Merge remote-tracking branch 'origin/released'

Conflicts:
	mesalib/src/mapi/glapi/gen/gl_gentable.py
	xorg-server/include/servermd.h

18 files changed, 155 insertions, 90 deletions

diff --git a/mesalib/src/gallium/Automake.inc b/mesalib/src/gallium/Automake.inc
index 4519c161c..70921ecd8 100644
--- a/mesalib/src/gallium/Automake.inc
+++ b/mesalib/src/gallium/Automake.inc
@@ -56,7 +56,9 @@ GALLIUM_DRI_LINKER_FLAGS = \
 	-shrext .so \
 	-module \
 	-avoid-version \
+if HAVE_LD_VERSION_SCRIPT
 	-Wl,--version-script=$(top_srcdir)/src/gallium/targets/dri.sym \
+endif
 	$(GC_SECTIONS)
 
 GALLIUM_VDPAU_LINKER_FLAGS = \
@@ -64,7 +66,9 @@ GALLIUM_VDPAU_LINKER_FLAGS = \
 	-module \
 	-no-undefined \
 	-version-number $(VDPAU_MAJOR):$(VDPAU_MINOR) \
+if HAVE_LD_VERSION_SCRIPT
 	-Wl,--version-script=$(top_srcdir)/src/gallium/targets/vdpau.sym \
+endif
 	$(GC_SECTIONS) \
 	$(LD_NO_UNDEFINED)
 
@@ -73,7 +77,9 @@ GALLIUM_XVMC_LINKER_FLAGS = \
 	-module \
 	-no-undefined \
 	-version-number $(XVMC_MAJOR):$(XVMC_MINOR) \
+if HAVE_LD_VERSION_SCRIPT
 	-Wl,--version-script=$(top_srcdir)/src/gallium/targets/xvmc.sym \
+endif
 	$(GC_SECTIONS) \
 	$(LD_NO_UNDEFINED)
 
@@ -82,7 +88,9 @@ GALLIUM_OMX_LINKER_FLAGS = \
 	-module \
 	-no-undefined \
 	-avoid-version \
+if HAVE_LD_VERSION_SCRIPT
 	-Wl,--version-script=$(top_srcdir)/src/gallium/targets/omx.sym \
+endif
 	$(GC_SECTIONS) \
 	$(LD_NO_UNDEFINED)
 
diff --git a/mesalib/src/mapi/glapi/Makefile.am b/mesalib/src/mapi/glapi/Makefile.am
index e425bacea..cadcd69d6 100644
--- a/mesalib/src/mapi/glapi/Makefile.am
+++ b/mesalib/src/mapi/glapi/Makefile.am
@@ -48,13 +48,13 @@ if HAVE_SHARED_GLAPI
 AM_CPPFLAGS += \
 	-DMAPI_MODE_BRIDGE \
 	-DMAPI_ABI_HEADER=\"glapi/glapi_mapi_tmp.h\"
-glapi_GLAPI_sources =
+glapi_GLAPI_sources = $(GLAPI_UTIL_SOURCES)
 glapi_ASM_sources =
 glapi_MAPI_sources = $(MAPI_BRIDGE_FILES)
 else
 AM_CPPFLAGS += \
 	-DMAPI_MODE_UTIL
-glapi_GLAPI_sources = $(GLAPI_SOURCES)
+glapi_GLAPI_sources = $(GLAPI_SOURCES) $(GLAPI_UTIL_SOURCES)
 glapi_ASM_sources = $(GLAPI_ASM_SOURCES)
 glapi_MAPI_sources = $(MAPI_UTIL_FILES)
 endif
diff --git a/mesalib/src/mapi/glapi/Makefile.sources b/mesalib/src/mapi/glapi/Makefile.sources
index 04fef6c9e..aae41a88d 100644
--- a/mesalib/src/mapi/glapi/Makefile.sources
+++ b/mesalib/src/mapi/glapi/Makefile.sources
@@ -1,9 +1,11 @@
 # src/mapi/glapi/Makefile.sources
 
