE_INLINE uint64x2_t vcgtzq_s64(int64x2_t a) {
+  vbool64_t cmp_res = __riscv_vmsgt_vx_i64m1_b64(a, 0, 2);
+  return __riscv_vmerge_vvm_u64m1(vdupq_n_u64(0x0), vdupq_n_u64(UINT64_MAX), cmp_res, 2);
+}
 
-// FORCE_INLINE uint32x2_t vcgtz_f32(float32x2_t a);
+FORCE_INLINE uint32x2_t vcgtz_f32(float32x2_t a) {
+  vbool32_t cmp_res = __riscv_vmfgt_vf_f32m1_b32(a, 0, 2);
+  return __riscv_vmerge_vvm_u32m1(vdup_n_u32(0x0), vdup_n_u32(UINT32_MAX), cmp_res, 2);
+}
 
-// FORCE_INLINE uint32x4_t vcgtzq_f32(float32x4_t a);
+FORCE_INLINE uint32x4_t vcgtzq_f32(float32x4_t a) {
+  vbool32_t cmp_res = __riscv_vmfgt_vf_f32m1_b32(a, 0, 4);
+  return __riscv_vmerge_vvm_u32m1(vdupq_n_u32(0x0), vdupq_n_u32(UINT32_MAX), cmp_res, 4);
+}
 
-// FORCE_INLINE uint64x1_t vcgtz_f64(float64x1_t a);
+FORCE_INLINE uint64x1_t vcgtz_f64(float64x1_t a) {
+  vbool64_t cmp_res = __riscv_vmfgt_vf_f64m1_b64(a, 0, 1);
+  return __riscv_vmerge_vvm_u64m1(vdupq_n_u64(0x0), vdupq_n_u64(UINT64_MAX), cmp_res, 1);
+}
 
-// FORCE_INLINE uint64x2_t vcgtzq_f64(float64x2_t a);
+FORCE_INLINE uint64x2_t vcgtzq_f64(float64x2_t a) {
+  vbool64_t cmp_res = __riscv_vmfgt_vf_f64m1_b64(a, 0, 2);
+  return __riscv_vmerge_vvm_u64m1(vdupq_n_u64(0x0), vdupq_n_u64(UINT64_MAX), cmp_res, 2);
+}
 
-// FORCE_INLINE uint64_t vcgtzd_s64(int64_t a);
+FORCE_INLINE uint64_t vcgtzd_s64(int64_t a) { return (a > 0) ? UINT64_MAX : 0x00; }
 
-// FORCE_INLINE uint32_t vcgtzs_f32(float32_t a);
+FORCE_INLINE uint32_t vcgtzs_f32(float32_t a) { return (a > 0) ? UINT32_MAX : 0x00; }
 
-// FORCE_INLINE uint64_t vcgtzd_f64(float64_t a);
+FORCE_INLINE uint64_t vcgtzd_f64(float64_t a) { return (a > 0) ? UINT64_MAX : 0x00; }
 
 FORCE_INLINE uint8x16_t vcgtq_u8(uint8x16_t a, uint8x16_t b) {
   vbool8_t cmp_res = __riscv_vmsgtu_vv_u8m1_b8(a, b, 16);