From a907c91e93cce88ee1929263c455fab541b8c4a3 Mon Sep 17 00:00:00 2001 From: "Juan A. Suarez Romero" Date: Fri, 23 Sep 2016 15:57:39 +0000 Subject: [PATCH] i965/vec4: consider subregister offset in live variables MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit Take into account offset values less than a full register (32 bytes) when getting the var from register. This is required when dealing with an operation that writes half of the register (like one d2x in IVB/BYT, which uses exec_size == 4). v2: - Take in account this offset < 32 in liveness analysis too (Curro) v3: - Change formula in var_from_reg() (Curro) - Remove useless changes (Curro) Signed-off-by: Samuel Iglesias Gonsálvez Reviewed-by: Francisco Jerez --- src/intel/compiler/brw_vec4_live_variables.h | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/src/intel/compiler/brw_vec4_live_variables.h b/src/intel/compiler/brw_vec4_live_variables.h index 2946b98aacc..92f79550fd4 100644 --- a/src/intel/compiler/brw_vec4_live_variables.h +++ b/src/intel/compiler/brw_vec4_live_variables.h @@ -91,7 +91,7 @@ var_from_reg(const simple_allocator &alloc, const src_reg ®, assert(reg.file == VGRF && reg.nr < alloc.count && c < 4); const unsigned csize = DIV_ROUND_UP(type_sz(reg.type), 4); unsigned result = - 8 * (alloc.offsets[reg.nr] + reg.offset / REG_SIZE) + + 8 * alloc.offsets[reg.nr] + reg.offset / 4 + (BRW_GET_SWZ(reg.swizzle, c) + k / csize * 4) * csize + k % csize; /* Do not exceed the limit for this register */ assert(result < 8 * (alloc.offsets[reg.nr] + alloc.sizes[reg.nr])); @@ -105,7 +105,7 @@ var_from_reg(const simple_allocator &alloc, const dst_reg ®, assert(reg.file == VGRF && reg.nr < alloc.count && c < 4); const unsigned csize = DIV_ROUND_UP(type_sz(reg.type), 4); unsigned result = - 8 * (alloc.offsets[reg.nr] + reg.offset / REG_SIZE) + + 8 * alloc.offsets[reg.nr] + reg.offset / 4 + (c + k / csize * 4) * csize + k % csize; /* Do not exceed the limit for this register */ assert(result < 8 * (alloc.offsets[reg.nr] + alloc.sizes[reg.nr])); -- 2.30.2