Petr Gajdos
eda5aea351
- Add patch to fix regression on non-SSE architectures https://github.com/AcademySoftwareFoundation/openexr/issues/1460: * 1488.patch - Also disable 'DWA[AB]Compression' test on aarch64 as the patch above fixes the issue for all targets, except aarch64 OBS-URL: https://build.opensuse.org/request/show/1099245 OBS-URL: https://build.opensuse.org/package/show/graphics/openexr?expand=0&rev=101
83 lines
3.6 KiB
Diff
83 lines
3.6 KiB
Diff
From 1461a92d198a8d7586bbb7a5f1c99311a175a223 Mon Sep 17 00:00:00 2001
|
|
From: Kimball Thurston <kdt3rd@gmail.com>
|
|
Date: Sat, 15 Jul 2023 10:38:22 +1200
|
|
Subject: [PATCH] Revert pre-computed values
|
|
|
|
It would appear the precomputed values cause differences from the
|
|
specific computed values on other architectures.
|
|
|
|
Signed-off-by: Kimball Thurston <kdt3rd@gmail.com>
|
|
---
|
|
src/lib/OpenEXRCore/internal_dwa_simd.h | 46 ++++++++++++-------------
|
|
1 file changed, 23 insertions(+), 23 deletions(-)
|
|
|
|
diff --git a/src/lib/OpenEXRCore/internal_dwa_simd.h b/src/lib/OpenEXRCore/internal_dwa_simd.h
|
|
index 6432c5fb3..64399ebf8 100644
|
|
--- a/src/lib/OpenEXRCore/internal_dwa_simd.h
|
|
+++ b/src/lib/OpenEXRCore/internal_dwa_simd.h
|
|
@@ -871,13 +871,13 @@ dctInverse8x8DcOnly (float* data)
|
|
static inline void
|
|
dctInverse8x8_scalar (float* data, int zeroedRows)
|
|
{
|
|
- const float a = 0.35355362513961314f;//.5f * cosf (3.14159f / 4.0f);
|
|
- const float b = 0.4903926563794112f;//.5f * cosf (3.14159f / 16.0f);
|
|
- const float c = 0.4619398297234211f;//.5f * cosf (3.14159f / 8.0f);
|
|
- const float d = 0.4157349443626743f;//.5f * cosf (3.f * 3.14159f / 16.0f);
|
|
- const float e = 0.2777854612564676f;//.5f * cosf (5.f * 3.14159f / 16.0f);
|
|
- const float f = 0.19134217585694352f;//.5f * cosf (3.f * 3.14159f / 8.0f);
|
|
- const float g = 0.09754573032714427f;//.5f * cosf (7.f * 3.14159f / 16.0f);
|
|
+ const float a = .5f * cosf (3.14159f / 4.0f);
|
|
+ const float b = .5f * cosf (3.14159f / 16.0f);
|
|
+ const float c = .5f * cosf (3.14159f / 8.0f);
|
|
+ const float d = .5f * cosf (3.f * 3.14159f / 16.0f);
|
|
+ const float e = .5f * cosf (5.f * 3.14159f / 16.0f);
|
|
+ const float f = .5f * cosf (3.f * 3.14159f / 8.0f);
|
|
+ const float g = .5f * cosf (7.f * 3.14159f / 16.0f);
|
|
|
|
float alpha[4], beta[4], theta[4], gamma[4];
|
|
|
|
@@ -1868,7 +1868,7 @@ dctInverse8x8_avx_7 (float* data)
|
|
// Default implementation
|
|
//
|
|
|
|
-void
|
|
+static void
|
|
dctForward8x8 (float* data)
|
|
{
|
|
float A0, A1, A2, A3, A4, A5, A6, A7;
|
|
@@ -1877,20 +1877,20 @@ dctForward8x8 (float* data)
|
|
float* srcPtr = data;
|
|
float* dstPtr = data;
|
|
|
|
- const float c1 = 0.9807853127588224f;//cosf (3.14159f * 1.0f / 16.0f);
|
|
- const float c2 = 0.9238796594468422f;//cosf (3.14159f * 2.0f / 16.0f);
|
|
- const float c3 = 0.8314698887253485f;//cosf (3.14159f * 3.0f / 16.0f);
|
|
- const float c4 = 0.7071072502792263f;//cosf (3.14159f * 4.0f / 16.0f);
|
|
- const float c5 = 0.5555709225129352f;//cosf (3.14159f * 5.0f / 16.0f);
|
|
- const float c6 = 0.38268435171388704f;//cosf (3.14159f * 6.0f / 16.0f);
|
|
- const float c7 = 0.19509146065428853f;//cosf (3.14159f * 7.0f / 16.0f);
|
|
-
|
|
- const float c1Half = 0.4903926563794112f;//.5f * c1;
|
|
- const float c2Half = 0.4619398297234211f;//.5f * c2;
|
|
- const float c3Half = 0.4157349443626743f;//.5f * c3;
|
|
- const float c5Half = 0.2777854612564676f;//.5f * c5;
|
|
- const float c6Half = 0.19134217585694352f;//.5f * c6;
|
|
- const float c7Half = 0.09754573032714427f;//.5f * c7;
|
|
+ const float c1 = cosf (3.14159f * 1.0f / 16.0f);
|
|
+ const float c2 = cosf (3.14159f * 2.0f / 16.0f);
|
|
+ const float c3 = cosf (3.14159f * 3.0f / 16.0f);
|
|
+ const float c4 = cosf (3.14159f * 4.0f / 16.0f);
|
|
+ const float c5 = cosf (3.14159f * 5.0f / 16.0f);
|
|
+ const float c6 = cosf (3.14159f * 6.0f / 16.0f);
|
|
+ const float c7 = cosf (3.14159f * 7.0f / 16.0f);
|
|
+
|
|
+ const float c1Half = .5f * c1;
|
|
+ const float c2Half = .5f * c2;
|
|
+ const float c3Half = .5f * c3;
|
|
+ const float c5Half = .5f * c5;
|
|
+ const float c6Half = .5f * c6;
|
|
+ const float c7Half = .5f * c7;
|
|
|
|
//
|
|
// First pass - do a 1D DCT over the rows and write the
|