Mesa (master): translate: Respect translate_buffer::max_index.

[Jose Fonseca]

Module: Mesa
Branch: master
Commit: 17bbc1f0425b3768e26473eccea5f2570dcb26d3
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=17bbc1f0425b3768e26473eccea5f2570dcb26d3

Author: José Fonseca <jfonseca at vmware.com>
Date:   Fri Apr  1 19:19:42 2011 +0100

translate: Respect translate_buffer::max_index.

---

 src/gallium/auxiliary/rtasm/rtasm_x86sse.c      |   10 ++++++++++
 src/gallium/auxiliary/rtasm/rtasm_x86sse.h      |    1 +
 src/gallium/auxiliary/translate/translate_sse.c |   19 +++++++++++++++++--
 3 files changed, 28 insertions(+), 2 deletions(-)

diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
index b03dd3a..5231bb0 100644
--- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
+++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
@@ -726,6 +726,16 @@ void x86_movzx16(struct x86_function *p, struct x86_reg dst, struct x86_reg src
    emit_modrm(p, dst, src);
 }
 
+void x86_cmovcc( struct x86_function *p,
+                 struct x86_reg dst,
+                 struct x86_reg src,
+                 enum x86_cc cc)
+{
+   DUMP_RRI( dst, src, cc );
+   emit_2ub( p, 0x0f, 0x40 + cc );
+   emit_modrm( p, dst, src );
+}
+
 void x86_xor( struct x86_function *p,
 	      struct x86_reg dst,
 	      struct x86_reg src )
diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.h b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h
index 2b9678b..af0565b 100644
--- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.h
+++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h
@@ -309,6 +309,7 @@ void sse_movmskps( struct x86_function *p, struct x86_reg dst, struct x86_reg sr
 
 void x86_add( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void x86_and( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_cmovcc( struct x86_function *p, struct x86_reg dst, struct x86_reg src, enum x86_cc cc );
 void x86_cmp( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void x86_dec( struct x86_function *p, struct x86_reg reg );
 void x86_inc( struct x86_function *p, struct x86_reg reg );
diff --git a/src/gallium/auxiliary/translate/translate_sse.c b/src/gallium/auxiliary/translate/translate_sse.c
index ef7f4be..b00242d 100644
--- a/src/gallium/auxiliary/translate/translate_sse.c
+++ b/src/gallium/auxiliary/translate/translate_sse.c
@@ -1067,6 +1067,8 @@ static boolean init_inputs( struct translate_sse *p,
       struct translate_buffer *buffer = &p->buffer[varient->buffer_index];
 
       if (!index_size || varient->instance_divisor) {
+         struct x86_reg buf_max_index = x86_make_disp(p->machine_EDI,
+                                                     get_offset(p, &buffer->max_index));
          struct x86_reg buf_stride   = x86_make_disp(p->machine_EDI,
                                                      get_offset(p, &buffer->stride));
          struct x86_reg buf_ptr      = x86_make_disp(p->machine_EDI,
@@ -1100,14 +1102,17 @@ static boolean init_inputs( struct translate_sse *p,
             x86_mov(p->func, tmp_EAX, elt);
          }
 
-         /*
-          * TODO: Respect translate_buffer::max_index.
+         /* Clamp to max_index
           */
+         x86_cmp(p->func, tmp_EAX, buf_max_index);
+         x86_cmovcc(p->func, tmp_EAX, buf_max_index, cc_AE);
 
          x86_imul(p->func, tmp_EAX, buf_stride);
          x64_rexw(p->func);
          x86_add(p->func, tmp_EAX, buf_base_ptr);
 
+         x86_cmp(p->func, p->count_EBP, p->tmp_EAX);
+
 
          /* In the linear case, keep the buffer pointer instead of the
           * index number.
@@ -1163,6 +1168,10 @@ static struct x86_reg get_buffer_ptr( struct translate_sse *p,
          x86_make_disp(p->machine_EDI,
                        get_offset(p, &p->buffer[varient->buffer_index].base_ptr));
 
+      struct x86_reg buf_max_index =
+         x86_make_disp(p->machine_EDI,
+                       get_offset(p, &p->buffer[varient->buffer_index].max_index));
+
 
 
       /* Calculate pointer to current attrib:
@@ -1179,6 +1188,12 @@ static struct x86_reg get_buffer_ptr( struct translate_sse *p,
          x86_mov(p->func, ptr, elt);
          break;
       }
+
+      /* Clamp to max_index
+       */
+      x86_cmp(p->func, ptr, buf_max_index);
+      x86_cmovcc(p->func, ptr, buf_max_index, cc_AE);
+
       x86_imul(p->func, ptr, buf_stride);
       x64_rexw(p->func);
       x86_add(p->func, ptr, buf_base_ptr);