From 90167d0ee94384327d4a8482b21fdd28ba2c7171 Mon Sep 17 00:00:00 2001 From: Jonathan Adamczewski Date: Mon, 5 Jan 2009 07:49:48 -0700 Subject: cell: SIMDize some subtractions Put edge.{dx,dy} into a union with a vector and perform subtractions in setup_sort_vertices() on vectors. Reduces spu_tri.o by ~300 bytes. --- src/gallium/drivers/cell/spu/spu_tri.c | 18 ++++++++++-------- 1 file changed, 10 insertions(+), 8 deletions(-) (limited to 'src/gallium/drivers/cell/spu') diff --git a/src/gallium/drivers/cell/spu/spu_tri.c b/src/gallium/drivers/cell/spu/spu_tri.c index 30531d38f81..1ef75b386fa 100644 --- a/src/gallium/drivers/cell/spu/spu_tri.c +++ b/src/gallium/drivers/cell/spu/spu_tri.c @@ -77,8 +77,13 @@ struct vertex_header { * Triangle edge info */ struct edge { - float dx; /**< X(v1) - X(v0), used only during setup */ - float dy; /**< Y(v1) - Y(v0), used only during setup */ + union { + struct { + float dx; /**< X(v1) - X(v0), used only during setup */ + float dy; /**< Y(v1) - Y(v0), used only during setup */ + }; + vec_float4 ds; /**< vector accessor for dx and dy */ + }; float dxdy; /**< dx/dy */ float sx, sy; /**< first sample point coord */ int lines; /**< number of lines on this edge */ @@ -517,12 +522,9 @@ setup_sort_vertices(const struct vertex_header *v0, spu_extract(setup.vmax->data[0], 0) > setup.cliprect_maxx) return FALSE; - setup.ebot.dx = spu_extract(setup.vmid->data[0], 0) - spu_extract(setup.vmin->data[0], 0); - setup.ebot.dy = spu_extract(setup.vmid->data[0], 1) - spu_extract(setup.vmin->data[0], 1); - setup.emaj.dx = spu_extract(setup.vmax->data[0], 0) - spu_extract(setup.vmin->data[0], 0); - setup.emaj.dy = spu_extract(setup.vmax->data[0], 1) - spu_extract(setup.vmin->data[0], 1); - setup.etop.dx = spu_extract(setup.vmax->data[0], 0) - spu_extract(setup.vmid->data[0], 0); - setup.etop.dy = spu_extract(setup.vmax->data[0], 1) - spu_extract(setup.vmid->data[0], 1); + setup.ebot.ds = spu_sub(setup.vmid->data[0], setup.vmin->data[0]); + setup.emaj.ds = spu_sub(setup.vmax->data[0], setup.vmin->data[0]); + setup.etop.ds = spu_sub(setup.vmax->data[0], setup.vmid->data[0]); /* * Compute triangle's area. Use 1/area to compute partial -- cgit v1.2.3