+GLAPI_UTIL_SOURCES = \
+	glapi_gentable.c
+
 GLAPI_SOURCES = \
 	glapi_dispatch.c \
 	glapi_entrypoint.c \
-	glapi_gentable.c \
 	glapi_getproc.c \
 	glapi_nop.c \
 	glapi.c \
diff --git a/mesalib/src/mapi/glapi/gen/glX_proto_send.py b/mesalib/src/mapi/glapi/gen/glX_proto_send.py
index a98f63e99..1419a7cc1 100755
--- a/mesalib/src/mapi/glapi/gen/glX_proto_send.py
+++ b/mesalib/src/mapi/glapi/gen/glX_proto_send.py
@@ -947,6 +947,7 @@ class PrintGlxProtoInit_c(gl_XML.gl_print_base):
 #include "glapi.h"
 #include <assert.h>
 
+#ifndef GLX_USE_APPLEGL
 
 /**
  * No-op function used to initialize functions that have no GLX protocol
@@ -984,6 +985,8 @@ struct _glapi_table * __glXNewIndirectAPI( void )
         print """
     return (struct _glapi_table *) table;
 }
+
+#endif
 """
         return
 
diff --git a/mesalib/src/mapi/glapi/gen/gl_gentable.py b/mesalib/src/mapi/glapi/gen/gl_gentable.py
index 1c704452f..bbf985d77 100755
--- a/mesalib/src/mapi/glapi/gen/gl_gentable.py
+++ b/mesalib/src/mapi/glapi/gen/gl_gentable.py
@@ -113,7 +113,7 @@ __glapi_gentable_set_remaining_noop(struct _glapi_table *disp) {
 
 struct _glapi_table *
 _glapi_create_table_from_handle(void *handle, const char *symbol_prefix) {
-    struct _glapi_table *disp = calloc(_glapi_get_dispatch_table_size(), sizeof(void*));
+    struct _glapi_table *disp = calloc(1, _glapi_get_dispatch_table_size() * sizeof(_glapi_proc));
     char symboln[512];
 
     if(!disp)
diff --git a/mesalib/src/mesa/Makefile.am b/mesalib/src/mesa/Makefile.am
index 0fdc8f315..88eeff9ee 100644
--- a/mesalib/src/mesa/Makefile.am
+++ b/mesalib/src/mesa/Makefile.am
@@ -25,7 +25,7 @@ if HAVE_X11_DRIVER
 SUBDIRS += drivers/x11
 endif
 
-if HAVE_DRI
+if HAVE_DRICOMMON
 SUBDIRS += drivers/dri
 endif
 
diff --git a/mesalib/src/mesa/SConscript b/mesalib/src/mesa/SConscript
index cd959be65..f56578672 100644
--- a/mesalib/src/mesa/SConscript
+++ b/mesalib/src/mesa/SConscript
@@ -392,6 +392,7 @@ if (env['gcc'] or env['clang']) and \
             'USE_X86_64_ASM',
         ])
         mesa_sources += [
+            'x86/common_x86.c',
             'x86-64/x86-64.c',
             'x86-64/xform4.S',
         ]
diff --git a/mesalib/src/mesa/drivers/common/meta.c b/mesalib/src/mesa/drivers/common/meta.c
index 3ef3f7971..fec0d2be5 100644
--- a/mesalib/src/mesa/drivers/common/meta.c
+++ b/mesalib/src/mesa/drivers/common/meta.c
@@ -242,10 +242,25 @@ _mesa_meta_setup_blit_shader(struct gl_context *ctx,
                              GLenum target,
                              struct blit_shader_table *table)
 {
-   const char *vs_source;
-   char *fs_source;
+   char *vs_source, *fs_source;
    void *const mem_ctx = ralloc_context(NULL);
    struct blit_shader *shader = choose_blit_shader(target, table);
+   const char *vs_input, *vs_output, *fs_input, *vs_preprocess, *fs_preprocess;
+
+   if (ctx->Const.GLSLVersion < 130) {
+      vs_preprocess = "";
+      vs_input = "attribute";
+      vs_output = "varying";
+      fs_preprocess = "#extension GL_EXT_texture_array : enable";
+      fs_input = "varying";
+   } else {
+      vs_preprocess = "#version 130";
+      vs_input = "in";
+      vs_output = "out";
+      fs_preprocess = "#version 130";
+      fs_input = "in";
+      shader->func = "texture";
+   }
 
    assert(shader != NULL);
 
