mesa: Resurrect SPARC asm code.

This rewrites the sparc GLAPI code so that it's PIC friendly and works
with all of the TLS/PTHREADS/64-bit/32-bit combinations properly.

As a result we can turn SPARC asm back on.  Currently it's only
enabled on Linux, as that's the only place where I can test this
stuff out.

For the moment the cliptest SPARC asm routines are disabled as they
are non-working.  The problem is that they use register %g7 as a
temporary which is where the threading libraries store the thread
pointer on SPARC.  I will fix that code up in a future change as it's
a pretty important routine to optimize.

Like x86 we do the runtime patch as a pthread once-invoked initializer
in init_glapi_relocs().

Unlike x86, however, our GLAPI stubs on SPARC are just two instruction
sequences that branch to a trampoline and put the GLAPI offset into a
register.  The trampoline is what we run-time patch.  The stubs thus
all look like:

glFoo:
	ba		__glapi_sparc_foo_stub
	 sethi		GLAPI_OFFSET(glFOO) * PTR_SIZE, %g3

This actually makes generate_entrypoint() a lot simpler on SPARC.  For
this case in generate_entrypoint() we generate stubs using a 'call'
instead of the 'ba' above to make sure it can reach.

In order to get a proper tail call going here, in the unpatched case,
we do several tricks.  To get the current PC, for example, we save the
return address register into a temporary, do a call, save the return
address register written by the call to another temporary, then
restore the original return address register value.  This is to
avoid having to allocate a stack frame.

This is necessary for PIC address formation.

This new GLAPI scheme lets us get rid of the ugly SPARC GLAPI hacks in
__glXInitialize() and one_time_init().

Signed-off-by: David S. Miller <davem@davemloft.net>

3 files changed, 328 insertions, 109 deletions

diff --git a/src/mesa/glapi/gl_SPARC_asm.py b/src/mesa/glapi/gl_SPARC_asm.py
index 178c00a227..33e752df38 100644
--- a/src/mesa/glapi/gl_SPARC_asm.py
+++ b/src/mesa/glapi/gl_SPARC_asm.py
@@ -39,80 +39,202 @@ class PrintGenericStubs(gl_XML.gl_print_base):
 
