summaryrefslogtreecommitdiff
path: root/src/gallium/auxiliary/tgsi/tgsi_sse2.c
diff options
context:
space:
mode:
Diffstat (limited to 'src/gallium/auxiliary/tgsi/tgsi_sse2.c')
-rw-r--r--src/gallium/auxiliary/tgsi/tgsi_sse2.c369
1 files changed, 293 insertions, 76 deletions
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sse2.c b/src/gallium/auxiliary/tgsi/tgsi_sse2.c
index ff869c8312..cac44af7f4 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_sse2.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_sse2.c
@@ -25,9 +25,16 @@
*
**************************************************************************/
+#include "pipe/p_config.h"
+
+#if defined(PIPE_ARCH_X86)
+
#include "pipe/p_debug.h"
#include "pipe/p_shader_tokens.h"
#include "util/u_math.h"
+#if defined(PIPE_ARCH_SSE)
+#include "util/u_sse.h"
+#endif
#include "tgsi/tgsi_parse.h"
#include "tgsi/tgsi_util.h"
#include "tgsi_exec.h"
@@ -35,8 +42,6 @@
#include "rtasm/rtasm_x86sse.h"
-#ifdef PIPE_ARCH_X86
-
/* for 1/sqrt()
*
* This costs about 100fps (close to 10%) in gears:
@@ -509,10 +514,31 @@ emit_coef_dady(
* Function call helpers.
*/
+/**
+ * NOTE: In gcc, if the destination uses the SSE intrinsics, then it must be
+ * defined with __attribute__((force_align_arg_pointer)), as we do not guarantee
+ * that the stack pointer is 16 byte aligned, as expected.
+ */
static void
-emit_push_gp(
- struct x86_function *func )
+emit_func_call_dst(
+ struct x86_function *func,
+ unsigned xmm_save,
+ unsigned xmm_dst,
+ void (PIPE_CDECL *code)() )
{
+ struct x86_reg ecx = x86_make_reg( file_REG32, reg_CX );
+ unsigned i, n;
+ unsigned xmm_mask;
+
+ /* Bitmask of the xmm registers to save */
+ xmm_mask = (1 << xmm_save) - 1;
+ xmm_mask &= ~(1 << xmm_dst);
+
+ sse_movaps(
+ func,
+ get_temp( TEMP_R0, 0 ),
+ make_xmm( xmm_dst ) );
+
x86_push(
func,
x86_make_reg( file_REG32, reg_AX) );
@@ -522,12 +548,49 @@ emit_push_gp(
x86_push(
func,
x86_make_reg( file_REG32, reg_DX) );
-}
+
+ for(i = 0, n = 0; i < 8; ++i)
+ if(xmm_mask & (1 << i))
+ ++n;
+
+ x86_sub_imm(
+ func,
+ x86_make_reg( file_REG32, reg_SP ),
+ n*16);
+
+ for(i = 0, n = 0; i < 8; ++i)
+ if(xmm_mask & (1 << i)) {
+ sse_movups(
+ func,
+ x86_make_disp( x86_make_reg( file_REG32, reg_SP ), n*16 ),
+ make_xmm( i ) );
+ ++n;
+ }
+
+ x86_lea(
+ func,
+ ecx,
+ get_temp( TEMP_R0, 0 ) );
+
+ x86_push( func, ecx );
+ x86_mov_reg_imm( func, ecx, (unsigned long) code );
+ x86_call( func, ecx );
+ x86_pop(func, ecx );
+
+ for(i = 0, n = 0; i < 8; ++i)
+ if(xmm_mask & (1 << i)) {
+ sse_movups(
+ func,
+ make_xmm( i ),
+ x86_make_disp( x86_make_reg( file_REG32, reg_SP ), n*16 ) );
+ ++n;
+ }
+
+ x86_add_imm(
+ func,
+ x86_make_reg( file_REG32, reg_SP ),
+ n*16);
-static void
-x86_pop_gp(
- struct x86_function *func )
-{
/* Restore GP registers in a reverse order.
