gallium: use new util_fast_exp2(), _log2(), pow() functions

New code surrounded with #if FAST_MATH to allow comparing against original
code if we need to debug.

2 files changed, 49 insertions, 0 deletions

diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c
index 88a34a6961..e28b56c842 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -57,6 +57,9 @@
 #include "tgsi/tgsi_parse.h"
 #include "tgsi/tgsi_util.h"
 #include "tgsi_exec.h"
+#include "util/u_math.h"
+
+#define FAST_MATH 1
 
 #define TILE_TOP_LEFT     0
 #define TILE_TOP_RIGHT    1
@@ -145,6 +148,8 @@ tgsi_exec_machine_bind_shader(
    tgsi_dump(tokens, 0);
 #endif
 
+   util_init_math();
+
    mach->Tokens = tokens;
    mach->Samplers = samplers;
 
@@ -448,10 +453,17 @@ micro_exp2(
    union tgsi_exec_channel *dst,
    const union tgsi_exec_channel *src)
 {
+#if FAST_MATH
+   dst->f[0] = util_fast_exp2( src->f[0] );
+   dst->f[1] = util_fast_exp2( src->f[1] );
+   dst->f[2] = util_fast_exp2( src->f[2] );
+   dst->f[3] = util_fast_exp2( src->f[3] );
+#else
    dst->f[0] = powf( 2.0f, src->f[0] );
    dst->f[1] = powf( 2.0f, src->f[1] );
    dst->f[2] = powf( 2.0f, src->f[2] );
    dst->f[3] = powf( 2.0f, src->f[3] );
+#endif
 }
 
 static void
@@ -528,10 +540,17 @@ micro_lg2(
    union tgsi_exec_channel *dst,
    const union tgsi_exec_channel *src )
 {
+#if FAST_MATH
+   dst->f[0] = util_fast_log2( src->f[0] );
+   dst->f[1] = util_fast_log2( src->f[1] );
+   dst->f[2] = util_fast_log2( src->f[2] );
+   dst->f[3] = util_fast_log2( src->f[3] );
+#else
    dst->f[0] = logf( src->f[0] ) * 1.442695f;
    dst->f[1] = logf( src->f[1] ) * 1.442695f;
    dst->f[2] = logf( src->f[2] ) * 1.442695f;
    dst->f[3] = logf( src->f[3] ) * 1.442695f;
+#endif
 }
 
 static void
@@ -796,10 +815,17 @@ micro_pow(
    const union tgsi_exec_channel *src0,
    const union tgsi_exec_channel *src1 )
 {
+#if FAST_MATH
+   dst->f[0] = util_fast_pow( src0->f[0], src1->f[0] );
+   dst->f[1] = util_fast_pow( src0->f[1], src1->f[1] );
+   dst->f[2] = util_fast_pow( src0->f[2], src1->f[2] );
+   dst->f[3] = util_fast_pow( src0->f[3], src1->f[3] );
+#else
    dst->f[0] = powf( src0->f[0], src1->f[0] );
    dst->f[1] = powf( src0->f[1], src1->f[1] );
    dst->f[2] = powf( src0->f[2], src1->f[2] );
    dst->f[3] = powf( src0->f[3], src1->f[3] );
+#endif
 }
 
 static void
@@ -2024,7 +2050,11 @@ exec_instruction(
     /* TGSI_OPCODE_EX2 */
       FETCH(&r[0], 0, CHAN_X);
 
+#if FAST_MATH
+      micro_exp2( &r[0], &r[0] );
+#else
       micro_pow( &r[0], &mach->Temps[TEMP_2_I].xyzw[TEMP_2_C], &r[0] );
+#endif
 
       FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
 	 STORE( &r[0], 0, chan_index );
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sse2.c b/src/gallium/auxiliary/tgsi/tgsi_sse2.c
index 485e5a0e6f..e390607023 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_sse2.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_sse2.c
@@ -27,6 +27,7 @@
 
 #include "pipe/p_util.h"
 #include "pipe/p_shader_tokens.h"
+#include "util/u_math.h"
 #include "tgsi/tgsi_parse.h"
 #include "tgsi/tgsi_util.h"
 #include "tgsi_exec.h"
@@ -42,6 +43,8 @@
  */
 #define HIGH_PRECISION 1
 
+#define FAST_MATH 1
+
 
 #define FOR_EACH_CHANNEL( CHAN )\
    for (CHAN = 0; CHAN < NUM_CHANNELS; CHAN++)
@@ -623,10 +626,17 @@ ex24f(
 {
    const unsigned X = 0;
 
+#if FAST_MATH
+   store[X + 0] = util_fast_exp2( store[X + 0] );
+   store[X + 1] = util_fast_exp2( store[X + 1] );
+   store[X + 2] = util_fast_exp2( store[X + 2] );
+   store[X + 3] = util_fast_exp2( store[X + 3] );
+#else
    store[X + 0] = powf( 2.0f, store[X + 0] );
    store[X + 1] = powf( 2.0f, store[X + 1] );
    store[X + 2] = powf( 2.0f, store[X + 2] );
    store[X + 3] = powf( 2.0f, store[X + 3] );
+#endif
 }
 
 static void
@@ -762,10 +772,17 @@ pow4f(
 {
    const unsigned X = 0;
 
+#if FAST_MATH
+   store[X + 0] = util_fast_pow( store[X + 0], store[X + 4] );
+   store[X + 1] = util_fast_pow( store[X + 1], store[X + 5] );
+   store[X + 2] = util_fast_pow( store[X + 2], store[X + 6] );
+   store[X + 3] = util_fast_pow( store[X + 3], store[X + 7] );
+#else
    store[X + 0] = powf( store[X + 0], store[X + 4] );
    store[X + 1] = powf( store[X + 1], store[X + 5] );
    store[X + 2] = powf( store[X + 2], store[X + 6] );
    store[X + 3] = powf( store[X + 3], store[X + 7] );
+#endif
 }
 
 static void
@@ -2235,6 +2252,8 @@ tgsi_emit_sse2(
    unsigned ok = 1;
    uint num_immediates = 0;
 
+   util_init_math();
+
    func->csr = func->store;
 
    tgsi_parse_init( &parse, tokens );