* liboil/colorspace/composite_i386.c: (composite_over_argb_sse2_2),

(composite_over_argb_sse2_3): hacking
* liboil/liboilcpu.c: (oil_cpu_i386_getflags_cpuid): Intel's SSE2
implies MMXEXT

3 files changed, 276 insertions, 0 deletions

diff --git a/ChangeLog b/ChangeLog
index 154a257..29d6b7e 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,10 @@
+2005-08-01  David Schleef  <ds@schleef.org>
+
+	* liboil/colorspace/composite_i386.c: (composite_over_argb_sse2_2),
+	(composite_over_argb_sse2_3): hacking
+	* liboil/liboilcpu.c: (oil_cpu_i386_getflags_cpuid): Intel's SSE2
+	implies MMXEXT
+
 2005-07-31  David Schleef  <ds@schleef.org>
 
 	* liboil/colorspace/composite_i386.c: (composite_over_argb_mmx_4),
diff --git a/liboil/colorspace/composite_i386.c b/liboil/colorspace/composite_i386.c
index 0e6cd3a..415ec65 100644
--- a/liboil/colorspace/composite_i386.c
+++ b/liboil/colorspace/composite_i386.c
@@ -295,6 +295,7 @@ composite_over_argb_mmx_3 (uint32_t *dest, uint32_t *src, int n)
 }
 OIL_DEFINE_IMPL_FULL (composite_over_argb_mmx_3, composite_over_argb, OIL_IMPL_FLAG_MMX);
 
+/* written for gromit */
 static void
 composite_over_argb_mmx_4 (uint32_t *dest, uint32_t *src, int n)
 {
@@ -440,6 +441,273 @@ composite_over_argb_sse2 (uint32_t *dest, uint32_t *src, int n)
 }
 OIL_DEFINE_IMPL_FULL (composite_over_argb_sse2, composite_over_argb, OIL_IMPL_FLAG_SSE2);
 
