RVV port to inline from intrinsics

- avoids intrinsics API changes
- build options for rvv not needed for inline instructions
- declare vl and vtype clobbers instead of cc
- tune for SiFive X280 using 16/32-bit loads and stores instead of
  vlseg/vsseg where more efficient
- tune LMUL for SpacemiT K3 (256-bit vectors) with max m4 (m2 where
  constrained by register pressure or vlseg/vsseg field rules)
- YUVTORGB use vmulhu

NV12ToARGB performance estimates:
CPU           VLEN      C       Intrinsics  Inline ASM  Speedup (Inline)
------------------------------------------------------------------------
SiFive X390  1024-bit  360 ms     29 ms       9.5 ms     37.89x (3.05x)
SiFive X280   512-bit  980 ms    115 ms      38.0 ms     25.79x (3.03x)
SpacemiT K3   256-bit  420 ms     68 ms      28.0 ms     15.00x (2.43x)
Google Coral  256-bit 1120 ms    190 ms      58.0 ms     19.31x (3.28x)
T-Head C920   128-bit  510 ms     88 ms      39.0 ms     13.08x (2.26x)
T-Head C908   128-bit  860 ms    145 ms      62.0 ms     13.87x (2.34x)

Bug: b/443984359
Change-Id: If4097b99ced4cc2306f12236b8605994e713eecd
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/8426529
Reviewed-by: richard winterton <rrwinterton@gmail.com>
Commit-Queue: Frank Barchard <fbarchard@google.com>

README.chromium

@@ -1,6 +1,6 @@
 Name: libyuv
 URL: https://chromium.googlesource.com/libyuv/libyuv/
-Version: 1971
+Version: 1972
 Revision: DEPS
 License: BSD-3-Clause
 License File: LICENSE

include/libyuv/row.h

@@ -627,14 +627,17 @@ extern "C" {
 #define HAS_ARGBTOYMATRIXROW_SVE2
 #define HAS_AYUVTOUVROW_SVE2
 #define HAS_AYUVTOVUROW_SVE2
-// #define HAS_CONVERT16TO8ROW_SVE2  // Disabled: NEON version is faster for 128 bit vectors.
+// Disabled: NEON version is faster for 128 bit vectors.
+// #define HAS_CONVERT16TO8ROW_SVE2
 #define HAS_CONVERT8TO8ROW_SVE2
 #define HAS_COPYROW_SVE2
 #define HAS_DIVIDEROW_16_SVE2
 #define HAS_HALFFLOATROW_SVE2
 #define HAS_HALFMERGEUVROW_SVE2
-// #define HAS_HALFROW_16TO8_SVE2  // Disabled: NEON version is faster for 128 bit vectors.
-// #define HAS_HALFWIDTHROW_16TO8_SVE2  // Disabled: NEON version is faster for 128 bit vectors.
+// Disabled: NEON version is faster for 128 bit vectors.
+// #define HAS_HALFROW_16TO8_SVE2
+// Disabled: NEON version is faster for 128 bit vectors.
+// #define HAS_HALFWIDTHROW_16TO8_SVE2
 #define HAS_I210ALPHATOARGBROW_SVE2
 #define HAS_I210TOAR30ROW_SVE2
 #define HAS_I210TOARGBROW_SVE2
@@ -857,20 +860,6 @@ extern "C" {
 #define HAS_YUY2TOYROW_LASX
 #endif
 
