Revert F16C series (MR 6774)

This reverts commit 4fb2eddfdf. This reverts commit 7a1deb16f8. This reverts commit 2b6a172343. This reverts commit 5af81393e4. This reverts commit 87900afe5b. A couple of problems were discovered after this series was merged that cause breakage in different configurations: (1) It seems that using -mf16c also enables AVX, leading to SIGILL on platforms that do not support AVX. (2) Since clang only warns about unknown flags, and as I understand it Meson's handling in cc.has_argument() is broken, the F16C code is wrongly enabled when clang is used, even for example on ARM, leading to a compilation error. Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3583 Reviewed-by: Eric Anholt <eric@anholt.net> Reviewed-by: Rob Clark <robdclark@chromium.org> Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6969>
2020-10-01 13:48:16 -07:00
parent 4a0164ed85
commit 1aac47db69
45 changed files with 266 additions and 216 deletions
@@ -360,7 +360,7 @@ def conversion_expr(src_channel,

    # Promote half to float
    if src_type == FLOAT and src_size == 16:
-        value = '_mesa_half_to_float(%s)' % value
+        value = 'util_half_to_float(%s)' % value
        src_size = 32

    # Special case for float <-> ubytes for more accurate results
@@ -436,7 +436,7 @@ def conversion_expr(src_channel,
            src_size = 32

        if dst_channel.size == 16:
-            value = '_mesa_float_to_float16_rtz(%s)' % value
+            value = 'util_float_to_half_rtz(%s)' % value
        elif dst_channel.size == 64 and src_size < 64:
            value = '(double)%s' % value

@@ -715,7 +715,7 @@ def generate(formats):
    print()
    print('#include "pipe/p_compiler.h"')
    print('#include "util/u_math.h"')
-    print('#include "util/half_float.h"')
+    print('#include "util/u_half.h"')
    print('#include "u_format.h"')
    print('#include "u_format_other.h"')
    print('#include "util/format_srgb.h"')
@@ -914,11 +914,8 @@ util_format_test_cases[] =
   {PIPE_FORMAT_R16_FLOAT, PACKED_1x16(0xffff), PACKED_1x16(0x03FF), UNPACKED_1x1( 6.09756E-5, 0.0, 0.0, 1.0)},
 #endif

-   /* This fails with _mesa_float_to_float16_rtz, but passes with _mesa_float_to_float16_rtne. */
-#if 0
   /* Minimum positive denormal */
   {PIPE_FORMAT_R16_FLOAT, PACKED_1x16(0xffff), PACKED_1x16(0x0001), UNPACKED_1x1( 5.96046E-8, 0.0, 0.0, 1.0)},
-#endif

   /* Min representable value */
   {PIPE_FORMAT_R16_FLOAT, PACKED_1x16(0xffff), PACKED_1x16(0xfbff), UNPACKED_1x1(   -65504.0, 0.0, 0.0, 1.0)},
@@ -28,10 +28,10 @@
 #include <math.h>
 #include <assert.h>
 #include "half_float.h"
+#include "util/u_half.h"
 #include "rounding.h"
 #include "softfloat.h"
 #include "macros.h"
-#include "u_math.h"

 typedef union { float f; int32_t i; uint32_t u; } fi_type;

@@ -54,7 +54,7 @@ typedef union { float f; int32_t i; uint32_t u; } fi_type;
 *     result in the same value as if the expression were executed on the GPU.
 */
 uint16_t
-_mesa_float_to_half_slow(float val)
+_mesa_float_to_half(float val)
 {
   const fi_type fi = {val};
   const int flt_m = fi.i & 0x7fffff;
@@ -129,9 +129,9 @@ _mesa_float_to_half_slow(float val)
 }

 uint16_t
-_mesa_float_to_float16_rtz_slow(float val)
+_mesa_float_to_float16_rtz(float val)
 {
-    return _mesa_float_to_half_rtz_slow(val);
+    return _mesa_float_to_half_rtz(val);
 }