+/* written for shaun */
+static void
+composite_over_argb_sse2_2 (uint32_t *dest, uint32_t *src, int n)
+{
+  int end;
+
+  __asm__ __volatile__ ("  pxor %%xmm7, %%xmm7\n"   // mm7 = { 0, 0, 0, 0 }
+      "  movl $0x80808080, %%eax\n"
+      "  movd %%eax, %%xmm6\n"  // mm6 = { 128, 128, 128, 128 }
+      "  punpcklbw %%xmm7, %%xmm6\n"
+      "  punpcklwd %%xmm6, %%xmm6\n"
+      "  movl $0xffffffff, %%eax\n" // mm5 = { 255, 255, 255, 255 }
+      "  movd %%eax, %%xmm5\n"
+      "  punpcklbw %%xmm7, %%xmm5\n"
+      "  punpcklwd %%xmm5, %%xmm5\n"
+      "  movl $0x02020202, %%eax\n"
+      "  movd %%eax, %%xmm4\n"
+      "  punpcklbw %%xmm7, %%xmm4\n"
+      "  paddw %%xmm5, %%xmm4\n" // mm4 = { 257, 257, 257, 257 }
+      "  punpcklwd %%xmm4, %%xmm4\n"
+      :
+      :
+      :"eax");
+
+  if (n&1 && ((unsigned long)dest & 0xf)) {
+    __asm__ __volatile__ (
+      "  movl (%1), %%eax\n"
+      "  testl $0xff000000, %%eax\n"
+      "  jz 1f\n"
+
+      "  movd (%1), %%xmm1\n"
+      "  punpcklbw %%xmm7, %%xmm1\n"
+      "  pshuflw $0xff, %%xmm1, %%xmm0\n"
+      "  pxor %%xmm5, %%xmm0\n"
+
+      "  movd (%0), %%xmm3\n"
+      "  punpcklbw %%xmm7, %%xmm3\n"
+      "  pmullw %%xmm0, %%xmm3\n"
+      "  paddw %%xmm6, %%xmm3\n"
+      "  pmulhuw %%xmm4, %%xmm3\n"
+
+      "  paddw %%xmm1, %%xmm3\n"
+      "  packuswb %%xmm3, %%xmm3\n"
+      "  movd %%xmm3, (%0)\n"
+
+      "1:\n"
+      "  addl $4, %0\n"
+      "  addl $4, %1\n"
+      :"+r" (dest), "+r" (src), "+r" (n)
+      :
+      :"eax");
+  }
+  end = n&1;
+  n>>=1;
+
+  if (n>0){
+    __asm__ __volatile__ ("\n"
+      "3:\n"
+      "  movl (%1), %%eax\n"
+      "  orl 4(%1), %%eax\n"
+      "  testl $0xff000000, %%eax\n"
+      "  jz 4f\n"
+
+      "  movq (%1), %%xmm1\n"
+      "  punpcklbw %%xmm7, %%xmm1\n"
+      "  pshuflw $0xff, %%xmm1, %%xmm0\n"
+      "  pshufhw $0xff, %%xmm0, %%xmm0\n"
+      "  pxor %%xmm5, %%xmm0\n"
+
+      "  movq (%0), %%xmm3\n"
+      "  punpcklbw %%xmm7, %%xmm3\n"
+      "  pmullw %%xmm0, %%xmm3\n"
+      "  paddw %%xmm6, %%xmm3\n"
+      "  pmulhuw %%xmm4, %%xmm3\n"
+      "  paddw %%xmm1, %%xmm3\n"
+      "  packuswb %%xmm3, %%xmm3\n"
+      "  movq %%xmm3, (%0)\n"
+      "4:\n"
+      "  addl $8, %0\n"
+      "  addl $8, %1\n"
+      "  subl $1, %2\n"
+      "  jnz 3b\n"
+      :"+r" (dest), "+r" (src), "+r" (n)
+      :
+      :"eax");
+  }
+  if (end) {
+    __asm__ __volatile__ (
+      "  movl (%1), %%eax\n"
+      "  testl $0xff000000, %%eax\n"
+      "  jz 1f\n"
+
+      "  movd (%1), %%xmm1\n"
+      "  punpcklbw %%xmm7, %%xmm1\n"
+      "  pshuflw $0xff, %%xmm1, %%xmm0\n"
+      "  pxor %%xmm5, %%xmm0\n"
+
+      "  movd (%0), %%xmm3\n"
+      "  punpcklbw %%xmm7, %%xmm3\n"
+      "  pmullw %%xmm0, %%xmm3\n"
+      "  paddw %%xmm6, %%xmm3\n"
+      "  pmulhuw %%xmm4, %%xmm3\n"
+
+      "  paddw %%xmm1, %%xmm3\n"
+      "  packuswb %%xmm3, %%xmm3\n"
+      "  movd %%xmm3, (%0)\n"
+
+      "1:\n"
+      "  addl $4, %0\n"
+      "  addl $4, %1\n"
+      :"+r" (dest), "+r" (src), "+r" (n)
+      :
+      :"eax");
+  }
+
+}
+OIL_DEFINE_IMPL_FULL (composite_over_argb_sse2_2, composite_over_argb, OIL_IMPL_FLAG_SSE2);
+
+/* written for shaun */
+static void
+composite_over_argb_sse2_3 (uint32_t *dest, uint32_t *src, int n)
+{
+  int begin;
+  int middle;
+  int end;
+
+  __asm__ __volatile__ ("  pxor %%xmm7, %%xmm7\n"   // mm7 = { 0, 0, 0, 0 }
+      "  movl $0x80808080, %%eax\n"
+      "  movd %%eax, %%xmm6\n"  // mm6 = { 128, 128, 128, 128 }
+      "  punpcklbw %%xmm7, %%xmm6\n"
+      "  punpcklwd %%xmm6, %%xmm6\n"
+      "  movl $0xffffffff, %%eax\n" // mm5 = { 255, 255, 255, 255 }
+      "  movd %%eax, %%xmm5\n"
+      "  punpcklbw %%xmm7, %%xmm5\n"
+      "  punpcklwd %%xmm5, %%xmm5\n"
+      "  movl $0x02020202, %%eax\n"