@@ -254,57 +269,30 @@ _mesa_meta_setup_blit_shader(struct gl_context *ctx,
       return;
    }
 
-   if (ctx->Const.GLSLVersion < 130) {
-      vs_source =
-         "attribute vec2 position;\n"
-         "attribute vec4 textureCoords;\n"
-         "varying vec4 texCoords;\n"
-         "void main()\n"
-         "{\n"
-         "   texCoords = textureCoords;\n"
-         "   gl_Position = vec4(position, 0.0, 1.0);\n"
-         "}\n";
-
-      fs_source = ralloc_asprintf(mem_ctx,
-                                  "#extension GL_EXT_texture_array : enable\n"
-                                  "#extension GL_ARB_texture_cube_map_array: enable\n"
-                                  "uniform %s texSampler;\n"
-                                  "varying vec4 texCoords;\n"
-                                  "void main()\n"
-                                  "{\n"
-                                  "   gl_FragColor = %s(texSampler, %s);\n"
-                                  "   gl_FragDepth = gl_FragColor.x;\n"
-                                  "}\n",
-                                  shader->type,
-                                  shader->func, shader->texcoords);
-   }
-   else {
-      vs_source = ralloc_asprintf(mem_ctx,
-                                  "#version 130\n"
-                                  "in vec2 position;\n"
-                                  "in vec4 textureCoords;\n"
-                                  "out vec4 texCoords;\n"
-                                  "void main()\n"
-                                  "{\n"
-                                  "   texCoords = textureCoords;\n"
-                                  "   gl_Position = vec4(position, 0.0, 1.0);\n"
-                                  "}\n");
-      fs_source = ralloc_asprintf(mem_ctx,
-                                  "#version 130\n"
-                                  "#extension GL_ARB_texture_cube_map_array: enable\n"
-                                  "uniform %s texSampler;\n"
-                                  "in vec4 texCoords;\n"
-                                  "out vec4 out_color;\n"
-                                  "\n"
-                                  "void main()\n"
-                                  "{\n"
-                                  "   out_color = texture(texSampler, %s);\n"
-                                  "   gl_FragDepth = out_color.x;\n"
-                                  "}\n",
-                                  shader->type,
-                                  shader->texcoords);
-   }
-
+   vs_source = ralloc_asprintf(mem_ctx,
+                "%s\n"
+                "%s vec2 position;\n"
+                "%s vec4 textureCoords;\n"
+                "%s vec4 texCoords;\n"
+                "void main()\n"
+                "{\n"
+                "   texCoords = textureCoords;\n"
+                "   gl_Position = vec4(position, 0.0, 1.0);\n"
+                "}\n",
+                vs_preprocess, vs_input, vs_input, vs_output);
+
+   fs_source = ralloc_asprintf(mem_ctx,
+                "%s\n"
+                "#extension GL_ARB_texture_cube_map_array: enable\n"
+                "uniform %s texSampler;\n"
+                "%s vec4 texCoords;\n"
+                "void main()\n"
+                "{\n"
+                "   gl_FragColor = %s(texSampler, %s);\n"
+                "   gl_FragDepth = gl_FragColor.x;\n"
+                "}\n",
+                fs_preprocess, shader->type, fs_input,
+                shader->func, shader->texcoords);
 
    _mesa_meta_compile_and_link_program(ctx, vs_source, fs_source,
                                        ralloc_asprintf(mem_ctx, "%s blit",
@@ -2860,13 +2848,13 @@ copytexsubimage_using_blit_framebuffer(struct gl_context *ctx, GLuint dims,
     * are too strict for CopyTexImage.  We know meta will be fine with format
     * changes.
     */
-   _mesa_meta_BlitFramebuffer(ctx, x, y,
-                              x + width, y + height,
-                              xoffset, yoffset,
-                              xoffset + width, yoffset + height,
-                              mask, GL_NEAREST);
+   mask = _mesa_meta_BlitFramebuffer(ctx, x, y,
+                                     x + width, y + height,
+                                     xoffset, yoffset,
+                                     xoffset + width, yoffset + height,
+                                     mask, GL_NEAREST);
    ctx->Meta->Blit.no_ctsi_fallback = false;
-   success = true;
+   success = mask == 0x0;
 