 
 	def printRealHeader(self):
-		print '#include "glapioffsets.h"'
+		print '#include "glapi/glapioffsets.h"'
 		print ''
 		print '#ifdef __arch64__'
-		print '#  define GL_STUB(fn,off)\t\t\t\t\\'
-		print 'fn:\t\t\t\t\t\\'
-		print '\tsethi\t%hi(0xDEADBEEF), %g4 ;\t\t\t\\'
-		print '\tsethi\t%hi(0xDEADBEEF), %g1 ;\t\t\t\\'
-		print '\tor\t%g4, %lo(0xDEADBEEF), %g4 ;\t\t\\'
-		print '\tor\t%g1, %lo(0xDEADBEEF), %g1 ;\t\t\\'
-		print '\tsllx\t%g4, 32, %g4 ;\t\t\t\t\\'
-		print '\tldx\t[%g1 + %g4], %g1 ;\t\t\t\\'
-		print '\tsethi\t%hi(8 * off), %g4 ;\t\t\t\\'
-		print '\tor\t%g4, %lo(8 * off), %g4 ;\t\t\\'
-		print '\tldx\t[%g1 + %g4], %g5 ;\t\t\t\\'
-		print '\tjmpl\t%g5, %g0 ;\t\t\t\t\\'
-		print '\tnop'
+		print '#define GL_OFF(N)\t((N) * 8)'
+		print '#define GL_LL\t\tldx'
+		print '#define GL_TIE_LD(SYM)\t%tie_ldx(SYM)'
+		print '#define GL_STACK_SIZE\t128'
 		print '#else'
-		print '#  define GL_STUB(fn,off)\t\t\t\t\\'
-		print 'fn:\t\t\t\t\t\\'
-		print '\tsethi\t%hi(0xDEADBEEF), %g1 ;\t\t\t\\'
-		print '\tld\t[%g1 + %lo(0xDEADBEEF)], %g1 ;\t\t\\'
-		print '\tld\t[%g1 + (4 * off)], %g5 ;\t\t\\'
-		print '\tjmpl\t%g5, %g0 ;\t\t\t\t\\'
-		print '\tnop'
+		print '#define GL_OFF(N)\t((N) * 4)'
+		print '#define GL_LL\t\tld'
+		print '#define GL_TIE_LD(SYM)\t%tie_ld(SYM)'
+		print '#define GL_STACK_SIZE\t64'
 		print '#endif'
 		print ''
-		print '#define GL_STUB_ALIAS(fn,alias) fn = alias'
+		print '#define GLOBL_FN(x) .globl x ; .type x, @function'
+		print '#define HIDDEN(x) .hidden x'
 		print ''
-		print '.text'
-		print '.align 32'
-		print '\t\t.globl __glapi_sparc_icache_flush ; .type __glapi_sparc_icache_flush,#function'
+		print '\t.register %g2, #scratch'
+		print '\t.register %g3, #scratch'
+		print ''
+		print '\t.text'
+		print ''
+		print '\tGLOBL_FN(__glapi_sparc_icache_flush)'
+		print '\tHIDDEN(__glapi_sparc_icache_flush)'
+		print '\t.type\t__glapi_sparc_icache_flush, @function'
 		print '__glapi_sparc_icache_flush: /* %o0 = insn_addr */'
 		print '\tflush\t%o0'
 		print '\tretl'
-		print '\tnop'
+		print '\t nop'
+		print ''
+		print '\t.align\t32'
+		print ''
+		print '\t.type\t__glapi_sparc_get_pc, @function'
+		print '__glapi_sparc_get_pc:'
+		print '\tretl'
+		print '\t add\t%o7, %g2, %g2'
+		print '\t.size\t__glapi_sparc_get_pc, .-__glapi_sparc_get_pc'
+		print ''
+		print '#ifdef GLX_USE_TLS'
+		print ''
+		print '\tGLOBL_FN(__glapi_sparc_get_dispatch)'
+		print '\tHIDDEN(__glapi_sparc_get_dispatch)'