-#if !defined(LIBYUV_DISABLE_RVV) && defined(__riscv_vector)
-#if defined(__riscv_v_intrinsic) && __riscv_v_intrinsic >= 100000
-// Since v1.0, vcreate intrinsic is introduced
-#define LIBYUV_RVV_HAS_VCREATE
-#endif
-#if defined(__riscv_v_intrinsic) && __riscv_v_intrinsic >= 12000
-// Since v0.12, TUPLE_TYPE is introduced for segment load and store.
-#define LIBYUV_RVV_HAS_TUPLE_TYPE
-// Since v0.12, VXRM(fixed-point rounding mode) is included in arguments of
-// fixed-point intrinsics.
-#define LIBYUV_RVV_HAS_VXRM_ARG
-#endif
-#endif
-
 #if !defined(LIBYUV_DISABLE_RVV) && defined(__riscv_vector)
 #define HAS_AR64TOARGBROW_RVV
 #define HAS_ARGBCOPYYTOALPHAROW_RVV
@@ -896,20 +885,6 @@ extern "C" {
 #define HAS_SPLITUVROW_RVV
 #define HAS_SPLITXRGBROW_RVV
 #define HAS_SWAPUVROW_RVV
-
-// The following are available for RVV v0.11 and RVV v1.0
-// TODO(fbarchard): Port to RVV v0.12 (tuple)
-// missing support for vcreate_v:
-//  __riscv_vcreate_v_u16m2x2
-//  __riscv_vcreate_v_u16m2x4
-//  __riscv_vcreate_v_u16m4x2
-//  __riscv_vcreate_v_u8m1x3
-//  __riscv_vcreate_v_u8m1x4
-//  __riscv_vcreate_v_u8m2x2
-//  __riscv_vcreate_v_u8m2x3
-//  __riscv_vcreate_v_u8m2x4
-//  __riscv_vcreate_v_u8m4x2
-#if defined(LIBYUV_RVV_HAS_VCREATE)
 #define HAS_AB64TOARGBROW_RVV
 #define HAS_AR64TOAB64ROW_RVV
 #define HAS_ARGBATTENUATEROW_RVV
@@ -944,7 +919,6 @@ extern "C" {
 #define HAS_RGB24TOARGBROW_RVV
 #define HAS_RGBATOARGBROW_RVV
 #endif
-#endif
 
 #if defined(_MSC_VER) && !defined(__CLR_VER) && !defined(__clang__)
 #if defined(VISUALC_HAS_AVX2)

include/libyuv/scale_row.h

@@ -151,19 +151,9 @@ extern "C" {
 #define HAS_SCALEROWDOWN34_LSX
 #endif
 
-// The following are available on RVV with 64 bit elements
-// TODO: Update compiler to support 64 bit
-#if !defined(LIBYUV_DISABLE_RVV) && defined(__riscv_vector) && \
-    defined(__riscv_zve64x)
+#if !defined(LIBYUV_DISABLE_RVV) && defined(__riscv_vector)
 #define HAS_SCALEUVROWDOWN4_RVV
 #define HAS_SCALEARGBROWDOWN2_RVV
-#endif
-
-#if !defined(LIBYUV_DISABLE_RVV) && defined(__riscv_vector) && \
-    defined(__riscv_v_intrinsic)
-// The following are available on RVV v0.11 and RVV v1.0
-// TODO: Port to RVV v0.12
-#if __riscv_v_intrinsic == 11000 || __riscv_v_intrinsic >= 100000
 #define HAS_SCALEROWDOWN34_0_BOX_RVV
 #define HAS_SCALEROWDOWN34_1_BOX_RVV
 #define HAS_SCALEROWDOWN38_2_BOX_RVV
@@ -190,13 +180,8 @@ extern "C" {
 #define HAS_SCALEUVROWDOWN2_RVV
 #define HAS_SCALEUVROWDOWN2BOX_RVV
 #define HAS_SCALEUVROWDOWN2LINEAR_RVV
-#endif
-
-// The following are available on RVV v0.11
-#if __riscv_v_intrinsic == 11000
 #define HAS_SCALEARGBFILTERCOLS_RVV
 #endif
-#endif
 
 // Scale ARGB vertically with bilinear interpolation.
 void ScalePlaneVertical(int src_height,

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 1971
+#define LIBYUV_VERSION 1972
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_

source/cpu_id.cc

@@ -304,7 +304,7 @@ LIBYUV_API SAFEBUFFERS int RiscvCpuCaps(const char* cpuinfo_name) {
   int flag = 0;
   FILE* f = fopen(cpuinfo_name, "re");
   if (!f) {
-#if defined(__riscv_vector)
+#if !defined(LIBYUV_DISABLE_RVV) && defined(__riscv) && __riscv_xlen == 64
     // Assume RVV if /proc/cpuinfo is unavailable.
     // This will occur for Chrome sandbox for Pepper or Render process.
     return kCpuHasRVV;
@@ -364,7 +364,7 @@ LIBYUV_API SAFEBUFFERS int RiscvCpuCaps(const char* cpuinfo_name) {
         }
       }
     }
-#if defined(__riscv_vector)
+#if !defined(LIBYUV_DISABLE_RVV) && defined(__riscv) && __riscv_xlen == 64
     // Assume RVV if /proc/cpuinfo is from x86 host running QEMU.
     else if ((memcmp(cpuinfo_line, "vendor_id\t: GenuineIntel", 24) == 0) ||
              (memcmp(cpuinfo_line, "vendor_id\t: AuthenticAMD", 24) == 0)) {