  out:
    _mesa_lock_texture(ctx, texObj);
diff --git a/mesalib/src/mesa/drivers/common/meta.h b/mesalib/src/mesa/drivers/common/meta.h
index 2186a39f8..007f1040b 100644
--- a/mesalib/src/mesa/drivers/common/meta.h
+++ b/mesalib/src/mesa/drivers/common/meta.h
@@ -422,13 +422,21 @@ _mesa_meta_setup_sampler(struct gl_context *ctx,
                          const struct gl_texture_object *texObj,
                          GLenum target, GLenum filter, GLuint srcLevel);
 
-extern void
+extern GLbitfield
 _mesa_meta_BlitFramebuffer(struct gl_context *ctx,
                            GLint srcX0, GLint srcY0, GLint srcX1, GLint srcY1,
                            GLint dstX0, GLint dstY0, GLint dstX1, GLint dstY1,
                            GLbitfield mask, GLenum filter);
 
 extern void
+_mesa_meta_and_swrast_BlitFramebuffer(struct gl_context *ctx,
+                                      GLint srcX0, GLint srcY0,
+                                      GLint srcX1, GLint srcY1,
+                                      GLint dstX0, GLint dstY0,
+                                      GLint dstX1, GLint dstY1,
+                                      GLbitfield mask, GLenum filter);
+
+extern void
 _mesa_meta_Clear(struct gl_context *ctx, GLbitfield buffers);
 
 extern void
diff --git a/mesalib/src/mesa/drivers/common/meta_blit.c b/mesalib/src/mesa/drivers/common/meta_blit.c
index e5a0a9ad0..707269dd6 100644
--- a/mesalib/src/mesa/drivers/common/meta_blit.c
+++ b/mesalib/src/mesa/drivers/common/meta_blit.c
@@ -325,10 +325,15 @@ setup_glsl_blit_framebuffer(struct gl_context *ctx,
                             struct gl_renderbuffer *src_rb,
                             GLenum target)
 {
+   unsigned texcoord_size;
+
    /* target = GL_TEXTURE_RECTANGLE is not supported in GLES 3.0 */
    assert(_mesa_is_desktop_gl(ctx) || target == GL_TEXTURE_2D);
 
-   _mesa_meta_setup_vertex_objects(&blit->VAO, &blit->VBO, true, 2, 2, 0);
+   texcoord_size = 2 + (src_rb->Depth > 1 ? 1 : 0);
+
+   _mesa_meta_setup_vertex_objects(&blit->VAO, &blit->VBO, true,
+                                   2, texcoord_size, 0);
 
    if (target == GL_TEXTURE_2D_MULTISAMPLE ||
        target == GL_TEXTURE_2D_MULTISAMPLE_ARRAY) {
@@ -533,12 +538,16 @@ blitframebuffer_texture(struct gl_context *ctx,
 
       verts[0].tex[0] = s0;
       verts[0].tex[1] = t0;
+      verts[0].tex[2] = readAtt->Zoffset;
       verts[1].tex[0] = s1;
       verts[1].tex[1] = t0;
+      verts[1].tex[2] = readAtt->Zoffset;
       verts[2].tex[0] = s1;
       verts[2].tex[1] = t1;
+      verts[2].tex[2] = readAtt->Zoffset;
       verts[3].tex[0] = s0;
       verts[3].tex[1] = t1;
+      verts[3].tex[2] = readAtt->Zoffset;
 