+		print '__glapi_sparc_get_dispatch:'
+		print '\tmov\t%o7, %g1'
+		print '\tsethi\t%hi(_GLOBAL_OFFSET_TABLE_-4), %g2'
+		print '\tcall\t__glapi_sparc_get_pc'
+		print '\tadd\t%g2, %lo(_GLOBAL_OFFSET_TABLE_+4), %g2'
+		print '\tmov\t%g1, %o7'
+		print '\tsethi\t%tie_hi22(_glapi_tls_Dispatch), %g1'
+		print '\tadd\t%g1, %tie_lo10(_glapi_tls_Dispatch), %g1'
+		print '\tGL_LL\t[%g2 + %g1], %g2, GL_TIE_LD(_glapi_tls_Dispatch)'
+		print '\tretl'
+		print '\t mov\t%g2, %o0'
+		print ''
+		print '\t.data'
+		print '\t.align\t32'
+		print ''
+		print '\t/* --> sethi %hi(_glapi_tls_Dispatch), %g1 */'
+		print '\t/* --> or %g1, %lo(_glapi_tls_Dispatch), %g1 */'
+		print '\tGLOBL_FN(__glapi_sparc_tls_stub)'
+		print '\tHIDDEN(__glapi_sparc_tls_stub)'
+		print '__glapi_sparc_tls_stub: /* Call offset in %g3 */'
+		print '\tmov\t%o7, %g1'
+		print '\tsethi\t%hi(_GLOBAL_OFFSET_TABLE_-4), %g2'
+		print '\tcall\t__glapi_sparc_get_pc'
+		print '\tadd\t%g2, %lo(_GLOBAL_OFFSET_TABLE_+4), %g2'
+		print '\tmov\t%g1, %o7'
+		print '\tsrl\t%g3, 10, %g3'
+		print '\tsethi\t%tie_hi22(_glapi_tls_Dispatch), %g1'
+		print '\tadd\t%g1, %tie_lo10(_glapi_tls_Dispatch), %g1'
+		print '\tGL_LL\t[%g2 + %g1], %g2, GL_TIE_LD(_glapi_tls_Dispatch)'
+		print '\tGL_LL\t[%g7+%g2], %g1'
+		print '\tGL_LL\t[%g1 + %g3], %g1'
+		print '\tjmp\t%g1'
+		print '\t nop'
+		print '\t.size\t__glapi_sparc_tls_stub, .-__glapi_sparc_tls_stub'
+		print ''
+		print '#define GL_STUB(fn, off)\t\t\t\t\\'
+		print '\tGLOBL_FN(fn);\t\t\t\t\t\\'
+		print 'fn:\tba\t__glapi_sparc_tls_stub;\t\t\t\\'
+		print '\t sethi\tGL_OFF(off), %g3;\t\t\t\\'
+		print '\t.size\tfn,.-fn;'
+		print ''
+		print '#elif defined(PTHREADS)'
+		print ''
+		print '\t/* 64-bit 0x00 --> sethi %hh(_glapi_Dispatch), %g1 */'
+		print '\t/* 64-bit 0x04 --> sethi %lm(_glapi_Dispatch), %g2 */'
+		print '\t/* 64-bit 0x08 --> or %g1, %hm(_glapi_Dispatch), %g1 */'
+		print '\t/* 64-bit 0x0c --> sllx %g1, 32, %g1 */'
+		print '\t/* 64-bit 0x10 --> add %g1, %g2, %g1 */'
+		print '\t/* 64-bit 0x14 --> ldx [%g1 + %lo(_glapi_Dispatch)], %g1 */'
+		print ''
+		print '\t/* 32-bit 0x00 --> sethi %hi(_glapi_Dispatch), %g1 */'
+		print '\t/* 32-bit 0x04 --> ld [%g1 + %lo(_glapi_Dispatch)], %g1 */'
+		print ''
+		print '\t.data'
+		print '\t.align\t32'
+		print ''
+		print '\tGLOBL_FN(__glapi_sparc_pthread_stub)'
+		print '\tHIDDEN(__glapi_sparc_pthread_stub)'
+		print '__glapi_sparc_pthread_stub: /* Call offset in %g3 */'
