9a885b27f3
Yet another surprising finding when looking at ARM code generation is that passing these values to functions by const& does make a difference, even when fully inlined. I can only guess that the compiler's somehow more sure that way that the values won't change? Anyway, convert all skvx functions that take Vec arguments to take const Vec& instead. This tweak is enough to let the natural implementation of mull() actually produce good code generation, so I've promoted that to SkVx.h and added a unit test. Notice in the NEON case we've got a base case at N=8 and two recursive cases, one down to 8 as usual when N > 8, but also one up to 8 when N < 8. This also is another big speedup for ARMv7 NEON, bringing it to nearly the same speed as ARMv8 NEON on the same device. Bug: chromium:952502 Change-Id: I0f19bab45cf02222ccc8090053ea2a4a380f1dfe Reviewed-on: https://skia-review.googlesource.com/c/skia/+/208582 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Auto-Submit: Mike Klein <mtklein@google.com> Reviewed-by: Michael Ludwig <michaelludwig@google.com>
182 lines
6.9 KiB
C++
182 lines
6.9 KiB
C++
/*
|
|
* Copyright 2019 Google Inc.
|
|
*
|
|
* Use of this source code is governed by a BSD-style license that can be
|
|
* found in the LICENSE file.
|
|
*/
|
|
|
|
#include "SkVx.h"
|
|
#include "Test.h"
|
|
|
|
using float2 = skvx::Vec<2,float>;
|
|
using float4 = skvx::Vec<4,float>;
|
|
using float8 = skvx::Vec<8,float>;
|
|
|
|
using double2 = skvx::Vec<2,double>;
|
|
using double4 = skvx::Vec<4,double>;
|
|
using double8 = skvx::Vec<8,double>;
|
|
|
|
using byte2 = skvx::Vec< 2,uint8_t>;
|
|
using byte4 = skvx::Vec< 4,uint8_t>;
|
|
using byte8 = skvx::Vec< 8,uint8_t>;
|
|
using byte16 = skvx::Vec<16,uint8_t>;
|
|
|
|
using int2 = skvx::Vec<2,int32_t>;
|
|
using int4 = skvx::Vec<4,int32_t>;
|
|
using int8 = skvx::Vec<8,int32_t>;
|
|
|
|
using long2 = skvx::Vec<2,int64_t>;
|
|
using long4 = skvx::Vec<4,int64_t>;
|
|
using long8 = skvx::Vec<8,int64_t>;
|
|
|
|
// These are unused, and just here so I can look at the disassembly.
|
|
float2 Sqrt(float2 x) { return sqrt(x); }
|
|
float4 Sqrt(float4 x) { return sqrt(x); }
|
|
float8 Sqrt(float8 x) { return sqrt(x); }
|
|
|
|
float4 RSqrt(float4 x) { return rsqrt(x); }
|
|
float4 Rcp(float4 x) { return rcp(x); }
|
|
float4 Ceil(float4 x) { return ceil(x); }
|
|
float4 Floor(float4 x) { return floor(x); }
|
|
float4 Trunc(float4 x) { return trunc(x); }
|
|
float4 Round(float4 x) { return round(x); }
|
|
float4 Abs(float4 x) { return abs(x); }
|
|
|
|
