glsl: remove rcp from sqrt()

Per a patch from Marek Olšák, we can simply multiply the incoming
value by 1/sqrt(x) instead of using rcp.

We're keeping the x==0 check to avoid generating NaN for sqrt(0).

diff --git a/src/mesa/shader/slang/library/slang_common_builtin.gc b/src/mesa/shader/slang/library/slang_common_builtin.gc
index a25ca55bc42592c06f1929bc147b0c569a7ee537..3d464630df2d8d5fb92e075f621b574e1d4a4a3b 100644 (file)
--- a/src/mesa/shader/slang/library/slang_common_builtin.gc
+++ b/src/mesa/shader/slang/library/slang_common_builtin.gc
@@ -605,7 +605,7 @@ float sqrt(const float x)
    const float nx = -x;
    float r;
    __asm float_rsq r, x;
-   __asm float_rcp r, r;
+   r = r * x;
    __asm vec4_cmp __retVal, nx, r, 0.0;
 }
 
@@ -615,8 +615,7 @@ vec2 sqrt(const vec2 x)
    vec2 r;
    __asm float_rsq r.x, x.x;
    __asm float_rsq r.y, x.y;
-   __asm float_rcp r.x, r.x;
-   __asm float_rcp r.y, r.y;
+   r = r * x;
    __asm vec4_cmp __retVal, nx, r, zero;
 }
 
@@ -627,9 +626,7 @@ vec3 sqrt(const vec3 x)
    __asm float_rsq r.x, x.x;
    __asm float_rsq r.y, x.y;
    __asm float_rsq r.z, x.z;
-   __asm float_rcp r.x, r.x;
-   __asm float_rcp r.y, r.y;
-   __asm float_rcp r.z, r.z;
+   r = r * x;
    __asm vec4_cmp __retVal, nx, r, zero;
 }
 
@@ -641,10 +638,7 @@ vec4 sqrt(const vec4 x)
    __asm float_rsq r.y, x.y;
    __asm float_rsq r.z, x.z;
    __asm float_rsq r.w, x.w;
-   __asm float_rcp r.x, r.x;
-   __asm float_rcp r.y, r.y;
-   __asm float_rcp r.z, r.z;
-   __asm float_rcp r.w, r.w;
+   r = r * x;
    __asm vec4_cmp __retVal, nx, r, zero;
 }