+		print '\tmov\t%o7, %g1'
+		print '\tsethi\t%hi(_GLOBAL_OFFSET_TABLE_-4), %g2'
+		print '\tcall\t__glapi_sparc_get_pc'
+		print '\t add\t%g2, %lo(_GLOBAL_OFFSET_TABLE_+4), %g2'
+		print '\tmov\t%g1, %o7'
+		print '\tsethi\t%hi(_glapi_Dispatch), %g1'
+		print '\tor\t%g1, %lo(_glapi_Dispatch), %g1'
+		print '\tsrl\t%g3, 10, %g3'
+		print '\tGL_LL\t[%g2+%g1], %g2'
+		print '\tGL_LL\t[%g2], %g1'
+		print '\tcmp\t%g1, 0'
+		print '\tbe\t2f'
+		print '\t nop'
+		print '1:\tGL_LL\t[%g1 + %g3], %g1'
+		print '\tjmp\t%g1'
+		print '\t nop'
+		print '2:\tsave\t%sp, GL_STACK_SIZE, %sp'
+		print '\tmov\t%g3, %l0'
+		print '\tcall\t_glapi_get_dispatch'
+		print '\t nop'
+		print '\tmov\t%o0, %g1'
+		print '\tmov\t%l0, %g3'
+		print '\tba\t1b'
+		print '\t restore %g0, %g0, %g0'
+		print '\t.size\t__glapi_sparc_pthread_stub, .-__glapi_sparc_pthread_stub'
+		print ''
+		print '#define GL_STUB(fn, off)\t\t\t\\'
+		print '\tGLOBL_FN(fn);\t\t\t\t\\'
+		print 'fn:\tba\t__glapi_sparc_pthread_stub;\t\\'
+		print '\t sethi\tGL_OFF(off), %g3;\t\t\\'
+		print '\t.size\tfn,.-fn;'
+		print ''
+		print '#else /* Non-threaded version. */'
 		print ''
-		print '.data'
-		print '.align 64'
+		print '\t.type	__glapi_sparc_nothread_stub, @function'
+		print '__glapi_sparc_nothread_stub: /* Call offset in %g3 */'
+		print '\tmov\t%o7, %g1'
+		print '\tsethi\t%hi(_GLOBAL_OFFSET_TABLE_-4), %g2'
+		print '\tcall\t__glapi_sparc_get_pc'
+		print '\t add\t%g2, %lo(_GLOBAL_OFFSET_TABLE_+4), %g2'
+		print '\tmov\t%g1, %o7'
+		print '\tsrl\t%g3, 10, %g3'
+		print '\tsethi\t%hi(_glapi_Dispatch), %g1'
+		print '\tor\t%g1, %lo(_glapi_Dispatch), %g1'
+		print '\tGL_LL\t[%g2+%g1], %g2'
+		print '\tGL_LL\t[%g2], %g1'
+		print '\tGL_LL\t[%g1 + %g3], %g1'
+		print '\tjmp\t%g1'
+		print '\t nop'
+		print '\t.size\t__glapi_sparc_nothread_stub, .-__glapi_sparc_nothread_stub'
+		print ''
+		print '#define GL_STUB(fn, off)\t\t\t\\'
+		print '\tGLOBL_FN(fn);\t\t\t\t\\'
+		print 'fn:\tba\t__glapi_sparc_nothread_stub;\t\\'
+		print '\t sethi\tGL_OFF(off), %g3;\t\t\\'
+		print '\t.size\tfn,.-fn;'
+		print ''
+		print '#endif'
+		print ''
+		print '#define GL_STUB_ALIAS(fn, alias)		\\'
+		print '	.globl	fn;				\\'
+		print '	.set	fn, alias'
+		print ''
+		print '\t.text'
+		print '\t.align\t32'
+		print ''
+		print '\t.globl\tgl_dispatch_functions_start'
+		print '\tHIDDEN(gl_dispatch_functions_start)'
+		print 'gl_dispatch_functions_start:'
 		print ''
 		return
 