float4 Min(float4 x, float4 y) { return min(x,y); }
|
|
float4 Max(float4 x, float4 y) { return max(x,y); }
|
|
|
|
float4 IfThenElse(int4 c, float4 t, float4 e) { return if_then_else(c,t,e); }
|
|
|
|
DEF_TEST(SkVx, r) {
|
|
static_assert(sizeof(float2) == 8, "");
|
|
static_assert(sizeof(float4) == 16, "");
|
|
static_assert(sizeof(float8) == 32, "");
|
|
|
|
static_assert(sizeof(byte2) == 2, "");
|
|
static_assert(sizeof(byte4) == 4, "");
|
|
static_assert(sizeof(byte8) == 8, "");
|
|
|
|
{
|
|
int4 mask = float4{1,2,3,4} < float4{1,2,4,8};
|
|
REPORTER_ASSERT(r, mask[0] == int32_t( 0));
|
|
REPORTER_ASSERT(r, mask[1] == int32_t( 0));
|
|
REPORTER_ASSERT(r, mask[2] == int32_t(-1));
|
|
REPORTER_ASSERT(r, mask[3] == int32_t(-1));
|
|
|
|
REPORTER_ASSERT(r, any(mask));
|
|
REPORTER_ASSERT(r, !all(mask));
|
|
}
|
|
|
|
{
|
|
long4 mask = double4{1,2,3,4} < double4{1,2,4,8};
|
|
REPORTER_ASSERT(r, mask[0] == int64_t( 0));
|
|
REPORTER_ASSERT(r, mask[1] == int64_t( 0));
|
|
REPORTER_ASSERT(r, mask[2] == int64_t(-1));
|
|
REPORTER_ASSERT(r, mask[3] == int64_t(-1));
|
|
|
|
REPORTER_ASSERT(r, any(mask));
|
|
REPORTER_ASSERT(r, !all(mask));
|
|
}
|
|
|
|
REPORTER_ASSERT(r, min(float4{1,2,3,4}) == 1);
|
|
REPORTER_ASSERT(r, max(float4{1,2,3,4}) == 4);
|
|
|
|
REPORTER_ASSERT(r, all(int4{1,2,3,4,5} == int4{1,2,3,4}));
|
|
REPORTER_ASSERT(r, all(int4{1,2,3,4} == int4{1,2,3,4}));
|
|
REPORTER_ASSERT(r, all(int4{1,2,3} == int4{1,2,3,0}));
|
|
REPORTER_ASSERT(r, all(int4{1,2} == int4{1,2,0,0}));
|
|
REPORTER_ASSERT(r, all(int4{1} == int4{1,0,0,0}));
|
|
REPORTER_ASSERT(r, all(int4(1) == int4{1,1,1,1}));
|
|
REPORTER_ASSERT(r, all(int4{} == int4{0,0,0,0}));
|
|
REPORTER_ASSERT(r, all(int4() == int4{0,0,0,0}));
|
|
|
|
REPORTER_ASSERT(r, all(int4{1,2,2,1} == min(int4{1,2,3,4}, int4{4,3,2,1})));
|
|
REPORTER_ASSERT(r, all(int4{4,3,3,4} == max(int4{1,2,3,4}, int4{4,3,2,1})));
|
|
|
|
REPORTER_ASSERT(r, all(if_then_else(float4{1,2,3,2} <= float4{2,2,2,2}, float4(42), float4(47))
|
|
== float4{42,42,47,42}));
|
|
|
|
REPORTER_ASSERT(r, all(floor(float4{-1.5f,1.5f,1.0f,-1.0f}) == float4{-2.0f,1.0f,1.0f,-1.0f}));
|
|
REPORTER_ASSERT(r, all( ceil(float4{-1.5f,1.5f,1.0f,-1.0f}) == float4{-1.0f,2.0f,1.0f,-1.0f}));
|
|
REPORTER_ASSERT(r, all(trunc(float4{-1.5f,1.5f,1.0f,-1.0f}) == float4{-1.0f,1.0f,1.0f,-1.0f}));
|
|
REPORTER_ASSERT(r, all(round(float4{-1.5f,1.5f,1.0f,-1.0f}) == float4{-2.0f,2.0f,1.0f,-1.0f}));
|
|
|
|
|
|
REPORTER_ASSERT(r, all(abs(float4{-2,-1,0,1}) == float4{2,1,0,1}));
|
|
|
|
// TODO(mtklein): these tests could be made less loose.