 /**
@@ -140,31 +140,9 @@ _mesa_float_to_float16_rtz_slow(float val)
 * http://www.opengl.org/discussion_boards/ubb/Forum3/HTML/008786.html
 */
 float
-_mesa_half_to_float_slow(uint16_t val)
+_mesa_half_to_float(uint16_t val)
 {
-   union fi infnan;
-   union fi magic;
-   union fi f32;
-
-   infnan.ui = 0x8f << 23;
-   infnan.f = 65536.0f;
-   magic.ui  = 0xef << 23;
-
-   /* Exponent / Mantissa */
-   f32.ui = (val & 0x7fff) << 13;
-
-   /* Adjust */
-   f32.f *= magic.f;
-   /* XXX: The magic mul relies on denorms being available */
-
-   /* Inf / NaN */
-   if (f32.f >= infnan.f)
-      f32.ui |= 0xff << 23;
-
-   /* Sign */
-   f32.ui |= (uint32_t)(val & 0x8000) << 16;
-
-   return f32.f;
+   return util_half_to_float(val);
 }

 /**
@@ -28,14 +28,6 @@

 #include <stdbool.h>
 #include <stdint.h>
-#include <string.h>
-#include "util/u_cpu_detect.h"
-
-#ifdef USE_F16C
-#include <immintrin.h>
-#define F16C_NEAREST 0
-#define F16C_TRUNCATE 3
-#endif

 #ifdef __cplusplus
 extern "C" {
@@ -44,48 +36,18 @@ extern "C" {
 #define FP16_ONE     ((uint16_t) 0x3c00)
 #define FP16_ZERO    ((uint16_t) 0)

-uint16_t _mesa_float_to_half_slow(float val);
-float _mesa_half_to_float_slow(uint16_t val);
+uint16_t _mesa_float_to_half(float val);
+float _mesa_half_to_float(uint16_t val);
 uint8_t _mesa_half_to_unorm8(uint16_t v);
 uint16_t _mesa_uint16_div_64k_to_half(uint16_t v);

 /*
- * _mesa_float_to_float16_rtz_slow is no more than a wrapper to the counterpart
+ * _mesa_float_to_float16_rtz is no more than a wrapper to the counterpart
 * softfloat.h call. Still, softfloat.h conversion API is meant to be kept
 * private. In other words, only use the API published here, instead of
 * calling directly the softfloat.h one.
 */
-uint16_t _mesa_float_to_float16_rtz_slow(float val);
-
-static inline uint16_t
-_mesa_float_to_half(float val)
-{
-#ifdef USE_F16C
-   if (util_cpu_caps.has_f16c)
-      return _cvtss_sh(val, F16C_NEAREST);
-#endif
-   return _mesa_float_to_half_slow(val);
-}
-
-static inline float
-_mesa_half_to_float(uint16_t val)
-{
-#ifdef USE_F16C
-   if (util_cpu_caps.has_f16c)
-      return _cvtsh_ss(val);
-#endif
-   return _mesa_half_to_float_slow(val);
-}
-
-static inline uint16_t
-_mesa_float_to_float16_rtz(float val)
-{
-#ifdef USE_F16C
-   if (util_cpu_caps.has_f16c)
-      return _cvtss_sh(val, F16C_TRUNCATE);
-#endif
-   return _mesa_float_to_float16_rtz_slow(val);
-}
+uint16_t _mesa_float_to_float16_rtz(float val);

 static inline uint16_t
 _mesa_float_to_float16_rtne(float val)
@@ -1435,7 +1435,7 @@ _mesa_double_to_f32(double val, bool rtz)
 * From f32_to_f16()
 */
 uint16_t
-_mesa_float_to_half_rtz_slow(float val)
+_mesa_float_to_half_rtz(float val)
 {
    const fi_type fi = {val};
    const uint32_t flt_m = fi.u & 0x7fffff;
@@ -56,7 +56,7 @@ double _mesa_double_mul_rtz(double a, double b);
 double _mesa_double_fma_rtz(double a, double b, double c);
 float _mesa_float_fma_rtz(float a, float b, float c);
 float _mesa_double_to_f32(double x, bool rtz);
-uint16_t _mesa_float_to_half_rtz_slow(float x);
+uint16_t _mesa_float_to_half_rtz(float x);

 #ifdef __cplusplus
 } /* extern C */
@@ -30,8 +30,7 @@
 #include <stdio.h>
 #include <float.h>

-#include "util/half_float.h"
-#include "util/u_math.h"
+#include "util/u_half.h"
 #include "util/format/u_format.h"
 #include "util/format/u_format_tests.h"
 #include "util/format/u_format_s3tc.h"