nir/algebraic: Add lowering for 64-bit usub_sat

v2: Rebase on 272e927d0e9 ("nir/spirv: initial handling of OpenCL.std
extension opcodes")

v3: Add a new lower_usub_sat64 flag that only applies to the 64-bit
version of the nir_op_usub_sat instruction.

v4: Also enable the lowering when nir_lower_iadd64 is set.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com> [v3]
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/767>

diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h
index d07e688b5a886a5a63fde15b99ff9ac46692a24e..eafd6fad8555bb2a6e55025553693a013cdd444c 100644 (file)
--- a/src/compiler/nir/nir.h
+++ b/src/compiler/nir/nir.h
@@ -2851,8 +2851,25 @@ typedef struct nir_shader_compiler_options {
     */
    bool lower_hadd64;
 
+   /**
+    * Set if nir_op_add_sat and nir_op_usub_sat should be lowered to simple
+    * arithmetic.
+    *
+    * If this flag is set, the lowering will be applied to all bit-sizes of
+    * these instructions.
+    *
+    * \sa ::lower_usub_sat64
+    */
    bool lower_add_sat;
 
+   /**
+    * Set if only 64-bit nir_op_usub_sat should be lowered to simple
+    * arithmetic.
+    *
+    * \sa ::lower_add_sat
+    */
+   bool lower_usub_sat64;
+
    /**
     * Should IO be re-vectorized?  Some scalar ISAs still operate on vec4's
     * for IO purposes and would prefer loads/stores be vectorized.

diff --git a/src/compiler/nir/nir_opt_algebraic.py b/src/compiler/nir/nir_opt_algebraic.py
index 97c4777a53414b4fa1f570149577b3e423ff9362..6289551939a4895a5b69fd927c01610e297af985 100644 (file)
--- a/src/compiler/nir/nir_opt_algebraic.py
+++ b/src/compiler/nir/nir_opt_algebraic.py
@@ -1010,6 +1010,7 @@ optimizations.extend([
 
    (('uadd_sat', a, b), ('bcsel', ('ult', ('iadd', a, b), a), -1, ('iadd', a, b)), 'options->lower_add_sat'),
    (('usub_sat', a, b), ('bcsel', ('ult', a, b), 0, ('isub', a, b)), 'options->lower_add_sat'),
+   (('usub_sat@64', a, b), ('bcsel', ('ult', a, b), 0, ('isub', a, b)), 'options->lower_usub_sat64 || (options->lower_int64_options & nir_lower_iadd64) != 0'),
 
    # Alternative lowering that doesn't rely on bfi.
    (('bitfield_insert', 'base', 'insert', 'offset', 'bits'),