       _mesa_BufferSubData(GL_ARRAY_BUFFER_ARB, 0, sizeof(verts), verts);
    }
@@ -644,7 +653,7 @@ _mesa_meta_setup_sampler(struct gl_context *ctx,
  * Meta implementation of ctx->Driver.BlitFramebuffer() in terms
  * of texture mapping and polygon rendering.
  */
-void
+GLbitfield
 _mesa_meta_BlitFramebuffer(struct gl_context *ctx,
                            GLint srcX0, GLint srcY0, GLint srcX1, GLint srcY1,
                            GLint dstX0, GLint dstY0, GLint dstX1, GLint dstY1,
@@ -669,7 +678,7 @@ _mesa_meta_BlitFramebuffer(struct gl_context *ctx,
    /* Multisample texture blit support requires texture multisample. */
    if (ctx->ReadBuffer->Visual.samples > 0 &&
        !ctx->Extensions.ARB_texture_multisample) {
-      goto fallback;
+      return mask;
    }
 
    /* Clip a copy of the blit coordinates. If these differ from the input
@@ -678,7 +687,7 @@ _mesa_meta_BlitFramebuffer(struct gl_context *ctx,
    if (!_mesa_clip_blit(ctx, &clip.srcX0, &clip.srcY0, &clip.srcX1, &clip.srcY1,
                         &clip.dstX0, &clip.dstY0, &clip.dstX1, &clip.dstY1)) {
       /* clipped/scissored everything away */
-      return;
+      return 0;
    }
 
    /* Only scissor affects blit, but we're doing to set a custom scissor if
@@ -705,10 +714,6 @@ _mesa_meta_BlitFramebuffer(struct gl_context *ctx,
                                   filter, dstFlipX, dstFlipY,
                                   use_glsl_version, false)) {
          mask &= ~GL_COLOR_BUFFER_BIT;
-         if (mask == 0x0) {
-            _mesa_meta_end(ctx);
-            return;
-         }
       }
    }
 
@@ -718,10 +723,6 @@ _mesa_meta_BlitFramebuffer(struct gl_context *ctx,
                                   filter, dstFlipX, dstFlipY,
                                   use_glsl_version, true)) {
          mask &= ~GL_DEPTH_BUFFER_BIT;
-         if (mask == 0x0) {
-            _mesa_meta_end(ctx);
-            return;
-         }
       }
    }
 
@@ -731,11 +732,7 @@ _mesa_meta_BlitFramebuffer(struct gl_context *ctx,
 
    _mesa_meta_end(ctx);
 
-fallback:
-   if (mask && !ctx->Meta->Blit.no_ctsi_fallback) {
-      _swrast_BlitFramebuffer(ctx, srcX0, srcY0, srcX1, srcY1,
-                              dstX0, dstY0, dstX1, dstY1, mask, filter);
-   }
+   return mask;
 }
 
 void
@@ -753,3 +750,24 @@ _mesa_meta_glsl_blit_cleanup(struct blit_state *blit)
    _mesa_DeleteTextures(1, &blit->depthTex.TexObj);
    blit->depthTex.TexObj = 0;
 }
+
+void
+_mesa_meta_and_swrast_BlitFramebuffer(struct gl_context *ctx,
+                                      GLint srcX0, GLint srcY0,
+                                      GLint srcX1, GLint srcY1,
+                                      GLint dstX0, GLint dstY0,
+                                      GLint dstX1, GLint dstY1,
+                                      GLbitfield mask, GLenum filter)
+{
+   mask = _mesa_meta_BlitFramebuffer(ctx,
+                                     srcX0, srcY0, srcX1, srcY1,
+                                     dstX0, dstY0, dstX1, dstY1,
+                                     mask, filter);
+   if (mask == 0x0)
+      return;
+
+   _swrast_BlitFramebuffer(ctx,
+                           srcX0, srcY0, srcX1, srcY1,
+                           dstX0, dstY0, dstX1, dstY1,
+                           mask, filter);
+}
diff --git a/mesalib/src/mesa/drivers/dri/Makefile.am b/mesalib/src/mesa/drivers/dri/Makefile.am
index e8076120d..70039f9ad 100644
--- a/mesalib/src/mesa/drivers/dri/Makefile.am
+++ b/mesalib/src/mesa/drivers/dri/Makefile.am
@@ -52,7 +52,7 @@ nodist_EXTRA_mesa_dri_drivers_la_SOURCES = dummy.cpp
 mesa_dri_drivers_la_SOURCES =
 mesa_dri_drivers_la_LDFLAGS = \
         -module -avoid-version -shared -shrext .so \
-        -Wl,-Bsymbolic \
+        $(BSYMBOLIC) \
         $(GC_SECTIONS) \
         $()
 mesa_dri_drivers_la_LIBADD = \
diff --git a/mesalib/src/mesa/main/compiler.h b/mesalib/src/mesa/main/compiler.h
index 97075f58e..600691724 100644
--- a/mesalib/src/mesa/main/compiler.h
+++ b/mesalib/src/mesa/main/compiler.h
@@ -264,6 +264,12 @@ static INLINE GLuint CPU_TO_LE32(GLuint x)
 #define unreachable()
 #endif
 
