nir: add lower_all_io_to_temps flag

This will be used for freedreno and vc4 which require all inputs
and outputs to be copied to temps.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>

src/broadcom/compiler/nir_to_vir.c

@@ -1874,6 +1874,7 @@ nir_to_vir(struct v3d_compile *c)
 }
 
 const nir_shader_compiler_options v3d_nir_options = {
+        .lower_all_io_to_temps = true,
         .lower_extract_byte = true,
         .lower_extract_word = true,
         .lower_bitfield_insert = true,

src/compiler/nir/nir.h

@@ -1873,6 +1873,8 @@ typedef struct nir_shader_compiler_options {
    bool lower_extract_byte;
    bool lower_extract_word;
 
+   bool lower_all_io_to_temps;
+
    /**
     * Does the driver support real 32-bit integers?  (Otherwise, integers
     * are simulated by floats.)

src/gallium/drivers/freedreno/ir3/ir3_nir.c

@@ -50,6 +50,7 @@ static const nir_shader_compiler_options options = {
 		.vertex_id_zero_based = true,
 		.lower_extract_byte = true,
 		.lower_extract_word = true,
+		.lower_all_io_to_temps = true,
 };
 
 struct nir_shader *

src/gallium/drivers/vc4/vc4_program.c

@@ -2180,6 +2180,7 @@ nir_to_qir(struct vc4_compile *c)
 }
 
 static const nir_shader_compiler_options nir_options = {
+        .lower_all_io_to_temps = true,
         .lower_extract_byte = true,
         .lower_extract_word = true,
         .lower_ffma = true,