*/
x86_pop(
@@ -539,39 +602,6 @@ x86_pop_gp(
x86_pop(
func,
x86_make_reg( file_REG32, reg_AX) );
-}
-
-static void
-emit_func_call_dst(
- struct x86_function *func,
- unsigned xmm_dst,
- void (PIPE_CDECL *code)() )
-{
- sse_movaps(
- func,
- get_temp( TEMP_R0, 0 ),
- make_xmm( xmm_dst ) );
-
- emit_push_gp(
- func );
-
- {
- struct x86_reg ecx = x86_make_reg( file_REG32, reg_CX );
-
- x86_lea(
- func,
- ecx,
- get_temp( TEMP_R0, 0 ) );
-
- x86_push( func, ecx );
- x86_mov_reg_imm( func, ecx, (unsigned long) code );
- x86_call( func, ecx );
- x86_pop(func, ecx );
- }
-
-
- x86_pop_gp(
- func );
sse_movaps(
func,
@@ -582,6 +612,7 @@ emit_func_call_dst(
static void
emit_func_call_dst_src(
struct x86_function *func,
+ unsigned xmm_save,
unsigned xmm_dst,
unsigned xmm_src,
void (PIPE_CDECL *code)() )
@@ -593,10 +624,119 @@ emit_func_call_dst_src(
emit_func_call_dst(
func,
+ xmm_save,
xmm_dst,
code );
}
+
+#if defined(PIPE_ARCH_SSE)
+
+/*
+ * Fast SSE2 implementation of special math functions.
+ */
+
+#define POLY0(x, c0) _mm_set1_ps(c0)
+#define POLY1(x, c0, c1) _mm_add_ps(_mm_mul_ps(POLY0(x, c1), x), _mm_set1_ps(c0))
+#define POLY2(x, c0, c1, c2) _mm_add_ps(_mm_mul_ps(POLY1(x, c1, c2), x), _mm_set1_ps(c0))
+#define POLY3(x, c0, c1, c2, c3) _mm_add_ps(_mm_mul_ps(POLY2(x, c1, c2, c3), x), _mm_set1_ps(c0))
+#define POLY4(x, c0, c1, c2, c3, c4) _mm_add_ps(_mm_mul_ps(POLY3(x, c1, c2, c3, c4), x), _mm_set1_ps(c0))
+#define POLY5(x, c0, c1, c2, c3, c4, c5) _mm_add_ps(_mm_mul_ps(POLY4(x, c1, c2, c3, c4, c5), x), _mm_set1_ps(c0))
+
+#define EXP_POLY_DEGREE 3
+#define LOG_POLY_DEGREE 5
+
+/**
+ * See http://www.devmaster.net/forums/showthread.php?p=43580
+ */
+static INLINE __m128
+exp2f4(__m128 x)
+{
+ __m128i ipart;
+ __m128 fpart, expipart, expfpart;
+
+ x = _mm_min_ps(x, _mm_set1_ps( 129.00000f));
+ x = _mm_max_ps(x, _mm_set1_ps(-126.99999f));
+
+ /* ipart = int(x - 0.5) */
+ ipart = _mm_cvtps_epi32(_mm_sub_ps(x, _mm_set1_ps(0.5f)));
+
+ /* fpart = x - ipart */
+ fpart = _mm_sub_ps(x, _mm_cvtepi32_ps(ipart));
+
+ /* expipart = (float) (1 << ipart) */
+ expipart = _mm_castsi128_ps(_mm_slli_epi32(_mm_add_epi32(ipart, _mm_set1_epi32(127)), 23));
+
+ /* minimax polynomial fit of 2**x, in range [-0.5, 0.5[ */
+#if EXP_POLY_DEGREE == 5
+ expfpart = POLY5(fpart, 9.9999994e-1f, 6.9315308e-1f, 2.4015361e-1f, 5.5826318e-2f, 8.9893397e-3f, 1.8775767e-3f);
+#elif EXP_POLY_DEGREE == 4
+ expfpart = POLY4(fpart, 1.0000026f, 6.9300383e-1f, 2.4144275e-1f, 5.2011464e-2f, 1.3534167e-2f);
+#elif EXP_POLY_DEGREE == 3
+ expfpart = POLY3(fpart, 9.9992520e-1f, 6.9583356e-1f, 2.2606716e-1f, 7.8024521e-2f);
+#elif EXP_POLY_DEGREE == 2
+ expfpart = POLY2(fpart, 1.0017247f, 6.5763628e-1f, 3.3718944e-1f);
+#else
+#error
+#endif
+
+ return _mm_mul_ps(expipart, expfpart);
+}
+
+
+/**
+ * See http://www.devmaster.net/forums/showthread.php?p=43580
+ */
+static INLINE __m128
+log2f4(__m128 x)
+{
+ __m128i expmask = _mm_set1_epi32(0x7f800000);
+ __m128i mantmask = _mm_set1_epi32(0x007fffff);
+ __m128 one = _mm_set1_ps(1.0f);
+
+ __m128i i = _mm_castps_si128(x);
+
+ /* exp = (float) exponent(x) */