+	def printRealFooter(self):
+		print ''
+		print '\t.globl\tgl_dispatch_functions_end'
+		print '\tHIDDEN(gl_dispatch_functions_end)'
+		print 'gl_dispatch_functions_end:'
+		return
 
 	def printBody(self, api):
 		for f in api.functionIterateByOffset():
-			if f.is_static_entry_point(f.name):
-				name = f.name
-			else:
-				name = "_dispatch_stub_%u" % (f.offset)
+			name = f.dispatch_name()
 
-			print '\t\t.globl gl%s ; .type gl%s,#function' % (name, name)
+			print '\tGL_STUB(gl%s, _gloffset_%s)' % (name, f.name)
 
-		print '\t\t.globl _mesa_sparc_glapi_begin ; .type _mesa_sparc_glapi_begin,#function'
-		print '_mesa_sparc_glapi_begin:'
-		print ''
+			if not f.is_static_entry_point(f.name):
+				print '\tHIDDEN(gl%s)' % (name)
 
 		for f in api.functionIterateByOffset():
-			if f.is_static_entry_point(f.name):
-				name = f.name
-			else:
-				name = "_dispatch_stub_%u" % (f.offset)
-
-			print '\tGL_STUB(gl%s, _gloffset_%s)' % (name, name)
+			name = f.dispatch_name()
 
-		print ''
-		print '\t\t.globl _mesa_sparc_glapi_end ; .type _mesa_sparc_glapi_end,#function'
-		print '_mesa_sparc_glapi_end:'
-		print ''
-
-
-		for f in api.functionIterateByOffset():
-			for n in f.entry_points:
-				if n != f.name:
-					if f.is_static_entry_point(n):
-						text = '\t.globl gl%s ; .type gl%s,#function ; gl%s = gl%s' % (n, n, n, f.name)
+			if f.is_static_entry_point(f.name):
+				for n in f.entry_points:
+					if n != f.name:
+						text = '\tGL_STUB_ALIAS(gl%s, gl%s)' % (n, f.name)
 
 						if f.has_different_protocol(n):
 							print '#ifndef GLX_INDIRECT_RENDERING'
diff --git a/src/mesa/glapi/glapi.c b/src/mesa/glapi/glapi.c
index e20618abc1..2b105d0f17 100644
--- a/src/mesa/glapi/glapi.c
+++ b/src/mesa/glapi/glapi.c
@@ -318,6 +318,14 @@ extern const GLubyte gl_dispatch_functions_start[];
 # endif
 #endif
 
