specialize arm64 allTrue()/anyTrue()

aarch64 added vector-wise add/mul/min/max instructions.
We can use min and max to implement allTrue() and anyTrue(),
respectively.

(This CL is mostly so I don't forget these intrinsics exist.)

In assembly, these actually compile to two instructions,
the folding operation into a vector register, then a move
from the vector register to a general purpose register.

Change-Id: Ia6a999ac250740de765e871094e911979a8711c7
Reviewed-on: https://skia-review.googlesource.com/116482
Reviewed-by: Chris Dalton <csmartdalton@google.com>
Commit-Queue: Mike Klein <mtklein@chromium.org>

2 files changed, 30 insertions, 0 deletions

diff --git a/src/opts/SkNx_neon.h b/src/opts/SkNx_neon.h
index f8170ef962..52247c8f66 100644
--- a/src/opts/SkNx_neon.h
+++ b/src/opts/SkNx_neon.h
@@ -112,12 +112,20 @@ public:
     }
 
     AI bool allTrue() const {
+    #if defined(__aarch64__)
+        return 0 != vminv_u32(vreinterpret_u32_f32(fVec));
+    #else
         auto v = vreinterpret_u32_f32(fVec);
         return vget_lane_u32(v,0) && vget_lane_u32(v,1);
+    #endif
     }
     AI bool anyTrue() const {
+    #if defined(__aarch64__)
+        return 0 != vmaxv_u32(vreinterpret_u32_f32(fVec));
+    #else
         auto v = vreinterpret_u32_f32(fVec);
         return vget_lane_u32(v,0) || vget_lane_u32(v,1);
+    #endif
     }
 
     AI SkNx thenElse(const SkNx& t, const SkNx& e) const {
@@ -229,14 +237,22 @@ public:
     }
 
     AI bool allTrue() const {
+    #if defined(__aarch64__)
+        return 0 != vminvq_u32(vreinterpretq_u32_f32(fVec));
+    #else
         auto v = vreinterpretq_u32_f32(fVec);
         return vgetq_lane_u32(v,0) && vgetq_lane_u32(v,1)
             && vgetq_lane_u32(v,2) && vgetq_lane_u32(v,3);
+    #endif
     }
     AI bool anyTrue() const {
+    #if defined(__aarch64__)
+        return 0 != vmaxvq_u32(vreinterpretq_u32_f32(fVec));
+    #else
         auto v = vreinterpretq_u32_f32(fVec);
         return vgetq_lane_u32(v,0) || vgetq_lane_u32(v,1)
             || vgetq_lane_u32(v,2) || vgetq_lane_u32(v,3);
+    #endif
     }
 
     AI SkNx thenElse(const SkNx& t, const SkNx& e) const {
diff --git a/tests/SkNxTest.cpp b/tests/SkNxTest.cpp
index 9e2c27e79c..9901f7a41d 100644
--- a/tests/SkNxTest.cpp
+++ b/tests/SkNxTest.cpp
@@ -454,3 +454,17 @@ DEF_TEST(Sk2f_Store4, r) {
     REPORTER_ASSERT(r, dst[6] == 6);
     REPORTER_ASSERT(r, dst[7] == 7);
 }
+
+DEF_TEST(SkNf_anyTrue_allTrue, r) {
+    REPORTER_ASSERT(r,  (Sk2f{1,2} < Sk2f{3,4}).anyTrue());
+    REPORTER_ASSERT(r,  (Sk2f{1,2} < Sk2f{3,4}).allTrue());
+    REPORTER_ASSERT(r,  (Sk2f{3,2} < Sk2f{1,4}).anyTrue());
+    REPORTER_ASSERT(r, !(Sk2f{3,2} < Sk2f{1,4}).allTrue());
+    REPORTER_ASSERT(r, !(Sk2f{3,4} < Sk2f{1,2}).anyTrue());
+
+    REPORTER_ASSERT(r,  (Sk4f{1,2,3,4} < Sk4f{3,4,5,6}).anyTrue());
+    REPORTER_ASSERT(r,  (Sk4f{1,2,3,4} < Sk4f{3,4,5,6}).allTrue());
+    REPORTER_ASSERT(r,  (Sk4f{1,2,3,4} < Sk4f{1,4,1,1}).anyTrue());
+    REPORTER_ASSERT(r, !(Sk4f{1,2,3,4} < Sk4f{1,4,1,1}).allTrue());
+    REPORTER_ASSERT(r, !(Sk4f{3,4,5,6} < Sk4f{1,2,3,4}).anyTrue());
+}