+ __m128 exp = _mm_cvtepi32_ps(_mm_sub_epi32(_mm_srli_epi32(_mm_and_si128(i, expmask), 23), _mm_set1_epi32(127)));
+
+ /* mant = (float) mantissa(x) */
+ __m128 mant = _mm_or_ps(_mm_castsi128_ps(_mm_and_si128(i, mantmask)), one);
+
+ __m128 logmant;
+
+ /* Minimax polynomial fit of log2(x)/(x - 1), for x in range [1, 2[
+ * These coefficients can be generate with
+ * http://www.boost.org/doc/libs/1_36_0/libs/math/doc/sf_and_dist/html/math_toolkit/toolkit/internals2/minimax.html
+ */
+#if LOG_POLY_DEGREE == 6
+ logmant = POLY5(mant, 3.11578814719469302614f, -3.32419399085241980044f, 2.59883907202499966007f, -1.23152682416275988241f, 0.318212422185251071475f, -0.0344359067839062357313f);
+#elif LOG_POLY_DEGREE == 5
+ logmant = POLY4(mant, 2.8882704548164776201f, -2.52074962577807006663f, 1.48116647521213171641f, -0.465725644288844778798f, 0.0596515482674574969533f);
+#elif LOG_POLY_DEGREE == 4
+ logmant = POLY3(mant, 2.61761038894603480148f, -1.75647175389045657003f, 0.688243882994381274313f, -0.107254423828329604454f);
+#elif LOG_POLY_DEGREE == 3
+ logmant = POLY2(mant, 2.28330284476918490682f, -1.04913055217340124191f, 0.204446009836232697516f);
+#else
+#error
+#endif
+
+ /* This effectively increases the polynomial degree by one, but ensures that log2(1) == 0*/
+ logmant = _mm_mul_ps(logmant, _mm_sub_ps(mant, one));
+
+ return _mm_add_ps(logmant, exp);
+}
+
+
+static INLINE __m128
+powf4(__m128 x, __m128 y)
+{
+ return exp2f4(_mm_mul_ps(log2f4(x), y));
+}
+
+#endif /* PIPE_ARCH_SSE */
+
+
+
/**
* Low-level instruction translators.
*/
@@ -639,38 +779,42 @@ cos4f(
static void
emit_cos(
struct x86_function *func,
+ unsigned xmm_save,
unsigned xmm_dst )
{
emit_func_call_dst(
func,
+ xmm_save,
xmm_dst,
cos4f );
}
static void PIPE_CDECL
+#if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_SSE)
+__attribute__((force_align_arg_pointer))
+#endif
ex24f(
float *store )
{
-#if FAST_MATH
+#if defined(PIPE_ARCH_SSE)
+ _mm_store_ps(&store[0], exp2f4( _mm_load_ps(&store[0]) ));
+#else
store[0] = util_fast_exp2( store[0] );
store[1] = util_fast_exp2( store[1] );
store[2] = util_fast_exp2( store[2] );
store[3] = util_fast_exp2( store[3] );
-#else
- store[0] = powf( 2.0f, store[0] );
- store[1] = powf( 2.0f, store[1] );
- store[2] = powf( 2.0f, store[2] );
- store[3] = powf( 2.0f, store[3] );
#endif
}
static void
emit_ex2(
struct x86_function *func,
+ unsigned xmm_save,
unsigned xmm_dst )
{
emit_func_call_dst(
func,
+ xmm_save,
xmm_dst,
ex24f );
}
@@ -710,10 +854,12 @@ flr4f(
static void
emit_flr(
struct x86_function *func,
+ unsigned xmm_save,
unsigned xmm_dst )
{
emit_func_call_dst(
func,
+ xmm_save,
xmm_dst,
flr4f );
}
@@ -731,31 +877,42 @@ frc4f(
static void
emit_frc(
struct x86_function *func,
+ unsigned xmm_save,
unsigned xmm_dst )
{
emit_func_call_dst(
func,
+ xmm_save,
xmm_dst,
frc4f );
}
static void PIPE_CDECL
+#if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_SSE)
+__attribute__((force_align_arg_pointer))
+#endif
lg24f(
float *store )
{
+#if defined(PIPE_ARCH_SSE)
+ _mm_store_ps(&store[0], log2f4( _mm_load_ps(&store[0]) ));
+#else
store[0] = util_fast_log2( store[0] );