+#ifdef USE_SPARC_ASM
+#ifdef GLX_USE_TLS
+extern unsigned int __glapi_sparc_tls_stub;
+#else
+extern unsigned int __glapi_sparc_pthread_stub;
+#endif
+#endif
+
 #if !defined(DISPATCH_FUNCTION_SIZE) && !defined(XFree86Server) && !defined(XGLServer)
 # define NEED_FUNCTION_POINTER
 #endif
@@ -344,6 +352,129 @@ init_glapi_relocs( void )
 	curr_func += DISPATCH_FUNCTION_SIZE;
     }
 #endif
+#ifdef USE_SPARC_ASM
+    extern void __glapi_sparc_icache_flush(unsigned int *);
+    static const unsigned int template[] = {
+#ifdef GLX_USE_TLS
+	0x05000000, /* sethi %hi(_glapi_tls_Dispatch), %g2 */
+	0x8730e00a, /* srl %g3, 10, %g3 */
+	0x8410a000, /* or %g2, %lo(_glapi_tls_Dispatch), %g2 */
+#ifdef __arch64__
+	0xc259c002, /* ldx [%g7 + %g2], %g1 */
+	0xc2584003, /* ldx [%g1 + %g3], %g1 */
+#else
+	0xc201c002, /* ld [%g7 + %g2], %g1 */
+	0xc2004003, /* ld [%g1 + %g3], %g1 */
+#endif
+	0x81c04000, /* jmp %g1 */
+	0x01000000, /* nop  */
+#else
+#ifdef __arch64__
+	0x03000000, /* 64-bit 0x00 --> sethi %hh(_glapi_Dispatch), %g1 */
+	0x05000000, /* 64-bit 0x04 --> sethi %lm(_glapi_Dispatch), %g2 */
+	0x82106000, /* 64-bit 0x08 --> or %g1, %hm(_glapi_Dispatch), %g1 */
+	0x8730e00a, /* 64-bit 0x0c --> srl %g3, 10, %g3 */
+	0x83287020, /* 64-bit 0x10 --> sllx %g1, 32, %g1 */
+	0x82004002, /* 64-bit 0x14 --> add %g1, %g2, %g1 */
+	0xc2586000, /* 64-bit 0x18 --> ldx [%g1 + %lo(_glapi_Dispatch)], %g1 */
+#else
+	0x03000000, /* 32-bit 0x00 --> sethi %hi(_glapi_Dispatch), %g1 */
+	0x8730e00a, /* 32-bit 0x04 --> srl %g3, 10, %g3 */
+	0xc2006000, /* 32-bit 0x08 --> ld [%g1 + %lo(_glapi_Dispatch)], %g1 */
+#endif
+	0x80a06000, /*             --> cmp %g1, 0 */
+	0x02800005, /*             --> be +4*5 */
+	0x01000000, /*             -->  nop  */
+#ifdef __arch64__
+	0xc2584003, /* 64-bit      --> ldx [%g1 + %g3], %g1 */
+#else
+	0xc2004003, /* 32-bit      --> ld [%g1 + %g3], %g1 */
+#endif
+	0x81c04000, /*             --> jmp %g1 */
+	0x01000000, /*             --> nop  */
+#ifdef __arch64__
+	0x9de3bf80, /* 64-bit      --> save  %sp, -128, %sp */
+#else
+	0x9de3bfc0, /* 32-bit      --> save  %sp, -64, %sp */
+#endif
+	0xa0100003, /*             --> mov  %g3, %l0 */
+	0x40000000, /*             --> call _glapi_get_dispatch */
+	0x01000000, /*             -->  nop */
+	0x82100008, /*             --> mov %o0, %g1 */
+	0x86100010, /*             --> mov %l0, %g3 */
+	0x10bffff7, /*             --> ba -4*9 */
+	0x81e80000, /*             -->  restore  */
+#endif
+    };
+#ifdef GLX_USE_TLS
+    extern unsigned long __glapi_sparc_get_dispatch(void);
+    unsigned int *code = &__glapi_sparc_tls_stub;
+    unsigned long dispatch = __glapi_sparc_get_dispatch();
+#else
+    unsigned int *code = &__glapi_sparc_pthread_stub;
+    unsigned long dispatch = (unsigned long) &_glapi_Dispatch;
+    unsigned long call_dest = (unsigned long ) &_glapi_get_dispatch;
+    int idx;
+#endif
+
+#if defined(GLX_USE_TLS)
+    code[0] = template[0] | (dispatch >> 10);
+    code[1] = template[1];
+    __glapi_sparc_icache_flush(&code[0]);
+    code[2] = template[2] | (dispatch & 0x3ff);
+    code[3] = template[3];
+    __glapi_sparc_icache_flush(&code[2]);
+    code[4] = template[4];
+    code[5] = template[5];
+    __glapi_sparc_icache_flush(&code[4]);
+    code[6] = template[6];
+    __glapi_sparc_icache_flush(&code[6]);
+#else
+#if defined(__arch64__)
+    code[0] = template[0] | (dispatch >> (32 + 10));
+    code[1] = template[1] | ((dispatch & 0xffffffff) >> 10);
+    __glapi_sparc_icache_flush(&code[0]);
+    code[2] = template[2] | ((dispatch >> 32) & 0x3ff);
+    code[3] = template[3];
+    __glapi_sparc_icache_flush(&code[2]);
+    code[4] = template[4];
+    code[5] = template[5];
+    __glapi_sparc_icache_flush(&code[4]);
+    code[6] = template[6] | (dispatch & 0x3ff);
+    idx = 7;
+#else
+    code[0] = template[0] | (dispatch >> 10);
+    code[1] = template[1];
+    __glapi_sparc_icache_flush(&code[0]);
+    code[2] = template[2] | (dispatch & 0x3ff);
+    idx = 3;
+#endif
+    code[idx + 0] = template[idx + 0];
+    __glapi_sparc_icache_flush(&code[idx - 1]);
+    code[idx + 1] = template[idx + 1];
+    code[idx + 2] = template[idx + 2];
+    __glapi_sparc_icache_flush(&code[idx + 1]);
+    code[idx + 3] = template[idx + 3];
+    code[idx + 4] = template[idx + 4];
+    __glapi_sparc_icache_flush(&code[idx + 3]);
+    code[idx + 5] = template[idx + 5];
+    code[idx + 6] = template[idx + 6];
+    __glapi_sparc_icache_flush(&code[idx + 5]);
+    code[idx + 7] = template[idx + 7];
+    code[idx + 8] = template[idx + 8] |
+	    (((call_dest - ((unsigned long) &code[idx + 8]))
+	      >> 2) & 0x3fffffff);
+    __glapi_sparc_icache_flush(&code[idx + 7]);
+    code[idx + 9] = template[idx + 9];
+    code[idx + 10] = template[idx + 10];
+    __glapi_sparc_icache_flush(&code[idx + 9]);
+    code[idx + 11] = template[idx + 11];
+    code[idx + 12] = template[idx + 12];
+    __glapi_sparc_icache_flush(&code[idx + 11]);
+    code[idx + 13] = template[idx + 13];
+    __glapi_sparc_icache_flush(&code[idx + 13]);
+#endif
+#endif
 }
 #endif /* defined(PTHREADS) || defined(GLX_USE_TLS) */
 