|
|
REPORTER_ASSERT(r, all( sqrt(float4{2,3,4,5}) < float4{2,2,3,3}));
|
|
REPORTER_ASSERT(r, all( rcp(float4{2,3,4,5}) < float4{1.0f,0.5f,0.5f,0.3f}));
|
|
REPORTER_ASSERT(r, all(rsqrt(float4{2,3,4,5}) < float4{1.0f,1.0f,1.0f,0.5f}));
|
|
|
|
REPORTER_ASSERT(r, all( sqrt(float2{2,3}) < float2{2,2}));
|
|
REPORTER_ASSERT(r, all( rcp(float2{2,3}) < float2{1.0f,0.5f}));
|
|
REPORTER_ASSERT(r, all(rsqrt(float2{2,3}) < float2{1.0f,1.0f}));
|
|
|
|
REPORTER_ASSERT(r, all(skvx::cast<int>(float4{-1.5f,0.5f,1.0f,1.5f}) == int4{-1,0,1,1}));
|
|
|
|
float buf[] = {1,2,3,4,5,6};
|
|
REPORTER_ASSERT(r, all(float4::Load(buf) == float4{1,2,3,4}));
|
|
float4{2,3,4,5}.store(buf);
|
|
REPORTER_ASSERT(r, buf[0] == 2
|
|
&& buf[1] == 3
|
|
&& buf[2] == 4
|
|
&& buf[3] == 5
|
|
&& buf[4] == 5
|
|
&& buf[5] == 6);
|
|
REPORTER_ASSERT(r, all(float4::Load(buf+0) == float4{2,3,4,5}));
|
|
REPORTER_ASSERT(r, all(float4::Load(buf+2) == float4{4,5,5,6}));
|
|
|
|
REPORTER_ASSERT(r, all(mad(float4{1,2,3,4}, 2.0f, 3.0f) == float4{5,7,9,11}));
|
|
|
|
REPORTER_ASSERT(r, all(skvx::shuffle<2,1,0,3> (float4{1,2,3,4}) == float4{3,2,1,4}));
|
|
REPORTER_ASSERT(r, all(skvx::shuffle<2,1> (float4{1,2,3,4}) == float2{3,2}));
|
|
REPORTER_ASSERT(r, all(skvx::shuffle<3,3,3,3> (float4{1,2,3,4}) == float4{4,4,4,4}));
|
|
REPORTER_ASSERT(r, all(skvx::shuffle<2,1,2,1,2,1,2,1>(float4{1,2,3,4})
|
|
== float8{3,2,3,2,3,2,3,2}));
|
|
|
|
// Test that mixed types can be used where they make sense. Mostly about ergonomics.
|
|
REPORTER_ASSERT(r, all(float4{1,2,3,4} < 5));
|
|
REPORTER_ASSERT(r, all( byte4{1,2,3,4} < 5));
|
|
REPORTER_ASSERT(r, all( int4{1,2,3,4} < 5.0f));
|
|
float4 five = 5;
|
|
REPORTER_ASSERT(r, all(five == 5.0f));
|
|
REPORTER_ASSERT(r, all(five == 5));
|
|
|
|
REPORTER_ASSERT(r, all(max(2, min(float4{1,2,3,4}, 3)) == float4{2,2,3,3}));
|
|
|
|
for (int x = 0; x < 256; x++)
|
|
for (int y = 0; y < 256; y++) {
|
|
uint8_t want = (uint8_t)( 255*(x/255.0 * y/255.0) + 0.5 );
|
|
|
|
{
|
|
uint8_t got = skvx::div255(skvx::Vec<8, uint16_t>(x) *
|
|
skvx::Vec<8, uint16_t>(y) )[0];
|
|
REPORTER_ASSERT(r, got == want);
|
|
}
|
|
|
|
{
|
|
uint8_t got = skvx::approx_scale(skvx::Vec<8,uint8_t>(x),
|
|
skvx::Vec<8,uint8_t>(y))[0];
|
|
|
|
REPORTER_ASSERT(r, got == want-1 ||
|
|
got == want ||
|
|
got == want+1);
|
|
if (x == 0 || y == 0 || x == 255 || y == 255) {
|
|
REPORTER_ASSERT(r, got == want);
|
|
}
|
|
}
|
|
}
|
|
|
|
for (int x = 0; x < 256; x++)
|
|
for (int y = 0; y < 256; y++) {
|
|
uint16_t xy = x*y;
|
|
|
|
// Make sure to cover implementation cases N=8, N<8, and N>8.
|
|
REPORTER_ASSERT(r, all(mull(byte2 (x), byte2 (y)) == xy));
|
|
REPORTER_ASSERT(r, all(mull(byte4 (x), byte4 (y)) == xy));
|
|
REPORTER_ASSERT(r, all(mull(byte8 (x), byte8 (y)) == xy));
|
|
REPORTER_ASSERT(r, all(mull(byte16(x), byte16(y)) == xy));
|
|
}
|
|
}
|