Mesa (gallium-0.1): tgsi: Implement OPCODE_SSG/SGN.

[Michał Król]

Module: Mesa
Branch: gallium-0.1
Commit: 1347439a87a26f261ab07c914ea4e69965703ee2
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=1347439a87a26f261ab07c914ea4e69965703ee2

Author: Michal Krol <michal at tungstengraphics.com>
Date:   Wed Nov 26 13:54:28 2008 +0100

tgsi: Implement OPCODE_SSG/SGN.

---

 src/gallium/auxiliary/tgsi/tgsi_exec.c |   18 +++++++++++++++++-
 src/gallium/auxiliary/tgsi/tgsi_sse2.c |   30 +++++++++++++++++++++++++++++-
 2 files changed, 46 insertions(+), 2 deletions(-)

diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c
index 7114119..65a0f39 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -840,6 +840,17 @@ micro_rnd(
 }
 
 static void
+micro_sgn(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] = src->f[0] < 0.0f ? -1.0f : src->f[0] > 0.0f ? 1.0f : 0.0f;
+   dst->f[1] = src->f[1] < 0.0f ? -1.0f : src->f[1] > 0.0f ? 1.0f : 0.0f;
+   dst->f[2] = src->f[2] < 0.0f ? -1.0f : src->f[2] > 0.0f ? 1.0f : 0.0f;
+   dst->f[3] = src->f[3] < 0.0f ? -1.0f : src->f[3] > 0.0f ? 1.0f : 0.0f;
+}
+
+static void
 micro_shl(
    union tgsi_exec_channel *dst,
    const union tgsi_exec_channel *src0,
@@ -2423,7 +2434,12 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_SSG:
-      assert (0);
+   /* TGSI_OPCODE_SGN */
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         micro_sgn( &r[0], &r[0] );
+         STORE( &r[0], 0, chan_index );
+      }
       break;
 
    case TGSI_OPCODE_CMP:
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sse2.c b/src/gallium/auxiliary/tgsi/tgsi_sse2.c
index 29442b4..499d865 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_sse2.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_sse2.c
@@ -902,6 +902,29 @@ emit_setsign(
 }
 
 static void PIPE_CDECL
+sgn4f(
+   float *store )
+{
+   store[0] = store[0] < 0.0f ? -1.0f : store[0] > 0.0f ? 1.0f : 0.0f;
+   store[1] = store[1] < 0.0f ? -1.0f : store[1] > 0.0f ? 1.0f : 0.0f;
+   store[2] = store[2] < 0.0f ? -1.0f : store[2] > 0.0f ? 1.0f : 0.0f;
+   store[3] = store[3] < 0.0f ? -1.0f : store[3] > 0.0f ? 1.0f : 0.0f;
+}
+
+static void
+emit_sgn(
+   struct x86_function *func,
+   unsigned xmm_save, 
+   unsigned xmm_dst )
+{
+   emit_func_call_dst(
+      func,
+      xmm_save,
+      xmm_dst,
+      sgn4f );
+}
+
+static void PIPE_CDECL
 sin4f(
    float *store )
 {
@@ -1904,7 +1927,12 @@ emit_instruction(
       break;
 
    case TGSI_OPCODE_SSG:
-      return 0;
+   /* TGSI_OPCODE_SGN */
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         emit_sgn( func, 0, 0 );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
       break;
 
    case TGSI_OPCODE_CMP: