d3dx9: Implement D3DXFloat16to32Array and D3DXFloat32to16Array.
Misha Koshelev
misha680 at gmail.com
Mon Feb 7 19:32:54 CST 2011
On Mon, Feb 7, 2011 at 6:20 PM, Stefan Dösinger <stefandoesinger at gmx.at> wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
>
> Am 07.02.2011 um 23:49 schrieb Misha Koshelev:
> I have no full explanation yet, just bits and pieces. By the way, if you haven't done so already I recommend a thorough study of http://en.wikipedia.org/wiki/Floating_point for background knowledge. http://en.wikipedia.org/wiki/IEEE_754-2008 may also be interesting, but I haven't read that wikipedia page so I don't really know what is in it.
>
>> I believe I've found that the threshold value is actually 65520.0 for
>> conversions from single to half precision
> That makes sense, it is halfway between 65536 and 65604. At this threshold values are rounded up or down
>
>> In any case, I have gotten the half to single precision conversion to
>> work in Wine by extending to 16 bits for the exponent (changing < 31
>> to < 32).
> You may as well replace it with TRUE because the exponent won't be >= 32. Too few bits to express anything bigger than 31.
>
>> However, I am still having trouble with the single to half precision
>> and believe I am spinning my wheels a bit at this point.
> I'm still trying to make sense of the Windows values :-/
>
>> math.c:2249: Test failed: Got 7800, expected 7bff or 7bff for index 3.
> Looks like the exponent is 1 too high and the mantissa zero instead of 1023. Probably a rounding / off by one bug in the code from wined3d. Different behavior wrt what happens when you exactly hit the rounding threshold. (Ie, do you round an exact 5 up or down(to 0 or 10)?)
>
>> math.c:2249: Test failed: Got ffff, expected fc00 or fce2 for index 8.
>> math.c:2249: Test failed: Got ffff, expected fc00 or fc00 for index 12.
>> math.c:2249: Test failed: Got ffff, expected fc00 or fc00 for index 13.
>> math.c:2249: Test failed: Got ffff, expected fc00 or fc00 for index 14.
>> math.c:2249: Test failed: Got ffff, expected fc00 or fc00 for index 15.
> This is this statement:
> + if (*in < -65520) return 0xFFFF;
> You probably wanted to return fc00 here, if you are trying to emulate case 1. Otherwise remove both CONFUSION HERE statements, change the if(exp > 30) case to exp > 31 and set ret = 0x7fff there. This will give you case 2.
>
>> math.c:2249: Test failed: Got 6400, expected 7fff or 7fff for index 18.
>> math.c:2249: Test failed: Got 6400, expected ffff or ffff for index 19.
> Hmm, I am afraid you can't remove the check for NaN entirely. If you try to any float operation on NaN undefined things are going to happen(Well, you get NaN, and later try to pick its bits apart)
>
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG/MacGPG2 v2.0.17 (Darwin)
>
> iQIcBAEBAgAGBQJNUH4mAAoJEN0/YqbEcdMwiNcP/R4sa+Z6L39FtN8qE1nX+Pwj
> KER3a4MWQ8DWDQgjXR09l3kYua9tM8kH5XmZNNpthH5Uck/CCdFOfCx+j3+CLuJq
> nbFkT5kwVA4qzv67o9qJxJhCaL2+9rXtB6ADOSU3QfLYn7L/UXzxwENzxgQlvFG5
> +0GzkdlimMGB8+Rj8mYte/ORFEQQnLoCFe3d1UT9RQQhOCqQvlDuaptBWQCiFiAc
> GgSQuBzC7YbZxLz8i0EDjfn50X5lW0iVo0p682RJzmod+oCeFztjY/G33wGyA1MA
> 45Ui6onTYpN1OES1esw/WtePkM8dKrSDlw+gu5rmU9c8JwhPFnNvs/1FUvHD1bGG
> WjHkv3hLCf3XVjMk1aT70ItFQ8ph2b6IlU3GOSvQvrYbDB8jfb7lbJnuqzd4hrd4
> ycTzAf5YULwuQ5m7Ie5QtWO+hG0iB/I0aqGIAnNj6tSO3b/bfmBlQKAauVnoUnDq
> sYgIEmW0GO15Ndsd/ipfbh414qQsNPyGQNdnhivkJK2PrVaiyXb/ztJf/vLDKjEw
> cRPoaAfQyIsfPD+cqKq4pnw2rwvr2tkOby11UQ8jfDeaTldp3redf0py0o37CGq1
> 520QC0CqC/A4MHawwbgcbFTi6XX/DCkHW4IOEirJvnweV8mw8V9IP4XyY/xmYepV
> ukB4hwY8XvIe4/Zxj8IP
> =bs4K
> -----END PGP SIGNATURE-----
>
Wow, thanks so much. That really helped.
The attached patch now passes all Wine tests and Windows tests as well:
https://testbot.winehq.org/JobDetails.pl?Key=9020
I had to use the signbit function as the > 0.0f check does not work for NaNs.
Any other suggestions for the patch before re-submission to
wine-patches? And sorry if I missed is Henri on vacation?
Thanks
Misha
-------------- next part --------------
From 06d992ea5e55a36222b2e426fe66be55fe2aaa96 Mon Sep 17 00:00:00 2001
From: Misha Koshelev <misha680 at gmail.com>
Date: Mon, 7 Feb 2011 20:28:01 -0500
Subject: d3dx9: Implement D3DXFloat32To16Array and D3DXFloat16To32Array.
To: wine-patches <wine-patches at winehq.org>
Reply-To: wine-devel <wine-devel at winehq.org>
---
dlls/d3dx9_36/d3dx9_36.spec | 4 +-
dlls/d3dx9_36/math.c | 87 +++++++++++++++++++++++++++++++++++++++++++
dlls/d3dx9_36/tests/math.c | 48 +++++++++++++++++++++++
include/d3dx9math.h | 18 +++++++++
include/d3dx9math.inl | 31 +++++++++++++++
5 files changed, 186 insertions(+), 2 deletions(-)
diff --git a/dlls/d3dx9_36/d3dx9_36.spec b/dlls/d3dx9_36/d3dx9_36.spec
index cbb6d20..55fe0cb 100644
--- a/dlls/d3dx9_36/d3dx9_36.spec
+++ b/dlls/d3dx9_36/d3dx9_36.spec
@@ -130,8 +130,8 @@
@ stub D3DXFillVolumeTextureTX
@ stdcall D3DXFilterTexture(ptr ptr long long)
@ stdcall D3DXFindShaderComment(ptr long ptr ptr)
-@ stub D3DXFloat16To32Array
-@ stub D3DXFloat32To16Array
+@ stdcall D3DXFloat16To32Array(ptr ptr long)
+@ stdcall D3DXFloat32To16Array(ptr ptr long)
@ stub D3DXFrameAppendChild
@ stub D3DXFrameCalculateBoundingSphere
@ stub D3DXFrameDestroy
diff --git a/dlls/d3dx9_36/math.c b/dlls/d3dx9_36/math.c
index fdb5f92..d2617ac 100644
--- a/dlls/d3dx9_36/math.c
+++ b/dlls/d3dx9_36/math.c
@@ -1769,3 +1769,90 @@ D3DXVECTOR4* WINAPI D3DXVec4TransformArray(D3DXVECTOR4* out, UINT outstride, CON
}
return out;
}
+
+static inline unsigned short float_32_to_16(const float *in)
+{
+ int exp = 0;
+ float tmp = fabs(*in);
+ unsigned int mantissa;
+ unsigned short ret;
+
+ /* Deal with special numbers */
+ if (*((unsigned int *) in) == 0x00000000) return 0x0000;
+ if (*((unsigned int *) in) == 0x80000000) return 0x8000;
+ if (isinf(*in)) return (signbit(*in) ? 0xffff : 0x7c00);
+ if (isnan(*in)) return (signbit(*in) ? 0xffff : 0x7fff);
+ if (fabs(*in) > 65520) return (signbit(*in) ? 0xfc00 : 0x7c00);
+
+ if (tmp < powf(2, 10)) {
+ do
+ {
+ tmp = tmp * 2.0f;
+ exp--;
+ } while (tmp < powf(2, 10));
+ } else if (tmp >= powf(2, 11)) {
+ do
+ {
+ tmp /= 2.0f;
+ exp++;
+ } while (tmp >= powf(2, 11));
+ }
+
+ mantissa = (unsigned int) tmp;
+ if (tmp - mantissa > 0.5f) mantissa++; /* round to nearest, away from zero */
+
+ exp += 10; /* Normalize the mantissa */
+ exp += 15; /* Exponent is encoded with excess 15 */
+
+ if (exp > 31) { /* too big */
+ ret = 0x7fff; /* INF */
+ } else if (exp <= 0) {
+ /* exp == 0: Non-normalized mantissa. Returns 0x0000 (=0.0) for too small numbers */
+ while (exp <= 0) {
+ mantissa = mantissa >> 1;
+ exp++;
+ }
+ ret = mantissa & 0x3ff;
+ } else {
+ ret = (exp << 10) | (mantissa & 0x3ff);
+ }
+
+ ret |= ((*in < 0.0f ? 1 : 0) << 15); /* Add the sign */
+ return ret;
+}
+
+D3DXFLOAT16 *WINAPI D3DXFloat32To16Array(D3DXFLOAT16 *pout, CONST FLOAT *pin, UINT n)
+{
+ unsigned int i;
+
+ for (i = 0; i < n; ++i) {
+ pout[i].value = float_32_to_16(&pin[i]);
+ }
+
+ return pout;
+}
+
+static inline float float_16_to_32(const unsigned short *in) {
+ const unsigned short s = ((*in) & 0x8000);
+ const unsigned short e = ((*in) & 0x7C00) >> 10;
+ const unsigned short m = (*in) & 0x3FF;
+ const float sgn = (s ? -1.0f : 1.0f);
+
+ if (e == 0) {
+ if (m == 0) return sgn * 0.0f; /* +0.0 or -0.0 */
+ else return sgn * powf(2, -14.0f) * ((float)m / 1024.0f);
+ } else {
+ return sgn * powf(2, (float)e - 15.0f) * (1.0f + ((float)m / 1024.0f));
+ }
+}
+
+FLOAT *WINAPI D3DXFloat16To32Array(FLOAT *pout, CONST D3DXFLOAT16 *pin, UINT n)
+{
+ unsigned int i;
+
+ for (i = 0; i < n; ++i) {
+ pout[i] = float_16_to_32(&pin[i].value);
+ }
+
+ return pout;
+}
diff --git a/dlls/d3dx9_36/tests/math.c b/dlls/d3dx9_36/tests/math.c
index e455a96..0d8c5eb 100644
--- a/dlls/d3dx9_36/tests/math.c
+++ b/dlls/d3dx9_36/tests/math.c
@@ -21,6 +21,7 @@
#include "wine/test.h"
#include "d3dx9.h"
+#include <math.h>
#define ARRAY_SIZE 5
@@ -2216,6 +2217,52 @@ static void test_D3DXVec_Array(void)
compare_planes(exp_plane, out_plane);
}
+static void test_D3DXFloat_Array(void)
+{
+ unsigned int i;
+ void *out = NULL;
+ D3DXFLOAT16 half_res[22];
+ FLOAT single[22] = { 80000.0f, 65503.0f, 65504.0f, 65520.0f, 65521.0f, 65534.0f, 65535.0f, 65536.0f,
+ -80000.0f, -65503.0f, -65504.0f, -65520.0f, -65521.0f, -65534.0f, -65535.0f, -65536.0f,
+ INFINITY, -INFINITY, NAN, -NAN, 0.0f, -0.0f },
+ single_exp[22] = { 65536.0f, 65504.0f, 65504.0f, 65504.0f, 65536.0f, 65536.0f, 65535.0f, 65536.0f,
+ -65536.0f, -65504.0f, -65504.0f, -65504.0f, -65536.0f, -65536.0f, -65535.0f, -65536.0f,
+ 65536.0f, -131008.0f, 131008.0f, -131008.0f, 0.0f, -0.0f },
+ single_res[22];
+ /* half_ver2 occurs on WXPPROSP3 (32 bit math), WVISTAADM (32 bit math), W7PRO (32 bit math) */
+ WORD half_ver1[22] = { 0x7c00, 0x7bff, 0x7bff, 0x7bff, 0x7c00, 0x7c00, 0x7c00, 0x7c00,
+ 0xfc00, 0xfbff, 0xfbff, 0xfbff, 0xfc00, 0xfc00, 0xfc00, 0xfc00,
+ 0x7c00, 0xffff, 0x7fff, 0xffff, 0x0, 0x8000 },
+ half_ver2[22] = { 0x7ce2, 0x7bff, 0x7bff, 0x7c00, 0x7c00, 0x7c00, 0x7c00, 0x7c00,
+ 0xfce2, 0xfbff, 0xfbff, 0xfc00, 0xfc00, 0xfc00, 0xfc00, 0xfc00,
+ 0x7fff, 0xffff, 0x7fff, 0xffff, 0x0, 0x8000 };
+
+ /* exception on NULL out or in parameter */
+ out = D3DXFloat32To16Array(half_res, single, 0);
+ ok(out == half_res, "Got %p, expected %p.\n", out, half_res);
+
+ out = D3DXFloat32To16Array(half_res, single, 22);
+ ok(out == half_res, "Got %p, expected %p.\n", out, half_res);
+
+ for (i = 0; i < 22; i++) {
+ ok(half_res[i].value == half_ver1[i] ||