+      "  movd %%eax, %%xmm4\n"
+      "  punpcklbw %%xmm7, %%xmm4\n"
+      "  paddw %%xmm5, %%xmm4\n" // mm4 = { 257, 257, 257, 257 }
+      "  punpcklwd %%xmm4, %%xmm4\n"
+      :
+      :
+      :"eax");
+
+  begin = 0x3 & (4 - (((unsigned long)dest & 0xf) >> 2));
+  if (begin>n) {
+    begin = n;
+    middle = 0;
+    end = 0;
+  } else {
+    middle = (n-begin)>>2;
+    end = n - begin - middle*4;
+  }
+
+  if (begin>0) {
+    __asm__ __volatile__ ("\n"
+      "1:\n"
+      "  movl (%1), %%eax\n"
+      "  testl $0xff000000, %%eax\n"
+      "  jz 2f\n"
+
+      "  movd (%1), %%xmm1\n"
+      "  punpcklbw %%xmm7, %%xmm1\n"
+      "  pshuflw $0xff, %%xmm1, %%xmm0\n"
+      "  pxor %%xmm5, %%xmm0\n"
+
+      "  movd (%0), %%xmm3\n"
+      "  punpcklbw %%xmm7, %%xmm3\n"
+      "  pmullw %%xmm0, %%xmm3\n"
+      "  paddw %%xmm6, %%xmm3\n"
+      "  pmulhuw %%xmm4, %%xmm3\n"
+
+      "  paddw %%xmm1, %%xmm3\n"
+      "  packuswb %%xmm3, %%xmm3\n"
+      "  movd %%xmm3, (%0)\n"
+
+      "2:\n"
+      "  addl $4, %0\n"
+      "  addl $4, %1\n"
+      "  subl $1, %2\n"
+      "  jnz 1b\n"
+      :"+r" (dest), "+r" (src), "+r" (begin)
+      :
+      :"eax");
+  }
+
+  if (middle>0){
+    __asm__ __volatile__ ("\n"
+      "1:\n"
+      "  movdqu (%1), %%xmm1\n"
+      "  movl (%1), %%eax\n"
+      "  orl 4(%1), %%eax\n"
+      "  orl 8(%1), %%eax\n"
+      "  orl 12(%1), %%eax\n"
+      "  test $0xff000000, %%eax\n"
+      "  jz 2f\n"
+      "  movdqa %%xmm1, %%xmm0\n"
+      "  punpcklbw %%xmm7, %%xmm1\n"
+      "  punpckhbw %%xmm7, %%xmm0\n"
+      "  pshuflw $0xff, %%xmm1, %%xmm1\n"
+      "  pshuflw $0xff, %%xmm0, %%xmm0\n"
+      "  pshufhw $0xff, %%xmm1, %%xmm1\n"
+      "  pshufhw $0xff, %%xmm0, %%xmm0\n"
+
+      "  pxor %%xmm5, %%xmm1\n"
+      "  pxor %%xmm5, %%xmm0\n"
+
+      "  movdqa (%0), %%xmm3\n"
+      "  movdqa %%xmm3, %%xmm2\n"
+      "  punpcklbw %%xmm7, %%xmm3\n"
+      "  punpckhbw %%xmm7, %%xmm2\n"
+
+      "  pmullw %%xmm1, %%xmm3\n"
+      "  paddw %%xmm6, %%xmm3\n"
+      "  pmulhuw %%xmm4, %%xmm3\n"
+      "  pmullw %%xmm0, %%xmm2\n"
+      "  paddw %%xmm6, %%xmm2\n"
+      "  pmulhuw %%xmm4, %%xmm2\n"
+      "  packuswb %%xmm2, %%xmm3\n"
+
+      "  movdqu (%1), %%xmm1\n"
+      "  paddb %%xmm1, %%xmm3\n"
+      "  movdqa %%xmm3, (%0)\n"
+      "2:\n"
+      "  addl $16, %0\n"
+      "  addl $16, %1\n"
+      "  subl $1, %2\n"
+      "  jnz 1b\n"
+      :"+r" (dest), "+r" (src), "+r" (middle)
+      :
+      :"eax");
+  }
+  if (end>0) {
+    __asm__ __volatile__ ("\n"
+      "1:\n"
+      "  movl (%1), %%eax\n"
+      "  testl $0xff000000, %%eax\n"
+      "  jz 2f\n"
+
+      "  movd (%1), %%xmm1\n"
+      "  punpcklbw %%xmm7, %%xmm1\n"
+      "  pshuflw $0xff, %%xmm1, %%xmm0\n"
+      "  pxor %%xmm5, %%xmm0\n"
+
+      "  movd (%0), %%xmm3\n"
+      "  punpcklbw %%xmm7, %%xmm3\n"
+      "  pmullw %%xmm0, %%xmm3\n"
+      "  paddw %%xmm6, %%xmm3\n"
+      "  pmulhuw %%xmm4, %%xmm3\n"
+
+      "  paddw %%xmm1, %%xmm3\n"
+      "  packuswb %%xmm3, %%xmm3\n"
+      "  movd %%xmm3, (%0)\n"
+
+      "2:\n"
+      "  addl $4, %0\n"
+      "  addl $4, %1\n"
+      "  subl $1, %2\n"
+      "  jnz 1b\n"
+      :"+r" (dest), "+r" (src), "+r" (end)
+      :
+      :"eax");
+  }
+
+}
+OIL_DEFINE_IMPL_FULL (composite_over_argb_sse2_3, composite_over_argb, OIL_IMPL_FLAG_SSE2);
+
 #if 0
 static void
 composite_over_argb_const_src_mmx (uint32_t *dest, uint32_t *src, int n)
diff --git a/liboil/liboilcpu.c b/liboil/liboilcpu.c
index 54968c4..1e33f68 100644
--- a/liboil/liboilcpu.c
+++ b/liboil/liboilcpu.c
@@ -180,6 +180,7 @@ oil_cpu_i386_getflags_cpuid (void)
   }
   if (edx & (1<<26)) {
     oil_cpu_flags |= OIL_IMPL_FLAG_SSE2;
+    oil_cpu_flags |= OIL_IMPL_FLAG_MMXEXT;
   }
   if (ecx & (1<<0)) {
     oil_cpu_flags |= OIL_IMPL_FLAG_SSE3;