+/*
+ * A trick to suppress uninitialized variable warning without generating any
+ * code
+ */
+#define uninitialized_var(x) x = x
+
 #if (__GNUC__ >= 3)
 #define PRINTFLIKE(f, a) __attribute__ ((format(__printf__, f, a)))
 #else
diff --git a/mesalib/src/mesa/main/cpuinfo.c b/mesalib/src/mesa/main/cpuinfo.c
index 8d482a6ac..0755d6b21 100644
--- a/mesalib/src/mesa/main/cpuinfo.c
+++ b/mesalib/src/mesa/main/cpuinfo.c
@@ -34,7 +34,7 @@
 void
 _mesa_get_cpu_features(void)
 {
-#ifdef USE_X86_ASM
+#if defined USE_X86_ASM || defined USE_X86_64_ASM
    _mesa_get_x86_features();
 #endif
 }
diff --git a/mesalib/src/mesa/main/cpuinfo.h b/mesalib/src/mesa/main/cpuinfo.h
index 367888905..57925e82b 100644
--- a/mesalib/src/mesa/main/cpuinfo.h
+++ b/mesalib/src/mesa/main/cpuinfo.h
@@ -27,7 +27,7 @@
 #define CPUINFO_H
 
 
-#if defined(USE_X86_ASM)
+#if defined USE_X86_ASM || defined USE_X86_64_ASM
 #include "x86/common_x86_asm.h"
 #endif
 
diff --git a/mesalib/src/mesa/main/fbobject.c b/mesalib/src/mesa/main/fbobject.c
index 97538bc7b..ae3a418cd 100644
--- a/mesalib/src/mesa/main/fbobject.c
+++ b/mesalib/src/mesa/main/fbobject.c
@@ -500,6 +500,12 @@ _mesa_framebuffer_renderbuffer(struct gl_context *ctx,
    }
    else {
       remove_attachment(ctx, att);
+      if (attachment == GL_DEPTH_STENCIL_ATTACHMENT) {
+         /* detach stencil (depth was detached above) */
+         att = get_attachment(ctx, fb, GL_STENCIL_ATTACHMENT_EXT);
+         assert(att);
+         remove_attachment(ctx, att);
+      }
    }
 
    invalidate_framebuffer(fb);
diff --git a/mesalib/src/mesa/vbo/vbo_exec_array.c b/mesalib/src/mesa/vbo/vbo_exec_array.c
index 07ce08fef..9c161ccca 100644
--- a/mesalib/src/mesa/vbo/vbo_exec_array.c
+++ b/mesalib/src/mesa/vbo/vbo_exec_array.c
@@ -581,9 +581,9 @@ vbo_handle_primitive_restart(struct gl_context *ctx,
 {
    struct vbo_context *vbo = vbo_context(ctx);
 
-   if ((ib != NULL) &&
-       ctx->Const.PrimitiveRestartInSoftware &&
-       ctx->Array._PrimitiveRestart) {
+   if (ctx->Const.PrimitiveRestartInSoftware &&
+       ctx->Array._PrimitiveRestart &&
+       (ib != NULL)) {
       /* Handle primitive restart in software */
       vbo_sw_primitive_restart(ctx, prim, nr_prims, ib, NULL);
    } else {
diff --git a/mesalib/src/mesa/x86/common_x86.c b/mesalib/src/mesa/x86/common_x86.c
index 3c1adc916..261f49170 100644
--- a/mesalib/src/mesa/x86/common_x86.c
+++ b/mesalib/src/mesa/x86/common_x86.c
@@ -47,6 +47,13 @@
 #include <sys/sysctl.h>
 #include <machine/cpu.h>
 #endif
+#if defined(USE_X86_64_ASM)
+#include <cpuid.h>
+#if !defined(bit_SSE4_1) && defined(bit_SSE41)
+/* XXX: clang defines bit_SSE41 instead of bit_SSE4_1 */
+#define bit_SSE4_1 bit_SSE41
+#endif
+#endif
 