+ half_res[i].value == half_ver2[i], "Got %x, expected %x or %x for index %d.\n",
+ half_res[i].value, half_ver1[i], half_ver2[i], i);
+ }
+
+ /* exception on NULL out or in parameter */
+ out = D3DXFloat16To32Array(single_res, (D3DXFLOAT16 *)half_ver1, 0);
+ ok(out == single_res, "Got %p, expected %p.\n", out, single_res);
+
+ out = D3DXFloat16To32Array(single_res, (D3DXFLOAT16 *)half_ver1, 22);
+ ok(out == single_res, "Got %p, expected %p.\n", out, single_res);
+
+ for (i = 0; i < 22; i++) {
+ ok(relative_error(single_res[i], single_exp[i]) < admitted_error, "Got %f, expected %f for index %d.\n",
+ single_res[i], single_exp[i], i);
+ }
+}
+
START_TEST(math)
{
D3DXColorTest();
@@ -2231,4 +2278,5 @@ START_TEST(math)
test_Matrix_Decompose();
test_Matrix_Transformation2D();
test_D3DXVec_Array();
+ test_D3DXFloat_Array();
}
diff --git a/include/d3dx9math.h b/include/d3dx9math.h
index f842e3e..cdb1deb 100644
--- a/include/d3dx9math.h
+++ b/include/d3dx9math.h
@@ -261,6 +261,21 @@ typedef struct D3DXCOLOR
FLOAT r, g, b, a;
} D3DXCOLOR, *LPD3DXCOLOR;
+typedef struct D3DXFLOAT16
+{
+#ifdef __cplusplus
+ D3DXFLOAT16();
+ D3DXFLOAT16(FLOAT f);
+ D3DXFLOAT16(CONST D3DXFLOAT16 &f);
+
+ operator FLOAT ();
+
+ BOOL operator == (CONST D3DXFLOAT16 &) const;
+ BOOL operator != (CONST D3DXFLOAT16 &) const;
+#endif /* __cplusplus */
+ WORD value;
+} D3DXFLOAT16, *LPD3DXFLOAT16;
+
#ifdef __cplusplus
extern "C" {
#endif
@@ -358,6 +373,9 @@ D3DXVECTOR4* WINAPI D3DXVec4Normalize(D3DXVECTOR4 *pout, CONST D3DXVECTOR4 *pv);
D3DXVECTOR4* WINAPI D3DXVec4Transform(D3DXVECTOR4 *pout, CONST D3DXVECTOR4 *pv, CONST D3DXMATRIX *pm);
D3DXVECTOR4* WINAPI D3DXVec4TransformArray(D3DXVECTOR4 *pout, UINT outstride, CONST D3DXVECTOR4 *pv, UINT vstride, CONST D3DXMATRIX *pm, UINT n);
+D3DXFLOAT16 *WINAPI D3DXFloat32To16Array(D3DXFLOAT16 *pout, CONST FLOAT *pin, UINT n);
+FLOAT *WINAPI D3DXFloat16To32Array(FLOAT *pout, CONST D3DXFLOAT16 *pin, UINT n);
+
#ifdef __cplusplus
}
#endif
diff --git a/include/d3dx9math.inl b/include/d3dx9math.inl
index 3cd078a..3f55aef 100644
--- a/include/d3dx9math.inl
+++ b/include/d3dx9math.inl
@@ -851,6 +851,37 @@ inline BOOL D3DXCOLOR::operator != (CONST D3DXCOLOR& col) const
return r != col.r || g != col.g || b != col.b || a != col.a;
}
+inline D3DXFLOAT16::D3DXFLOAT16()
+{
+}
+
+inline D3DXFLOAT16::D3DXFLOAT16(FLOAT f)
+{
+ D3DXFloat32To16Array(this, &f, 1);
+}
+
+inline D3DXFLOAT16::D3DXFLOAT16(CONST D3DXFLOAT16 &f)
+{
+ value = f.value;
+}
+
+inline D3DXFLOAT16::operator FLOAT ()
+{
+ FLOAT f;
+ D3DXFloat16To32Array(&f, this, 1);
+ return f;
+}
+
+inline BOOL D3DXFLOAT16::operator == (CONST D3DXFLOAT16 &f) const
+{
+ return value == f.value;
+}
+
+inline BOOL D3DXFLOAT16::operator != (CONST D3DXFLOAT16 &f) const
+{
+ return value != f.value;
+}
+
#endif /* __cplusplus */
/*_______________D3DXCOLOR_____________________*/
--
1.7.4
More information about the wine-devel
mailing list