Evas: Neon: Explicitly set the .fpu flag for neon.

This means you should not need to set any special compiler flags; which should
mean gcc will generate non-neon specific asm (unless you tell it to).  This
means it is possible to build an armv6 binary with neon suppor (as we always
meant to to do).

git-svn-id: svn+ssh://svn.enlightenment.org/var/svn/e/trunk/evas@55307 7cbeb6ba-43b4-40fd-8cce-4c39aea84d33

diff --git a/src/lib/engines/common/evas_cpu.c b/src/lib/engines/common/evas_cpu.c
index 9ccd40d..ee92bb8 100644 (file)
--- a/src/lib/engines/common/evas_cpu.c
+++ b/src/lib/engines/common/evas_cpu.c
@@ -78,6 +78,7 @@ evas_common_cpu_neon_test(void)
 //#if defined(__ARM_ARCH__) && (__ARM_ARCH__ >= 70)
 #ifdef BUILD_NEON
    asm volatile (
+               ".fpu neon           \n\t"
                  "vqadd.u8 d0, d1, d0\n"
                  );
 #endif

diff --git a/src/lib/engines/common/evas_op_blend/op_blend_color_neon.c b/src/lib/engines/common/evas_op_blend/op_blend_color_neon.c
index fd0192a..53b9991 100644 (file)
--- a/src/lib/engines/common/evas_op_blend/op_blend_color_neon.c
+++ b/src/lib/engines/common/evas_op_blend/op_blend_color_neon.c
@@ -6,6 +6,7 @@ _op_blend_c_dp_neon(DATA32 *s __UNUSED__, DATA8 *m __UNUSED__, DATA32 c, DATA32
        DATA32 *e, *tmp = 0;
 #define AP     "B_C_DP"
    asm volatile (
+       ".fpu neon                                      \n\t"
        "vdup.u32       q6, %[c]                        \n\t"
        "vmov.i8        q5, #1                          \n\t"
        "vmvn.u8        q7,q6                           \n\t"

diff --git a/src/lib/engines/common/evas_op_blend/op_blend_mask_color_neon.c b/src/lib/engines/common/evas_op_blend/op_blend_mask_color_neon.c
index a97a20a..46929f3 100644 (file)
--- a/src/lib/engines/common/evas_op_blend/op_blend_mask_color_neon.c
+++ b/src/lib/engines/common/evas_op_blend/op_blend_mask_color_neon.c
@@ -25,6 +25,7 @@ _op_blend_mas_c_dp_neon(DATA32 *s __UNUSED__, DATA8 *m, DATA32 c, DATA32 *d, int
 
 #define AP "blend_mas_c_dp_"
      asm volatile (
+       ".fpu neon                                      \n\t"
        "       vdup.i32        q15, %[c]                       \n\t"
        "       vmov.i8         q14,    #1                      \n\t"
 
@@ -205,6 +206,7 @@ _op_blend_mas_can_dp_neon(DATA32 *s __UNUSED__, DATA8 *m, DATA32 c, DATA32 *d, i
 
 #define AP     "_blend_mas_can_dp_neon_"
      asm volatile (
+       ".fpu neon                                      \n\t"
                "vdup.u32       q9,     %[c]            \n\t"
                "vmov.i8        q15,    #1              \n\t"
                "vmov.i8        q14,    #0              \n\t"

diff --git a/src/lib/engines/common/evas_op_blend/op_blend_pixel_color_neon.c b/src/lib/engines/common/evas_op_blend/op_blend_pixel_color_neon.c
index a25fcd0..51925c3 100644 (file)
--- a/src/lib/engines/common/evas_op_blend/op_blend_pixel_color_neon.c
+++ b/src/lib/engines/common/evas_op_blend/op_blend_pixel_color_neon.c
@@ -6,6 +6,7 @@ static void
 _op_blend_p_c_dp_neon(DATA32 *s, DATA8 *m __UNUSED__, DATA32 c, DATA32 *d, int l) {
 #define AP     "blend_p_c_dp_"
    asm volatile (
+       ".fpu neon                                      \n\t"
                // Load 'c'
                "vdup.u32       q7, %[c]                        \n\t"
                "vmov.i8        q6, #1                          \n\t"
@@ -202,6 +203,7 @@ _op_blend_pan_caa_dp_neon(DATA32 *s, DATA8 *m __UNUSED__, DATA32 c, DATA32 *d, i
 #define AP     "_op_blend_pan_caa_dp_"
    DATA32 *e = d + l, *tmp = (void*)73;
       asm volatile (
+       ".fpu neon                                      \n\t"
                /* Set up 'c' */
                "vdup.u8     d14, %[c]          \n\t"
                "vmov.i8     d15, #1            \n\t"

diff --git a/src/lib/engines/common/evas_op_blend/op_blend_pixel_neon.c b/src/lib/engines/common/evas_op_blend/op_blend_pixel_neon.c
index 19fe98b..5fcae8d 100644 (file)
--- a/src/lib/engines/common/evas_op_blend/op_blend_pixel_neon.c
+++ b/src/lib/engines/common/evas_op_blend/op_blend_pixel_neon.c
@@ -5,6 +5,7 @@ static void
 _op_blend_p_dp_neon(DATA32 *s, DATA8 *m, DATA32 c, DATA32 *d, int l) {
 #define AP "blend_p_dp_"
   asm volatile (
+       ".fpu neon                                      \n\t"
        //** init
        "vmov.i8        q8,     $0x1            \n\t"
 
@@ -226,6 +227,7 @@ _op_blend_pas_dp_neon(DATA32 *s, DATA8 *m, DATA32 c, DATA32 *d, int l) {
 #define AP "blend_pas_dp_"
    DATA32 *e = d + l,*tmp  = e + 32,*pl=(void*)912;
       asm volatile (
+        ".fpu neon                                     \n\t"
                "vmov.i8        q8,     #1                      \n\t"
        AP"loopchoose:                                          \n\t"
                // If aliged - go as fast we can

diff --git a/src/lib/engines/common/evas_op_copy/op_copy_color_neon.c b/src/lib/engines/common/evas_op_copy/op_copy_color_neon.c
index 2891c06..96310cd 100644 (file)
--- a/src/lib/engines/common/evas_op_copy/op_copy_color_neon.c
+++ b/src/lib/engines/common/evas_op_copy/op_copy_color_neon.c
@@ -6,6 +6,7 @@ _op_copy_c_dp_neon(DATA32 *s, DATA8 *m, DATA32 c, DATA32 *d, int l) {
 #define AP "COPY_C_DP_"
    uint32_t *e = d + l,*tmp;
    asm volatile (
+      ".fpu neon                                       \n\t"
 
                "vdup.i32       q0,     %[c]            \n\t"
 

diff --git a/src/lib/engines/common/evas_op_copy/op_copy_pixel_neon.c b/src/lib/engines/common/evas_op_copy/op_copy_pixel_neon.c
index 5850611..bc3748d 100644 (file)
--- a/src/lib/engines/common/evas_op_copy/op_copy_pixel_neon.c
+++ b/src/lib/engines/common/evas_op_copy/op_copy_pixel_neon.c
@@ -35,6 +35,7 @@ _op_copy_p_dp_neon(DATA32 *s, DATA8 *m __UNUSED__, DATA32 c __UNUSED__, DATA32 *
         s3 = s + 8;
         s4 = s + 12;
         asm volatile (
+       ".fpu neon                                      \n\t"
                       "asmloop2:\n\t"
                       "cmp %[e], %[d]\n\t"
                       "vld1.32 {d16-d17}, [%[s]]!\n\t"