store[1] = util_fast_log2( store[1] );
store[2] = util_fast_log2( store[2] );
store[3] = util_fast_log2( store[3] );
+#endif
}
static void
emit_lg2(
struct x86_function *func,
+ unsigned xmm_save,
unsigned xmm_dst )
{
emit_func_call_dst(
func,
+ xmm_save,
xmm_dst,
lg24f );
}
@@ -797,30 +954,32 @@ emit_neg(
}
static void PIPE_CDECL
+#if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_SSE)
+__attribute__((force_align_arg_pointer))
+#endif
pow4f(
float *store )
{
-#if FAST_MATH
+#if defined(PIPE_ARCH_SSE)
+ _mm_store_ps(&store[0], powf4( _mm_load_ps(&store[0]), _mm_load_ps(&store[4]) ));
+#else
store[0] = util_fast_pow( store[0], store[4] );
store[1] = util_fast_pow( store[1], store[5] );
store[2] = util_fast_pow( store[2], store[6] );
store[3] = util_fast_pow( store[3], store[7] );
-#else
- store[0] = powf( store[0], store[4] );
- store[1] = powf( store[1], store[5] );
- store[2] = powf( store[2], store[6] );
- store[3] = powf( store[3], store[7] );
#endif
}
static void
emit_pow(
struct x86_function *func,
+ unsigned xmm_save,
unsigned xmm_dst,
unsigned xmm_src )
{
emit_func_call_dst_src(
func,
+ xmm_save,
xmm_dst,
xmm_src,
pow4f );
@@ -855,10 +1014,12 @@ rnd4f(
static void
emit_rnd(
struct x86_function *func,
+ unsigned xmm_save,
unsigned xmm_dst )
{
emit_func_call_dst(
func,
+ xmm_save,
xmm_dst,
rnd4f );
}
@@ -935,10 +1096,12 @@ sgn4f(
static void
emit_sgn(
struct x86_function *func,
+ unsigned xmm_save,
unsigned xmm_dst )
{
emit_func_call_dst(
func,
+ xmm_save,
xmm_dst,
sgn4f );
}
@@ -955,10 +1118,12 @@ sin4f(
static void
emit_sin (struct x86_function *func,
+ unsigned xmm_save,
unsigned xmm_dst)
{
emit_func_call_dst(
func,
+ xmm_save,
xmm_dst,
sin4f );
}
@@ -1378,7 +1543,7 @@ emit_instruction(
get_temp(
TGSI_EXEC_TEMP_MINUS_128_I,
TGSI_EXEC_TEMP_MINUS_128_C ) );
- emit_pow( func, 1, 2 );
+ emit_pow( func, 3, 1, 2 );
FETCH( func, *inst, 0, 0, CHAN_X );
sse_xorps(
func,
@@ -1424,11 +1589,11 @@ emit_instruction(
if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) ||
IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y )) {
emit_MOV( func, 1, 0 );
- emit_flr( func, 1 );
+ emit_flr( func, 2, 1 );
/* dst.x = ex2(floor(src.x)) */
if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X )) {
emit_MOV( func, 2, 1 );
- emit_ex2( func, 2 );
+ emit_ex2( func, 3, 2 );
STORE( func, *inst, 2, 0, CHAN_X );
}
/* dst.y = src.x - floor(src.x) */
@@ -1440,7 +1605,7 @@ emit_instruction(
}
/* dst.z = ex2(src.x) */
if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z )) {
- emit_ex2( func, 0 );
+ emit_ex2( func, 3, 0 );
STORE( func, *inst, 0, 0, CHAN_Z );
}
}
@@ -1458,21 +1623,21 @@ emit_instruction(
FETCH( func, *inst, 0, 0, CHAN_X );
emit_abs( func, 0 );
emit_MOV( func, 1, 0 );
- emit_lg2( func, 1 );
+ emit_lg2( func, 2, 1 );
/* dst.z = lg2(abs(src.x)) */
if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z )) {
STORE( func, *inst, 1, 0, CHAN_Z );
}
if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) ||
IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y )) {
- emit_flr( func, 1 );
+ emit_flr( func, 2, 1 );
/* dst.x = floor(lg2(abs(src.x))) */