diff --git a/src/mesa/glapi/glapi_getproc.c b/src/mesa/glapi/glapi_getproc.c
index 4adc1aece0..764e2ef98e 100644
--- a/src/mesa/glapi/glapi_getproc.c
+++ b/src/mesa/glapi/glapi_getproc.c
@@ -259,53 +259,30 @@ generate_entrypoint(GLuint functionOffset)
    }
 
    return (_glapi_proc) code;
-#elif defined(USE_SPARC_ASM)
-
-#ifdef __arch64__
-   static const unsigned int insn_template[] = {
-	   0x05000000,	/* sethi	%uhi(_glapi_Dispatch), %g2	*/
-	   0x03000000,	/* sethi	%hi(_glapi_Dispatch), %g1	*/
-	   0x8410a000,	/* or		%g2, %ulo(_glapi_Dispatch), %g2	*/
-	   0x82106000,	/* or		%g1, %lo(_glapi_Dispatch), %g1	*/
-	   0x8528b020,	/* sllx		%g2, 32, %g2			*/
-	   0xc2584002,	/* ldx		[%g1 + %g2], %g1		*/
-	   0x05000000,	/* sethi	%hi(8 * glapioffset), %g2	*/
-	   0x8410a000,	/* or		%g2, %lo(8 * glapioffset), %g2	*/
-	   0xc6584002,	/* ldx		[%g1 + %g2], %g3		*/
-	   0x81c0c000,	/* jmpl		%g3, %g0			*/
-	   0x01000000	/*  nop						*/
+#elif defined(USE_SPARC_ASM) && (defined(PTHREADS) || defined(GLX_USE_TLS))
+   static const unsigned int template[] = {
+      0x07000000, /* sethi %hi(0), %g3 */
+      0x8210000f, /* mov  %o7, %g1 */
+      0x40000000, /* call */
+      0x9e100001, /* mov  %g1, %o7 */
    };
+#ifdef GLX_USE_TLS
+   extern unsigned int __glapi_sparc_tls_stub;
+   unsigned long call_dest = (unsigned long ) &__glapi_sparc_tls_stub;
 #else
-   static const unsigned int insn_template[] = {
-	   0x03000000,	/* sethi	%hi(_glapi_Dispatch), %g1	  */
-	   0xc2006000,	/* ld		[%g1 + %lo(_glapi_Dispatch)], %g1 */
-	   0xc6006000,	/* ld		[%g1 + %lo(4*glapioffset)], %g3	  */
-	   0x81c0c000,	/* jmpl		%g3, %g0			  */
-	   0x01000000	/*  nop						  */
-   };
-#endif /* __arch64__ */
-   unsigned int *code = (unsigned int *) malloc(sizeof(insn_template));
-   unsigned long glapi_addr = (unsigned long) &_glapi_Dispatch;
+   extern unsigned int __glapi_sparc_pthread_stub;
+   unsigned long call_dest = (unsigned long ) &__glapi_sparc_pthread_stub;
+#endif
+   unsigned int *code = (unsigned int *) malloc(sizeof(template));
    if (code) {
-      memcpy(code, insn_template, sizeof(insn_template));
-
-#ifdef __arch64__
-      code[0] |= (glapi_addr >> (32 + 10));
-      code[1] |= ((glapi_addr & 0xffffffff) >> 10);
+      code[0] = template[0] | (functionOffset & 0x3fffff);
+      code[1] = template[1];
       __glapi_sparc_icache_flush(&code[0]);
-      code[2] |= ((glapi_addr >> 32) & ((1 << 10) - 1));
-      code[3] |= (glapi_addr & ((1 << 10) - 1));
+      code[2] = template[2] |
+         (((call_dest - ((unsigned long) &code[2]))
+	   >> 2) & 0x3fffffff);
+      code[3] = template[3];
       __glapi_sparc_icache_flush(&code[2]);
-      code[6] |= ((functionOffset * 8) >> 10);
-      code[7] |= ((functionOffset * 8) & ((1 << 10) - 1));
-      __glapi_sparc_icache_flush(&code[6]);
-#else
-      code[0] |= (glapi_addr >> 10);
-      code[1] |= (glapi_addr & ((1 << 10) - 1));
-      __glapi_sparc_icache_flush(&code[0]);
-      code[2] |= (functionOffset * 4);
-      __glapi_sparc_icache_flush(&code[2]);
-#endif /* __arch64__ */
    }
    return (_glapi_proc) code;
 #else
@@ -337,21 +314,10 @@ fill_in_entrypoint_offset(_glapi_proc entrypoint, GLuint offset)
 #endif
 
 #elif defined(USE_SPARC_ASM)
-
-   /* XXX this hasn't been tested! */
    unsigned int *code = (unsigned int *) entrypoint;
-#ifdef __arch64__
-   code[6] = 0x05000000;  /* sethi	%hi(8 * glapioffset), %g2	*/
-   code[7] = 0x8410a000;  /* or		%g2, %lo(8 * glapioffset), %g2	*/
-   code[6] |= ((offset * 8) >> 10);
-   code[7] |= ((offset * 8) & ((1 << 10) - 1));
-   __glapi_sparc_icache_flush(&code[6]);
-#else /* __arch64__ */
-   code[2] = 0xc6006000;  /* ld		[%g1 + %lo(4*glapioffset)], %g3	  */
-   code[2] |= (offset * 4);
-   __glapi_sparc_icache_flush(&code[2]);
-#endif /* __arch64__ */
-
+   code[0] &= ~0x3fffff;
+   code[0] |= (offset * sizeof(void *)) & 0x3fffff;
+   __glapi_sparc_icache_flush(&code[0]);
 #else
 
    /* an unimplemented architecture */