 #include "main/imports.h"
 #include "common_x86_asm.h"
@@ -223,7 +230,7 @@ _mesa_get_x86_features(void)
        _mesa_debug(NULL, "CPUID not detected\n");
    }
    else {
-       GLuint cpu_features;
+       GLuint cpu_features, cpu_features_ecx;
        GLuint cpu_ext_features;
        GLuint cpu_ext_info;
        char cpu_vendor[13];
@@ -238,6 +245,7 @@ _mesa_get_x86_features(void)
 
        /* get cpu features */
        cpu_features = _mesa_x86_cpuid_edx(1);
+       cpu_features_ecx = _mesa_x86_cpuid_ecx(1);
 
        if (cpu_features & X86_CPU_FPU)
 	   _mesa_x86_cpu_features |= X86_FEATURE_FPU;
@@ -254,6 +262,8 @@ _mesa_get_x86_features(void)
 	   _mesa_x86_cpu_features |= X86_FEATURE_XMM;
        if (cpu_features & X86_CPU_XMM2)
 	   _mesa_x86_cpu_features |= X86_FEATURE_XMM2;
+       if (cpu_features_ecx & X86_CPU_SSE4_1)
+	   _mesa_x86_cpu_features |= X86_FEATURE_SSE4_1;
 #endif
 
        /* query extended cpu features */
@@ -330,7 +340,18 @@ _mesa_get_x86_features(void)
    }
 #endif
 
-#endif /* USE_X86_ASM */
+#elif defined(USE_X86_64_ASM)
+   unsigned int uninitialized_var(eax), uninitialized_var(ebx),
+                uninitialized_var(ecx), uninitialized_var(edx);
+
+   /* Always available on x86-64. */
+   _mesa_x86_cpu_features |= X86_FEATURE_XMM | X86_FEATURE_XMM2;
+
+   __get_cpuid(1, &eax, &ebx, &ecx, &edx);
+
+   if (ecx & bit_SSE4_1)
+      _mesa_x86_cpu_features |= X86_FEATURE_SSE4_1;
+#endif /* USE_X86_64_ASM */
 
    (void) detection_debug;
 }
diff --git a/mesalib/src/mesa/x86/common_x86_features.h b/mesalib/src/mesa/x86/common_x86_features.h
index 862548440..66f2cf651 100644
--- a/mesalib/src/mesa/x86/common_x86_features.h
+++ b/mesalib/src/mesa/x86/common_x86_features.h
@@ -43,6 +43,7 @@
 #define X86_FEATURE_XMM2	(1<<6)
 #define X86_FEATURE_3DNOWEXT	(1<<7)
 #define X86_FEATURE_3DNOW	(1<<8)
+#define X86_FEATURE_SSE4_1	(1<<9)
 
 /* standard X86 CPU features */
 #define X86_CPU_FPU		(1<<0)
@@ -50,6 +51,8 @@
 #define X86_CPU_MMX		(1<<23)
 #define X86_CPU_XMM		(1<<25)
 #define X86_CPU_XMM2		(1<<26)
+/* ECX. */
+#define X86_CPU_SSE4_1		(1<<19)
 
 /* extended X86 CPU features */
 #define X86_CPUEXT_MMX_EXT	(1<<22)
@@ -62,6 +65,7 @@
 #define cpu_has_xmm2		(_mesa_x86_cpu_features & X86_FEATURE_XMM2)
 #define cpu_has_3dnow		(_mesa_x86_cpu_features & X86_FEATURE_3DNOW)
 #define cpu_has_3dnowext	(_mesa_x86_cpu_features & X86_FEATURE_3DNOWEXT)
+#define cpu_has_sse4_1		(_mesa_x86_cpu_features & X86_FEATURE_SSE4_1)
 
 #endif