if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X )) {
STORE( func, *inst, 1, 0, CHAN_X );
}
/* dst.x = abs(src)/ex2(floor(lg2(abs(src.x)))) */
if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y )) {
- emit_ex2( func, 1 );
+ emit_ex2( func, 2, 1 );
emit_rcp( func, 1, 1 );
emit_mul( func, 0, 1 );
STORE( func, *inst, 0, 0, CHAN_Y );
@@ -1647,7 +1812,18 @@ emit_instruction(
case TGSI_OPCODE_DOT2ADD:
/* TGSI_OPCODE_DP2A */
- return 0;
+ FETCH( func, *inst, 0, 0, CHAN_X ); /* xmm0 = src[0].x */
+ FETCH( func, *inst, 1, 1, CHAN_X ); /* xmm1 = src[1].x */
+ emit_mul( func, 0, 1 ); /* xmm0 = xmm0 * xmm1 */
+ FETCH( func, *inst, 1, 0, CHAN_Y ); /* xmm1 = src[0].y */
+ FETCH( func, *inst, 2, 1, CHAN_Y ); /* xmm2 = src[1].y */
+ emit_mul( func, 1, 2 ); /* xmm1 = xmm1 * xmm2 */
+ emit_add( func, 0, 1 ); /* xmm0 = xmm0 + xmm1 */
+ FETCH( func, *inst, 1, 2, CHAN_X ); /* xmm1 = src[2].x */
+ emit_add( func, 0, 1 ); /* xmm0 = xmm0 + xmm1 */
+ FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+ STORE( func, *inst, 0, 0, chan_index ); /* dest[ch] = xmm0 */
+ }
break;
case TGSI_OPCODE_INDEX:
@@ -1662,7 +1838,7 @@ emit_instruction(
/* TGSI_OPCODE_FRC */
FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
FETCH( func, *inst, 0, 0, chan_index );
- emit_frc( func, 0 );
+ emit_frc( func, 0, 0 );
STORE( func, *inst, 0, 0, chan_index );
}
break;
@@ -1675,7 +1851,7 @@ emit_instruction(
/* TGSI_OPCODE_FLR */
FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
FETCH( func, *inst, 0, 0, chan_index );
- emit_flr( func, 0 );
+ emit_flr( func, 0, 0 );
STORE( func, *inst, 0, 0, chan_index );
}
break;
@@ -1683,7 +1859,7 @@ emit_instruction(
case TGSI_OPCODE_ROUND:
FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
FETCH( func, *inst, 0, 0, chan_index );
- emit_rnd( func, 0 );
+ emit_rnd( func, 0, 0 );
STORE( func, *inst, 0, 0, chan_index );
}
break;
@@ -1691,7 +1867,7 @@ emit_instruction(
case TGSI_OPCODE_EXPBASE2:
/* TGSI_OPCODE_EX2 */
FETCH( func, *inst, 0, 0, CHAN_X );
- emit_ex2( func, 0 );
+ emit_ex2( func, 0, 0 );
FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
STORE( func, *inst, 0, 0, chan_index );
}
@@ -1700,7 +1876,7 @@ emit_instruction(
case TGSI_OPCODE_LOGBASE2:
/* TGSI_OPCODE_LG2 */
FETCH( func, *inst, 0, 0, CHAN_X );
- emit_lg2( func, 0 );
+ emit_lg2( func, 0, 0 );
FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
STORE( func, *inst, 0, 0, chan_index );
}
@@ -1710,7 +1886,7 @@ emit_instruction(
/* TGSI_OPCODE_POW */
FETCH( func, *inst, 0, 0, CHAN_X );
FETCH( func, *inst, 1, 1, CHAN_X );
- emit_pow( func, 0, 1 );
+ emit_pow( func, 0, 0, 1 );
FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
STORE( func, *inst, 0, 0, chan_index );
}
@@ -1801,7 +1977,7 @@ emit_instruction(
case TGSI_OPCODE_COS:
FETCH( func, *inst, 0, 0, CHAN_X );
- emit_cos( func, 0 );
+ emit_cos( func, 0, 0 );
FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
STORE( func, *inst, 0, 0, chan_index );
}
@@ -1860,7 +2036,7 @@ emit_instruction(
case TGSI_OPCODE_SIN:
FETCH( func, *inst, 0, 0, CHAN_X );
- emit_sin( func, 0 );
+ emit_sin( func, 0, 0 );
FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
