summaryrefslogtreecommitdiffstats
path: root/src/mesa/drivers/dri/i965/brw_ir_fs.h
diff options
context:
space:
mode:
authorFrancisco Jerez <currojerez@riseup.net>2015-02-06 01:11:18 +0200
committerFrancisco Jerez <currojerez@riseup.net>2015-02-10 16:05:50 +0200
commitbfbb0e84e11e06af3d751701f157a21915976ca1 (patch)
tree718a2aab4abd08bd6bb4ba23419ce652be315bdc /src/mesa/drivers/dri/i965/brw_ir_fs.h
parent447879eb88b8df41ad32cf4406cc636b112b72d9 (diff)
downloadexternal_mesa3d-bfbb0e84e11e06af3d751701f157a21915976ca1.zip
external_mesa3d-bfbb0e84e11e06af3d751701f157a21915976ca1.tar.gz
external_mesa3d-bfbb0e84e11e06af3d751701f157a21915976ca1.tar.bz2
i965: Move IR object definitions to separate header files.
One should be able to manipulate i965 IR without pulling the whole FS/VEC4 visitor classes -- Optimization passes and other transformations would ideally be visitor-agnostic. Among other issues this avoids a circular dependency between the header file where such visitor-agnostic code will be defined and the main FS/VEC4 header where both IR (layer below) and visitor (layer above) happen to be defined. Reviewed-by: Matt Turner <mattst88@gmail.com>
Diffstat (limited to 'src/mesa/drivers/dri/i965/brw_ir_fs.h')
-rw-r--r--src/mesa/drivers/dri/i965/brw_ir_fs.h255
1 files changed, 255 insertions, 0 deletions
diff --git a/src/mesa/drivers/dri/i965/brw_ir_fs.h b/src/mesa/drivers/dri/i965/brw_ir_fs.h
new file mode 100644
index 0000000..2369834
--- /dev/null
+++ b/src/mesa/drivers/dri/i965/brw_ir_fs.h
@@ -0,0 +1,255 @@
+/* -*- c++ -*- */
+/*
+ * Copyright © 2010-2015 Intel Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ */
+
+#ifndef BRW_IR_FS_H
+#define BRW_IR_FS_H
+
+#include "brw_shader.h"
+
+class fs_inst;
+class fs_visitor;
+
+class fs_reg : public backend_reg {
+public:
+ DECLARE_RALLOC_CXX_OPERATORS(fs_reg)
+
+ void init();
+
+ fs_reg();
+ explicit fs_reg(float f);
+ explicit fs_reg(int32_t i);
+ explicit fs_reg(uint32_t u);
+ explicit fs_reg(uint8_t vf[4]);
+ explicit fs_reg(uint8_t vf0, uint8_t vf1, uint8_t vf2, uint8_t vf3);
+ fs_reg(struct brw_reg fixed_hw_reg);
+ fs_reg(enum register_file file, int reg);
+ fs_reg(enum register_file file, int reg, enum brw_reg_type type);
+ fs_reg(enum register_file file, int reg, enum brw_reg_type type, uint8_t width);
+
+ bool equals(const fs_reg &r) const;
+ bool is_contiguous() const;
+
+ /** Smear a channel of the reg to all channels. */
+ fs_reg &set_smear(unsigned subreg);
+
+ /**
+ * Offset in bytes from the start of the register. Values up to a
+ * backend_reg::reg_offset unit are valid.
+ */
+ int subreg_offset;
+
+ fs_reg *reladdr;
+
+ /**
+ * The register width. This indicates how many hardware values are
+ * represented by each virtual value. Valid values are 1, 8, or 16.
+ * For immediate values, this is 1. Most of the rest of the time, it
+ * will be equal to the dispatch width.
+ */
+ uint8_t width;
+
+ /**
+ * Returns the effective register width when used as a source in the
+ * given instruction. Registers such as uniforms and immediates
+ * effectively take on the width of the instruction in which they are
+ * used.
+ */
+ uint8_t effective_width;
+
+ /** Register region horizontal stride */
+ uint8_t stride;
+};
+
+static inline fs_reg
+negate(fs_reg reg)
+{
+ assert(reg.file != HW_REG && reg.file != IMM);
+ reg.negate = !reg.negate;
+ return reg;
+}
+
+static inline fs_reg
+retype(fs_reg reg, enum brw_reg_type type)
+{
+ reg.fixed_hw_reg.type = reg.type = type;
+ return reg;
+}
+
+static inline fs_reg
+byte_offset(fs_reg reg, unsigned delta)
+{
+ switch (reg.file) {
+ case BAD_FILE:
+ break;
+ case GRF:
+ case ATTR:
+ reg.reg_offset += delta / 32;
+ break;
+ case MRF:
+ reg.reg += delta / 32;
+ break;
+ default:
+ assert(delta == 0);
+ }
+ reg.subreg_offset += delta % 32;
+ return reg;
+}
+
+static inline fs_reg
+horiz_offset(fs_reg reg, unsigned delta)
+{
+ switch (reg.file) {
+ case BAD_FILE:
+ case UNIFORM:
+ case IMM:
+ /* These only have a single component that is implicitly splatted. A
+ * horizontal offset should be a harmless no-op.
