skcpu: sse4.1 floor, f16c f16<->f32

  -  floor with roundps is about 4.5x faster when available
  -  f16 srcover_n is similar to but a little faster than the version in https://codereview.chromium.org/1884683002.  This new one fuses the dst load/stores into the f16<->f32 conversions:

+0x180	    movups              (%r15), %xmm1
+0x184	    vcvtph2ps           (%rbx), %xmm2
+0x189	    movaps              %xmm1, %xmm3
+0x18c	    shufps              $255, %xmm3, %xmm3
+0x190	    movaps              %xmm0, %xmm4
+0x193	    subps               %xmm3, %xmm4
+0x196	    mulps               %xmm2, %xmm4
+0x199	    addps               %xmm1, %xmm4
+0x19c	    vcvtps2ph           $0, %xmm4, (%rbx)
+0x1a2	    addq                $16, %r15
+0x1a6	    addq                $8, %rbx
+0x1aa	    decl                %r14d
+0x1ad	    jne                 +0x180

If we decide to land this it'd be a good idea to convert most or all users of SkFloatToHalf_01 and SkHalfToFloat_01 over to the pointer-based versions.

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=1891513002
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot

Review URL: https://codereview.chromium.org/1891513002

1 files changed, 30 insertions, 0 deletions

diff --git a/src/core/SkHalf.h b/src/core/SkHalf.h
index 5f5575ae1a..e56c78c212 100644
--- a/src/core/SkHalf.h
+++ b/src/core/SkHalf.h
@@ -8,6 +8,7 @@
 #ifndef SkHalf_DEFINED
 #define SkHalf_DEFINED
 
+#include "SkCpu.h"
 #include "SkNx.h"
 #include "SkTypes.h"
 
@@ -122,3 +123,32 @@ static inline uint64_t SkFloatToHalf_01(const Sk4f& fs) {
 }
 
 #endif
+
+static inline Sk4f SkHalfToFloat_01(const uint64_t* hs) {
+#if !defined(SKNX_NO_SIMD) && SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSE2
+    if (SkCpu::Supports(SkCpu::F16C)) {
+        __m128 fs;
+    #if defined(_MSC_VER)
+        fs = _mm_cvtph_ps(_mm_loadl_epi64((const __m128i*)hs));
+    #else
+        asm("vcvtph2ps %[hs], %[fs]" : [fs]"=x"(fs) : [hs]"m"(*hs));
+    #endif
+        return fs;
+    }
+#endif
+    return SkHalfToFloat_01(*hs);
+}
+
+static inline void SkFloatToHalf_01(const Sk4f& fs, uint64_t* hs) {
+#if !defined(SKNX_NO_SIMD) && SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSE2
+    if (SkCpu::Supports(SkCpu::F16C)) {
+    #if defined(_MSC_VER)
+        _mm_storel_epi64((__m128i*)hs, _mm_cvtps_ph(fs.fVec, 0));
+    #else
+        asm("vcvtps2ph $0, %[fs], %[hs]" : [hs]"=m"(*hs) : [fs]"x"(fs.fVec));
+    #endif
+        return;
+    }
+#endif
+    *hs = SkFloatToHalf_01(fs);
+}