STORE( func, *inst, 0, 0, chan_index );
}
@@ -1927,7 +2103,7 @@ emit_instruction(
case TGSI_OPCODE_ARR:
FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
FETCH( func, *inst, 0, 0, chan_index );
- emit_rnd( func, 0 );
+ emit_rnd( func, 0, 0 );
emit_f2it( func, 0 );
STORE( func, *inst, 0, 0, chan_index );
}
@@ -1952,7 +2128,7 @@ emit_instruction(
/* TGSI_OPCODE_SGN */
FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
FETCH( func, *inst, 0, 0, chan_index );
- emit_sgn( func, 0 );
+ emit_sgn( func, 0, 0 );
STORE( func, *inst, 0, 0, chan_index );
}
break;
@@ -1964,12 +2140,12 @@ emit_instruction(
case TGSI_OPCODE_SCS:
IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) {
FETCH( func, *inst, 0, 0, CHAN_X );
- emit_cos( func, 0 );
+ emit_cos( func, 0, 0 );
STORE( func, *inst, 0, 0, CHAN_X );
}
IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) {
FETCH( func, *inst, 0, 0, CHAN_X );
- emit_sin( func, 0 );
+ emit_sin( func, 0, 0 );
STORE( func, *inst, 0, 0, CHAN_Y );
}
IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ) {
@@ -1995,7 +2171,39 @@ emit_instruction(
break;
case TGSI_OPCODE_NRM:
- return 0;
+ /* fall-through */
+ case TGSI_OPCODE_NRM4:
+ /* 3 or 4-component normalization */
+ {
+ uint dims = (inst->Instruction.Opcode == TGSI_OPCODE_NRM) ? 3 : 4;
+ /* note: cannot use xmm regs 2/3 here (see emit_rsqrt() above) */
+ FETCH( func, *inst, 4, 0, CHAN_X ); /* xmm4 = src[0].x */
+ FETCH( func, *inst, 5, 0, CHAN_Y ); /* xmm5 = src[0].y */
+ FETCH( func, *inst, 6, 0, CHAN_Z ); /* xmm6 = src[0].z */
+ if (dims == 4) {
+ FETCH( func, *inst, 7, 0, CHAN_W ); /* xmm7 = src[0].w */
+ }
+ emit_MOV( func, 0, 4 ); /* xmm0 = xmm3 */
+ emit_mul( func, 0, 4 ); /* xmm0 *= xmm3 */
+ emit_MOV( func, 1, 5 ); /* xmm1 = xmm4 */
+ emit_mul( func, 1, 5 ); /* xmm1 *= xmm4 */
+ emit_add( func, 0, 1 ); /* xmm0 += xmm1 */
+ emit_MOV( func, 1, 6 ); /* xmm1 = xmm5 */
+ emit_mul( func, 1, 6 ); /* xmm1 *= xmm5 */
+ emit_add( func, 0, 1 ); /* xmm0 += xmm1 */
+ if (dims == 4) {
+ emit_MOV( func, 1, 7 ); /* xmm1 = xmm7 */
+ emit_mul( func, 1, 7 ); /* xmm1 *= xmm7 */
+ emit_add( func, 0, 0 ); /* xmm0 += xmm1 */
+ }
+ emit_rsqrt( func, 1, 0 ); /* xmm1 = 1/sqrt(xmm0) */
+ FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+ if (chan_index < dims) {
+ emit_mul( func, 4+chan_index, 1); /* xmm[4+ch] *= xmm1 */
+ STORE( func, *inst, 4+chan_index, 0, chan_index );
+ }
+ }
+ }
break;
case TGSI_OPCODE_DIV:
@@ -2003,7 +2211,16 @@ emit_instruction(
break;
case TGSI_OPCODE_DP2:
- return 0;
+ FETCH( func, *inst, 0, 0, CHAN_X ); /* xmm0 = src[0].x */
+ FETCH( func, *inst, 1, 1, CHAN_X ); /* xmm1 = src[1].x */
+ emit_mul( func, 0, 1 ); /* xmm0 = xmm0 * xmm1 */
+ FETCH( func, *inst, 1, 0, CHAN_Y ); /* xmm1 = src[0].y */
+ FETCH( func, *inst, 2, 1, CHAN_Y ); /* xmm2 = src[1].y */
+ emit_mul( func, 1, 2 ); /* xmm1 = xmm1 * xmm2 */
+ emit_add( func, 0, 1 ); /* xmm0 = xmm0 + xmm1 */
+ FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+ STORE( func, *inst, 0, 0, chan_index ); /* dest[ch] = xmm0 */
+ }
break;
case TGSI_OPCODE_TXL:
generated by cgit v1.2.3 (git 2.39.1) at 2025-01-06 06:57:24 +0000