+ */
+ break;
+ case GRF:
+ case MRF:
+ case ATTR:
+ return byte_offset(reg, delta * reg.stride * type_sz(reg.type));
+ default:
+ assert(delta == 0);
+ }
+ return reg;
+}
+
+static inline fs_reg
+offset(fs_reg reg, unsigned delta)
+{
+ assert(reg.stride > 0);
+ switch (reg.file) {
+ case BAD_FILE:
+ break;
+ case GRF:
+ case MRF:
+ case ATTR:
+ return byte_offset(reg, delta * reg.width * reg.stride * type_sz(reg.type));
+ case UNIFORM:
+ reg.reg_offset += delta;
+ break;
+ default:
+ assert(delta == 0);
+ }
+ return reg;
+}
+
+static inline fs_reg
+component(fs_reg reg, unsigned idx)
+{
+ assert(reg.subreg_offset == 0);
+ assert(idx < reg.width);
+ reg.subreg_offset = idx * type_sz(reg.type);
+ reg.width = 1;
+ return reg;
+}
+
+/**
+ * Get either of the 8-component halves of a 16-component register.
+ *
+ * Note: this also works if \c reg represents a SIMD16 pair of registers.
+ */
+static inline fs_reg
+half(fs_reg reg, unsigned idx)
+{
+ assert(idx < 2);
+
+ if (reg.file == UNIFORM)
+ return reg;
+
+ assert(idx == 0 || (reg.file != HW_REG && reg.file != IMM));
+ assert(reg.width == 16);
+ reg.width = 8;
+ return horiz_offset(reg, 8 * idx);
+}
+
+static const fs_reg reg_undef;
+
+class fs_inst : public backend_instruction {
+ fs_inst &operator=(const fs_inst &);
+
+ void init(enum opcode opcode, uint8_t exec_width, const fs_reg &dst,
+ fs_reg *src, int sources);
+
+public:
+ DECLARE_RALLOC_CXX_OPERATORS(fs_inst)
+
+ fs_inst();
+ fs_inst(enum opcode opcode, uint8_t exec_size);
+ fs_inst(enum opcode opcode, const fs_reg &dst);
+ fs_inst(enum opcode opcode, uint8_t exec_size, const fs_reg &dst,
+ const fs_reg &src0);
+ fs_inst(enum opcode opcode, const fs_reg &dst, const fs_reg &src0);
+ fs_inst(enum opcode opcode, uint8_t exec_size, const fs_reg &dst,
+ const fs_reg &src0, const fs_reg &src1);
+ fs_inst(enum opcode opcode, const fs_reg &dst, const fs_reg &src0,
+ const fs_reg &src1);
+ fs_inst(enum opcode opcode, uint8_t exec_size, const fs_reg &dst,
+ const fs_reg &src0, const fs_reg &src1, const fs_reg &src2);
+ fs_inst(enum opcode opcode, const fs_reg &dst, const fs_reg &src0,
+ const fs_reg &src1, const fs_reg &src2);
+ fs_inst(enum opcode opcode, const fs_reg &dst, fs_reg src[], int sources);
+ fs_inst(enum opcode opcode, uint8_t exec_size, const fs_reg &dst,
+ fs_reg src[], int sources);
+ fs_inst(const fs_inst &that);
+
+ void resize_sources(uint8_t num_sources);
+
+ bool equals(fs_inst *inst) const;
+ bool overwrites_reg(const fs_reg &reg) const;
+ bool is_send_from_grf() const;
+ bool is_partial_write() const;
+ int regs_read(fs_visitor *v, int arg) const;
+ bool can_do_source_mods(struct brw_context *brw);
+
+ bool reads_flag() const;
+ bool writes_flag() const;
+
+ fs_reg dst;
+ fs_reg *src;
+
+ uint8_t sources; /**< Number of fs_reg sources. */
+
+ /**
+ * Execution size of the instruction. This is used by the generator to
+ * generate the correct binary for the given fs_inst. Current valid
+ * values are 1, 8, 16.
+ */
+ uint8_t exec_size;
+
+ /* Chooses which flag subregister (f0.0 or f0.1) is used for conditional
+ * mod and predication.
+ */
+ uint8_t flag_subreg;
+
+ uint8_t regs_written; /**< Number of vgrfs written by a SEND message, or 1 */
+ bool eot:1;
+ bool force_uncompressed:1;
+ bool force_sechalf:1;
+ bool pi_noperspective:1; /**< Pixel interpolator noperspective flag */
+};
+
+#endif